Vous êtes sur la page 1sur 79

Examensarbete vid Institutionen fr geovetenskaper ISSN 1650-6553 Nr 228

Rainfall-runoff Model Application in Ungauged Catchments in Scotland

Alexander Peter Anthony Fionda

Abstract
Rainfall-runoff model application for ungauged catchments in Scotland
Alexander Peter Anthony Fionda

Department of Earth Sciences, Uppsala University Villavgen 16, SE-752 36 Uppsala, Sweden. The conceptual rainfall-runoff model Hysim is used to estimate the flow in ungauged catchments in Scotland by Scottish Water. However, there are non-quantified uncertainties associated with the outcomes of the modelling strategy used. In order to identify and quantify these uncertainties it was necessary to use the framework of proxy-basin validation in order to evaluate the performance of different modelling strategies. The proxy-basin validation test requires hydrologically analogous catchments for the evaluation of models, a Region Of Influence regionalisation method was used in order group selected catchments by Q95(%MF). Four groups of four catchments were established, which covered Q95(%MF) 5-7%, 7-9%, 9-11% and 11-13%. The allocation of donor catchment and target catchment for each Q95(%MF) group was accomplished through discussion with Scottish Water with respect to existing Scottish Water modelled catchments. A single donor catchment and three target catchments were therefore indicated for each group. Two modelling strategies were developed by the study; the first full transposition method used the entire optimised parameter-set from the donor catchment with the exception of the target catchments catchment area parameter. The second partial transposition method used the entire optimal parameter-set with the exception of the target catchments interception storage, time to peak, rooting depth and catchment area parameters. It was found that the full transposition method had the least uncertainty associated its use for flow estimation when the parameter-set was derived from a donor catchment calibration that was excellent. Contrarily, it was found that the partial transposition model method had the least uncertainty associated with flow estimation for parameter-sets that were derived from a relatively poor donor catchment calibration. Encouraged by this testing framework, this study has suggested the use of catalogue of donor parameter-sets that can be used to estimate flow for catchments that are hydrologically similar. This strategy of hydrological modelling has been recommended to improve existing Scottish Water Hysim methodology. Keywords ungauged catchment, proxy-basin validation, region of interest, transposition method, hysim, rainfall-runoff model, sepa, scottish water, scotland.

Referat
Anvndning av en avrinningsmodell i ett skotskt avrinningsomrde utan vattenfringsmtningar
Alexander Peter Anthony Fionda

Institutionen fr geovetenskaper, Uppsala universitet Villavgen 16, 752 36 UPPSALA. Scottish Water anvnder den begreppsmssiga avrinningsmodellen Hysim fr att uppskatta vattenfringen i skotska avrinningsomrden utan vattenfringsdata. Den valda modelleringsstrategin har emellertid resulterat i icke-kvantifierade oskerheter i berknade vattenfringar. Fr att identifiera och kvantifiera de oskerheter som r frbundna med olika modelleringsstrategier var det ndvndigt att anvnda sig av information frn likartade avrinningsomrden. Den valda regionaliseringsmetoden anvnde hydrologiskt analoga avrinningsomrden som definition p likhet. Analogin grundades p inflytanderegion (Region of Influence) som erhlls genom att gruppera utvalda avrinningsomrden utefter Q95 (% medelflde). Fyra grupper med fyra avrinningsomrden valdes ut grundat p fljande Q95-grnser (% medelflde): 5-7%, 7-9%, 9-11% and 11-13%. Frdelningen av analoga avrinningsomrden (omrden med vattenfringsmtningar vars parametervrdesuppsttningar skulle verflyttas) och mlomrden (utan mtningar) fr varje Q95-grupp erhlls efter diskussion med Scottish Water frn omrden dr Scottish Water modellerat vattenfringen. Ett analogt omrde och tre mlomrden valdes ut fr varje grupp. Studien anvnde tv modelleringsstrategier. Den frsta metoden, total verflyttning, anvnde hela parametervrdesuppsttningen frn det analoga omrdet med undantag av mlomrdets area. Den andra metoden, partiell verflyttning, anvnde hela parametervrdesuppsttningen med undantag fr mlomrdets interceptionslager, tid till hgflde, rotdjup och area. Den totala verflyttningsmetoden hade lgst oskerhet nr parametervrdesuppsttningen hrleddes frn ett omrde med utmrkt kalibrering. Den partiella verflyttningsmetoden hade, andra sidan, lgst oskerhet nr parametervrdesuppsttningen hrleddes frn ett omrde med dlig kalibrering. Efter att ha provat de tv metoderna utmynnade studien i ett frslag till en katalog med parametervrdesuppsttningar fr omrden som kan bedmas som hydrologiskt lika. Denna strategi fr hydrologisk modellering har rekommenderats som frbttring av befintlig Hysimmetodik hos Scottish Water. Nyckelord Avrinningsomrde utan vattenfringsdata, validering mot likartade omrden, inflytanderegion, verflyttningsmetod, hysim, avrinningsmodell, SEPA, Scottish Water, Skottland.

II

Contents
1. Introduction ................................................................................................................. 1 1.1 Research objectives ................................................................................................. 1 1.1.2 1.1.2 1.2.1 1.2.2 2 Scottish Water and its resource systems .......................................................... 2 The role of hydrologic modelling in Scottish Water ........................................ 5 Quantifying the uncertainty associated with parameterisation ...................... 13 Modelling flow in ungauged catchments ....................................................... 15

1.2 The use of Hysim rainfall-runoff modelling by Scottish Water ............................ 12

1.3 Key questions and summary of methods ............................................................... 16 Materials and Methods ................................................................................................. 18 2.1 Analogue and target site selection from SEPA catchments .................................. 18 2.2 The Hysim conceptual rainfall-runoff model ........................................................ 21 2.3 Derivation of inputs ............................................................................................... 24 2.4 Hysim model calibration ....................................................................................... 26 2.5 Development of parameter transposition methods ................................................ 27 2.6 Evaluating model performance using the proxy-basin test ................................... 28 3 Results .......................................................................................................................... 29 3.1 Hydrological statistics of mega-zones and SEPA catchments .............................. 29 3.2 Calibration quality of donor catchments ............................................................... 31 3.3 Evaluating model performance with transposition method chosen ....................... 39 3.4 Evaluating model performance with selection of target catchment ...................... 44 4 Discussion .................................................................................................................... 47 4.1 Hydrological statistics of mega-zones and SEPA catchments .............................. 47 4.2 Calibration quality of donor catchments ............................................................... 48 4.3 Uncertainty identified with selection of target catchment ..................................... 53 4.4 A more pragmatic methodology for estimations of flow ...................................... 54 5 6 7 8 Conclusion ................................................................................................................... 56 Acknowledgements ...................................................................................................... 58 References .................................................................................................................... 59 Appendices ................................................................................................................... 63 Appendix A: Hysim operational notes ........................................................................... 63
III

Appendix B: Parameter-set references ........................................................................... 64 Appendix C: Results of validation ................................................................................. 65

Definition of terms
MF - the mean flow. Q95 - the 95th percentile of mean flow; the flow exceeded or equalled 95 % of the time. Q95(%MF) - the 95th percentile of mean flow as a percentage of mean flow. Source catchment - a catchment containing source of water, which is utilised by Scottish Water. Donor catchment - the catchment for which an optimal parameter-set is achieved through calibration. Target catchment/analogue - a catchment chosen through a method of regionalisation to be similar in character to the donor catchment. Model a software based representation of a physical system. Model software consists of a programmed framework, into which physical data and estimated parameters are placed, in order to represent a physical system. This study evaluates model performance, where a model consists of the programming, input data and parameters as a whole. This status is stored by Hysim the model programming- as a single project file, which is referred to as a model in its own right. Parameter-set - a set of estimated parameter values that may be adjusted in order to manipulate the outcomes of a model. Optimal parameter-set - a set of parameter values that provide the best estimation of flow, commonly achieved through the calibration of a model. Transposition the process of transferring parameter values from a donor catchment optimal parameter-set to a target catchment parameter-set. Full Transposition Method (FTM) - a method describing the transposition of every parameter from the donor catchments optimal parameter-set to the target catchment parameter-set. The catchment area of the target is maintained as a parameter for the target catchment parameter-set. Partial Transposition Method (PTM) - a method describing the transposition of part of the donor catchments optimal parameter-set to the target catchment parameter-set. catchment area, time to peak, rooting depth and interception storage of the target are maintained as parameters for the target catchment parameter-set. Uncertainty - A state of having limited knowledge where it is impossible to exactly describe existing state or future outcome; in this study this is quantified by evaluating model performance, using accuracy between estimated and recorded flow.
IV

1. Introduction
The benign human curiosity in the future drifts in and out of focus in society. It can enthral as the subject of films and can spell boon or doom in the media. As a species capable of producing much of what we utilise in our day to day existence it is our privilege to be able to successfully predict the outcomes of what we create and control. In order to do so, we rely on the continual development of the mathematical model. However, when we attempt to utilise the environment around us, there is the desire, and often assumption, that a similar level of prediction is available. We necessitate accurate environmental prediction, whether it may concern the local weather next week or global climate in the next century. Unfortunately, the natural world is almost infinite in its scale of complexity and cannot be represented in its entirety by any model. As such, the outcomes of mathematical models that attempt to tell us more about the future is discussed more as a form of prophecy than prediction (Beven, 1993). Hydrological variables are but one aspect of the natural world. Mathematical models, especially conceptual rainfall-runoff models, are a capable means of narrowing down future states of hydrological variables for a given area. For water management companies this is essential, as predictions of the likely states of variables are invaluable in resource planning. It is within the realms of prediction that rainfall-runoff models, capable of simulating flow in areas that are ungauged, are best suited. Models of hydrological systems have been progressing for the best part of three decades. One branch of development of modelling tools leads to the prediction rainfall and consequential runoff in a hydrological system. Conceptual rainfall-runoff models are among the most ubiquitously used tools in hydrology. Input data is more readily available for their application unlike their counterparts: the complex, physically based, distributed models. Conceptual models are often comparatively simple and easy to use, that said, the drawbacks of model parameters being inter-correlated or overparameterised is not uncommon. It is the case that some model parameters will have a physical bias that ties directly to variations on the catchment scale. Due to the fact that such variations are virtually unquantifiable in the field, calibration is an essential step in representing real runoff calculations. This leads to the pursuit of the optimal parameter-set that produces the greatest closeness to reality and a process of parameter alteration that inevitably brings about multiple solutions with different sets of parameters. Uncertainty therefore arises in modelling, it is discussed as the confusion as to which set of parameters to choose for application by Beck, 1987. This study aims to elaborate upon the uncertainty associated with parameter selection by testing parameter-sets that have been derived by various methods. It will then be possible to quantify this uncertainty by the comparison of the accuracy of these methods.

1.1

Research objectives

This study is undertaken in cooperation with Scottish Water -the publically owned water authority for Scotland, who expressed considerable interest in improving the efficiency of their rainfall-runoff modelling strategy for various operations. This study aims to evaluate model performance, where a model consists of the programming, input data and parameters as a whole. In doing so, the focus of evaluation will be on changes made to the parameter-set and potentially data. In literature surrounding model evaluation, the model software itself is usually under scrutiny and described as the model; such analysis is not the focus of this study. A review of the current internal and external publications on Scottish Waters modelling strategy

reveals non-quantified uncertainties in the input, parameterisation and calibration of their modelling scheme that require addressing. This paper attempts to identify and quantify the uncertainty surrounding parameterisation by testing the accuracy of various methods of parameter-set derivation. Furthermore, this uncertainty evaluation may then be used to infer an improved, more pragmatic method for modelling the flow in ungauged catchments. Using the framework of proxy-basin validation to evaluate the uncertainties associated with flow estimations in ungauged catchments requires the following aims to be fulfilled: i. Select catchments for experimentation that are both approved by a monitoring agency in terms of quality and that represent typical Scottish Water source catchments. Use a method of regionalisation to group hydrologically analogous catchments in compliance with the proxybasin framework Identify catchments that are suitable for deriving parameter-sets and those that are suitable as the target of the evaluation process; so called donor and target catchments. Update the input data and data selection periods and improve the calibration of existing Scottish Water models for those catchments identified as donor catchments. Develop two methods of parameter transposition and test parameter-sets upon target catchments in order to evaluate accuracy and quantify uncertainty associated with parameterset selection. Interpret whether uncertainties are quantified enough for the recommendation of using a single method of parameter-set derivation for the estimation of flow in all hydrological analogous, ungauged catchments.

ii.

iii.

In completing these objectives, it is possible to identify a single donor parameter-set for each hydrologically similar group that can be used to estimate flow in ungauged catchments with hydrological similarity to a quantified level of accuracy. A library of models would then exist that would each represent a range of hydrological similarity that could be used whenever flow was needed to be estimated in an ungauged catchment. This builds upon suggestions by Jacobs (2010); the ability to approve this as an outcome would recommend a more pragmatic Hysim methodology for Scottish Waters estimation of flow in ungauged catchments. For the objectives of this report to be upheld it is important to address some additional vulnerability within the current scheme of Hysim modelling that Scottish Water employs. A detailed method for the calibration of Hysim models must be documented and made consistent with Scottish Water guidelines; however the method should be seen to improve existing modelling procedure in order to assist with future Hysim modelling studies. Where there are pre-existing calibrations models for catchments, it is an aim of this to update or improve these models where possible. This may be achieved through taking advantage of the improved rating and record of evapotranspiration or precipitation data records or by alterations in the model construction process.

1.1.2 Scottish Water and its resource systems


Scotland, with respect to global water availability, is a water rich country. In terms of actual water availability Northern Europe has 34.6 x 103m3 per year per capita average for the past 60 years; when compared to the average for the entire of Europe (4.9 x 103m3) it is clear that there is a uneven geographic distribution of available water throughout Europe (Gleik, 1993). It is important not to
2

construe this data as a reflection of unlimited water resource capability; there are problems with water resources in relation to the public supply of water. A wide variability exists in the ability for the water authority, Scottish Water, to maintain water supply during peak demands and during droughts. In 2002, Scottish Water was crated by the merger of three water authorities in accordance with the Water Industry [Scotland] Act 2002. Scottish Water is accountable to the Scottish Parliament through the Scottish Ministers, it is publically owned. It remains a product of the amalgamation of 210 water boards and local councils since 1968. This unification provides the authority with a unified, consistent and strategic approach to Water Resource Planning that strengthens the operations it defines from its Water Resource Plan (WRP). The WRP (Scottish Water, 2009) is a regulatory document that has been developed in collaboration with the Scottish Environmental Protection Agency (SEPA). Its aims are to: Define Scottish Waters long term water resources strategy to ensure the consistent supply of drinking water to protect public health and facilitate economic growth, while abstracting and using water in a sustainable way to provide a value for money service for customers. Provide a twenty five year assessment of the Supply Demand Balance across Scotland at a zone-level that is consistent with good practice in the UK. Justify investment to restore deficits in the Supply Demand Balance in a prioritised water resource zones during the next investment period and beyond. The WRP therefore represents the interests of: environmental and water resource regulation, economic regulation, customer interests and consumer quality respectively (Scottish Water, 2009). The Water Resource Plan is subject to the model of planning guidance SEPA provides. As such, Scottish Water is requested to produce data for all Water Resource Zones (WRZs) defined within Scotland. WRZs are defined as the largest possible zone in which all customers experience the same risk of supply failure from a resource shortfall (Scottish Water, 2009). For the 2007/2008 period, 230 water resource zones exist across Scotland. Due to the low population density in Scotland, there is a large variation in the distribution of WRZs. A large quantity of WRZs are located in the Highlands and Islands, which supply isolated communities; contrasting with the eleven centrally located WRZs that supply almost half the population of Scotland (Scottish Water, 2009). Such an extensive collation of WRZs is unfamiliar to the majority of water management authorities; in England and Wales companies usually have one to ten WRZs. Therefore, the environmental agency guidelines that request data on all WRZs seems a task implicated with difficulties on a number of levels: specifically the collation of data for 230 WRZs and their constituent water sources.

Average Demand (Ml/d) Argyll and Bute Ayrshire and Inverclyde Central Scotland Dumfries and Galloway East Lothian and Borders Fife Fort William Grampian Inverness and Central Highlands North West Coast Orkney Shetland Skye and Lochalsh Tayside and Rural Forth Valley Western Isles Wick Scotland Total 2009/10

Average Population Number Number of Number of Demand (ml/d) (000s) of WRZs WTWs Sources 41.8 65.9 32 33 46 255.8 1,265.40 78.3 59.5 143 8.9 148.5 86.2 3.3 8.7 10.7 6.7 127 13.4 13.9 2,044 440.4 2,712.20 131.1 145.3 357.3 17.9 420.2 201.6 7.5 19.6 22 14.6 372.6 26.8 28.7 5,035 8 11 5 11 1 19 11 20 19 10 13 28 13 22 7 220 14 30 18 17 11 19 17 24 19 11 14 28 16 22 7 278 66 110 31 25 26 21 36 28 21 18 23 32 18 23 8 481

FIGURE 1.2.1: SCOTTISH WATER MEGA-ZONE REGIONAL GROUPING WITH ALLOCATED WATER RESOURCE ZONES (WRZS). IMAGES USED WITH PERMISSION (SCOTTISH WATER, 2009).

Water resource zones are grouped geographically into sixteen mega-zones, shown in figure 1.2.1. The disparity of population density across Scotland is notably significant, elucidating the need for an additional WRZs for every small pocket of population across a large area; these are classified as standalone zones. In studying the population given in thousands it is a frequent trend that a smaller population per mega-zone have a greater number of WRZs i.e. the population of central Scotland: 2,712,200, which is supplied by 11 WRZs whereas Argyle and Bute have a population of 41,800 and are supplied by 32 WRZs. However, this is not a rule as such; some low populations also have a low number of WRZs i.e. Wick, a population of 28,700 and 7 WRZs (Scottish Water, 2009). Water resource zones are supplied by Water Treatment Works (WTW), the distribution of which is directly influenced by the occurance of standalone zones. Each standalone zone is supplied directly by a single WTW, making 202 WTW zones that are supplied by a sole WTW across Scotland. The remaining 28 WRZs have more than one WTW. The Central Scotland mega-zone incorporates the cities of Edinburgh and Glasgow; the 11 WRZs within Central Scotland contain 30 WTW and serve 54% of the household population of Scotland (Scottish Water, 2009). The interconnectivity provided between these zones reduces the risk of supply failure within the mega-zone; although, there is a difference in risk between certain zones over others (Scottish Water, 2009). The risk of supply failure is considerably greater in the standalone zones as there is limited or no connectivity between the WTW. Efforts are being made by Scottish Water to further plans that would ensure a greater interconnectivity between standalone zones and reduce the risk of supply failure amongst these areas.

The supply demand balance from raw water sources to the water treatment works output for each water resource zone is essential for effective water management across Scotland. A suppy system incorporates the assets of collection, storage, transfer and treatment up to the output of the water treatment works (Scottish Water, 2009). It is the part of the supply system concerning collection that is of greatest interest for defining water sources in Scotland. Scottish Water Report that for 2007/2008 there were 532 sources providing water for supply to the population of Scotland; see figure 1.2.2. Large population centres in Central Scotland, such as Edinburgh, Glasgow, Dundee and Stirling are supplied by a small number of large reservoirs, whereas isolated communities in remote parts of Scotland rely upon numerous small reservoirs. This follows the trends identified from the disparate state of WRZ distribution across Scotland.

FIGURE 1.2.2: SURFACE WATER SOURCES UTILISED BY SCOTTISH WATER. IMAGE USED WITH PERMISSION (SCOTTISH WATER, 2009).

The distributions of raw water sources across Scotland are illustrated by map of WTW localities across the country; see figure 1.2.3. The majority of the 59 loch sources are located in the northwest of Scotland. Groundwater sources are found throughout Scotland; there are 42 spring sources and 54 borehole systems that make up 96 in total. 207 river sources are divided into: 103 indirect sources, which feed reservoirs and 104 pure river sources, which are generally larger in the east and smaller in the west. Impounding reservoirs, of which there are 170, and their contributing feeder river sources provide 82% of raw water to water treatment works in Scotland. Direct river sources provide 10% of raw water, whilst lochs and groundwater each provide 4% and 4% respectively (Scottish Water, 2009).

1.1.2 The role of hydrologic modelling in Scottish Water


Hydrological assessment occurs on a variety of levels dependent upon the project at hand. Scottish Water (2009) identified various scenarios where hydrological assessment is required for a water management authority. There is a division highlighted between internal projects i.e. a Scottish Water capital project with water quality or growth considerations and Scottish Water capital projects with environmental consideration. This study will focus on hydrological assessment associated with the eventual calculation of yield; a requisite for the supply-demand balance for all Scottish Water capital projects including Scottish Waters Water Resource Plan (Scottish Water, 2009). Yield is expressed in terms of the maximum continuous output that can be supplied in drought severe enough that on average its occurrence would cause a failure of supply one in forty years (Scottish Water, 2009). The use of conceptual rainfall-runoff models, such as Hysim, for estimation of stream flows is universal in water management authorities. This flow data requires some method of transformation before yield can be calculated. The estimation of yield requires either the estimation of
5

a natural flow duration curve (FDC) or its 95 flow percentile (Q95). Software such as Hysim-Aquator is capable of yield estimates directly from Hysim simulation data, whilst the Report 108 based Method (Institute of Hydrology, 1992) may also be used to estimate yield using one regression equation (Gustard et al., 1992). In addition, Scottish Government Directions on Environmental Standards (SGES) determine an allowance of abstraction given as a percentage of the FDC (Scottish Water, 2009). There is therefore a great necessity to represent catchment flow data in its Q95 and FDC form. Techniques from which an FDC may be obtained are: gauged records, Hysim modelling and Low Flows Enterprise calculation. A long term record of gauged flow for the focus catchment is undoubtedly the most accurate, reliable and practical method of FDC production. Empirical observations will always be of greatest value to the hydrologist, yet lengthy continuous gauged flow data for Scotland, and indeed much of the world, is not available. Furthermore, in the context of individual water management authorities such as Scottish Water, their abstraction sites are not close enough to long term gauges for representative FDCs to be derived. Where funding and time permits, it is beneficial to initialise flow gauging for sites (Mott Macdonald, 2010). It is suggested that there is suitability in short term direct flow gauging if enough analysis into finding a suitable analogue is undertaken. For the implementation of flow gauging to be effective in a project there must be a local, long term analogue. If such an analogue cannot be found then the gauging period for the catchment in focus must be greater than four years, which may extend beyond practical means for the project. If a local, long term analogue can be found and gauged data is provided that is over three years in length then transposition will be used between catchments and allow a revised FDC that better represents the focus catchments. Methods of transposition between catchments are detailed by Jacobs (2010); however there is no comparison between the efficacies of this procedure compared to the representation of the focus catchment by a rainfall-runoff model. A modelling strategy would inevitably require, and use, the same proximal, long term target catchment. Comparison between the resulting FDC would illustrate the value of initiating flow gauging at a focus site. Low Flows Enterprise (LFE), developed by Centre for Ecology and Hydrology (CEH), is a software package that is used to estimate the flow duration curve at ungauged sites. Wallingford Hydrosolutions currently maintain this software. The Scottish Environmental Protection Agency use LFE as the elected method for FDC derivation at ungauged sites in Scotland. Scottish Water has purchased LFE and is capable of providing LFE estimates on request. LFE obtains FDC through the selection of 5 Region-Of-Influence (ROI) gauged catchment sites, which must be determined to be similar to the donor catchments hydrological statistics. These five ROI provide an individual FDC, which is rescaled by the mean flow for the subject site; this is calculated by a separate model within the LFE software. Hysim-Aquator permits the transfer of flow data and its derivative FDC or Q95 to calculate a yield. Aquator achieves this through the simulation of daily transfers and abstractions for a given WRZ and represents this as a one in forty yield. The Hysim-Aquator method for yield calculation was developed in 2001, as a Scotland and Northern Ireland Forum for Environmental Research (SNIFFER) project. It is a combined software package comprising of the hydrological rainfall-runoff simulation model Hysim and Aquator, which is a water resource system model. Hysim as stand-alone software is a daily rainfall-runoff model. Its intended use is to simulate a historic daily river flow series based on historic daily rainfall and potential evapotranspiration; whilst taking into account artificial influences such as: groundwater abstractions, river abstractions or river discharges (Manley, 1978).
6

Aquator was developed especially for the aforementioned SNIFFER project. It uses output from the Hysim model as an input for the simulation of a water resource system; Hysim was also specially adapted for this project. The daily storage of a reservoir or loch may be simulated based upon a balance of inputs and outputs in terms of demands, compensation and freshets. Aquator is capable of modelling a number of demand centres as well as the key components of a resource system, such as: pumping stations, water treatment works, pipelines, hydro-generators, river abstractions and groundwater abstractions (Manley, 1978). The application and accuracy of Hysim-Aquator is limited by the availability of good quality input variables and parameters; guidance is provided by Scottish Water on the processing of input data.

Background of Scottish Waters Hysim models


31 individual Hysim rainfall-runoff models are currently in use for 70 WRZs. These models are consequently responsible for covering 250 source catchments, which in turn feed 90 WTWs For the 31 independent donor catchments there are three catchments that provide gauged data for the implementation of 40% of the Hysim models; these are: Green Burn located at Loch Dee, data is used for 29 catchments and 8 WRZ models, River Creed located at Creed Bridge, data is used for 18 catchments and 12 WRZ models, River Calder located at Muirshiel, data is used for 35 catchments and 6 WRZ models. Other source catchments are used for Hysim model calibration; however these catchments have been applied to two or three models only (Scottish Water, 2009). The 31 Hysim models were developed as part of larger studies than the models themselves; in these studies it was thought pragmatic to apply a single calibrated model to a range of catchments, despite more representative catchments being available for calibration. The models that use Green Burn and River Calder for calibration amongst others- have not been critically reviewed in order to assess the on-going validity of these calibrations since there original development in 2001 and 2002. However, the necessity of applying a single model to a number of hydrologically different catchments such as Creed Bridge illustrates the lack of alternative gauged catchments available on the Western Isles and Northern Isles. To continue the discussion of validity, the data quality upon which the models are based is also in question. The River Calder gauging station at Muirshiel is noted to be downstream of the River Calder abstraction intake and is therefore artificially impacted; the catchment is also identified by SEPA as unsuitable for use as an analogue. This issue is not brought to attention in the 2001 report by Camphill, from which the River Calder calibration is derived. There is significant reason to question the validity and revisit the calibration considering the wide scope of its application. Short term gauges have been used for the calibration of Hysim models: using one year of gauged data, the Geimisgarve and Clibh catchments are applied. These short term gauges were developed specifically for the Water Framework Directive (WFD) WR1 SR06 project (Scottish Water, 2006), which required models for a large number of remote islands in Scotland. These short term gauges were used as alternative calibrations for comparison with an adopted Creed Bridge model. The calibration for Clibh was accepted in three models and Geimisgarve was accepted for a single model. This position highlights the difficulty in establishing good Hysim donor parameter-sets for the large number of remote sites in the North Western Isle, the Western Isles and the Northern Isles. It has meant that the normal practise for Hysim calibration cannot always be followed i.e. the recommended record length would usually require at least 5 years of representative, gauged data.

A program of additional Scottish Water flow gauging sites was implemented since 2006 as a direct result of the conclusions drawn by the WFD WR1 SR06 project. This aided the confirmation of river flows at key project sites, which was not previously possible and helped strengthen observations made in those catchments. The resulting flow records cannot be used for direct calibration of Hysim models until the representative record length exceeds 5 years; ideally 7 years. However, the flow records may be used for indirect validation of existing Hysim flow records in order to help agreement upon a FDC for specific water sources during consultation with SEPA. SEPA will use the LFE instead of this FDC unless flow gauging can provide a high level of confidence to the Hysim modelled flow. Hysim-Aquator models are developed for reservoir or loch multiple-source system and generally not used for WRZs that are only supplied by river intakes. The criteria for their disuse is a system for which there is no storage available; exceptions do exist, such as the River Dee sources, which are used to extend gauged flow records. The rationale for excluding rivers is that river sources are generally smaller with low yields and therefore lower priority. It has been a concern that Hysim models do not perform well around the 99th percentile of flow (Q99). This issue is not as critical in systems with low storage as it is normally the combined impact of the whole flow regime and storage capacity available that determines the system yield. In contrast, a river with no storage has a yield that is determined based on the lowest daily flow values from the driest 3 or 4 years within the flow record. Therefore, any poor model performance at these very lowest flows can have a significant impact on yield sensitivity for river-only systems (Scottish Water, 2009).

FIGURE 1.2.3: THE DISTRIBUTION OF SCOTTISH WATER CALIBRATION GAUGES AND MODELLED WATER
TREATMENT WORKS THROUGHOUT MAINLAND SCOTLAND AND ISLANDS. IMAGE TAKEN FROM JACOBS (2010).

Region of influence: identifying hydrological similarity without geographical constraints


SEPA quality approved catchments are presented on figure 1.2.4. These catchments are categorised by their LFE derived Q95(%MF) flow descriptor, which is used to identify LFE ROI groupings across Scotland. The purpose of using Q95(%MF) is to eliminate the requisite for regional boundary grouping and can allow a number of catchments to be regionally grouped without boundaries. This is important due to the number of isolated source catchments in Scotland, such as islands, which would be unaccounted for if boundary regionalisation of catchments was pursued. LFE ROI catchments are mapped based upon Q95(%MF) values, which reflect regional variation in hydrological regimes. Five main regional groups are established, grouped by Hydrometric Area (HA) boundaries. Such a simplification of grouping causes a few stations to be in the wrong grouping such as Killing and Cultybraggan (Scottish Water, 2009). These stations have more hydrological similarity to stations in the North West region, yet are included in the central region due to the HA being of the Tay. In addition, Alness and Diriebught House stations have a better hydrological fit with the North East Region (Scottish Water, 2009). Further elaborations on using ROI as an alternative for regionalisation are discussed in subsequent chapters that discuss the literature surrounding catchment selection for parameter derivation and application. Scottish Water adopts this scheme of LFE ROI groups across Scotland in order to identify suitable catchments for use in the validation of optimal parameter-sets. If the desire is to estimate flows for an ungauged catchment with a Q95(%MF) of 6% it is possible to refer to the 5% 10% Q95(%MF) group and establish a number of target catchments for validation. This is useful tool as there is a reliable potential analogue gauge available that represents natural flow regimes that are mostly checked for hydrometric quality. In this LFE approach for obtaining suitable catchments, a distance factor is neglected unlike the SEPA analogue selection tool as it was developed to be reliant on proximity between catchments.

10

FIGURE 1.2.4: LOW FLOWS ENTERPRISE (LFE) REGION OF INFLUENCE (ROI) STATIONS AND SUGGESTED th ANALOGUES BY THE SCOTTISH ENVIRONMENTAL PROTECTION AGENCY (SEPA). 95 PERCENTILE OF FLOW AS A PERCENTAGE OF MEAN FLOW (Q95(%MF)) IS ILLUSTRATED BY THE COLOUR AND SIZE KEY. TAKEN FROM JACOBS (2010). 11

1.2 The use of Hysim rainfall-runoff modelling by Scottish Water


Scottish Water rely upon the use of a conceptual hydrological rainfall-runoff that is calibrated to nearby hydrologically analogous catchments in order to produce yield estimations for the majority of surface water supply systems in Scotland. Yield is defined as the maximum continuous output for given surface water source that can be supplied during a dry period of a stated severity. Yield estimations require flow data and, due to the lack of long term site specific flow gauging within a reasonable proximity to abstraction sites, representative target catchments are required for flow estimation in ungauged rivers. Unlike other locations in Britain such as England, there is not the same length or level of detail to historical flow records that affords the direct use of flow gauging records for yield estimation. These direct flow gauging installations are usually restricted to timescales under three years and are not suitable for direct application in model calibration. A requisite for model calibration is a good record of at least seven years of gauged flow data (Scottish Water, 2009). Therefore, direct flow gauge installations are usually used exclusively to provide validation of the optimal parameter-set for the catchment. The conceptual rainfall-runoff hydrological simulation modelling software used by Scottish Water is Hysim, which is continuously developed by Water Resource Associates. Hysim can be integrated within Oxford Scientifics water resource system model Aquator in order to produce estimations of yield for a given catchment. It is the case that uncertainties in Hysim modelling strategy and procedures have the potential to significantly undermine the confidence in Scottish Waters yield estimates. This has the implication of making any planning or investment schemes, based on the estimation of yield, less reliable. The Hysim-Aquator yield modelling process has been used by Scottish Water for over 10 years and it is understood that there is a lack of repeatability in some of their Hysim models. It is assumed that this is due to the number of times certain models have been updated or even the lack of a consistent guidance framework for application, which is often protracted by the use of different consultants. The data input, parameterisation and calibration processes for Hysim are aimed to be as objective and consistent as possible, yet these uncertainties are still apparent. The uncertainty and related sensitivity associated with these three key processes of modelling are not quantified. It would seem pertinent to quantify uncertainties and sensitivities within these processes in order to strengthen the reliability and confidence in model flow estimations and thus gain a more accurate yield estimate. Scottish Water identifies potential uncertainties sourced from inconsistencies in modelling procedure that are related to the project specific circumstances of the model genesis (Scottish Water, 2009). For instance, when genesis lies in large projects, the focus of the project can lay beyond the scope of detailed flow estimation and appraisal of models. Such projects often produce models that have less importance placed upon the quality of the model calibration and input data. It is also the case that Hysim models created and adapted by different companies that offer different approaches to the construction of models and weight internal protocol over guidance available from Scottish Water. Neuman (2003), states that the bias and uncertainty that result from an inadequate conceptual mathematical model are typically larger than those introduced through an inadequate choice of parameter values; it is essential to choose the correct donor catchment and target catchment for model calibration and transposition respectively. In light of this, there is a strong need to create a clear and pragmatic methodology for the selection of a parameter-set for use on a target catchment that is ungauged. In order to avoid the aforementioned caveats of model construction it would be beneficial to construct a library of Scottish Water acknowledged Hysim models that could be applied to

12

ungauged catchments with a good degree of confidence; as outlined in the closing notes of Jacobs (2010). There is considerable justification for an improvement in the method in which a model is calibrated and applied to an ungauged catchment using an analogue catchment. Without improvement, any future work where Hysim models are updated or new models developed will continue to provide inaccurate estimations of flow for a given catchment. A common recommendation, based on a poor correlation between calibration and direct gauged flows, is to scale the estimated Hysim flow series to agree with SEPAs low flows enterprise flow duration curve (Scottish Water, 2011). LFE estimates may also be uncertain and lead to misleading yield estimates and be an inadequate result for the estimation of yield by Scottish Water.

1.2.1 Quantifying the uncertainty associated with parameterisation


A number of literature sources discuss methods of validation for models in order to evaluate the uncertainty that exists in a particular model. This paper requires the evaluation of uncertainty associated with parameterisation. Seibert (1999a) gives a thorough review of the meaning and application of the term validation in a hydrological modelling context. A series of applications incorporating all current methods of validation with specific outcomes is detailed. A method for gaining a measure of model parameter uncertainty in between hydrologically similar, gauged catchments is identified by Seibert (1999a) as the proxy-basin test. Calibration takes place on a single catchment and validation of the optimal parameter-set is achieved by the transposition of these parameters to another gauged catchment. Seibert et al. (1999b) used a conceptual rainfall-runoff model, the Hydrologiska Byrns model (HBV), to calibrate a single catchment and validate this calibration on a further two catchments of similar character in the Black Forest, Germany. An expression of model efficiency was studied for every application of the calibrated parameter-set. In the optimisation of one parameter-set and application on the similar two catchments the average measure of efficiency was 0.76 (1 corresponding to a perfect fit). When calibrated in the hydrologically analogous catchments and parameters were applied to the original catchment the measure of efficiency was 0.84. These steps are characteristic of the proxy basin method and elucidate that there is less uncertainty associated with the model with 0.84 thus quantifying uncertainties associated with parameter choices. The proxy-basin test of validation provides a significant solution to the main objective of this investigation: to quantify the uncertainty associated with parameterisation when estimating flow in ungauged catchments. It is essential to use the framework of the proxy-basin test in order to evaluate the uncertainty associated with the parameter-set construction methods that are proposed. As indicated by the aforementioned study by Seibert et al. (2009b) a measure of model efficiency according to the Nash-Sutcliffe Efficiency Criterion is the preferred method of evaluating the performance of a model. In previous studies commissioned by Scottish Water i.e. Jacobs (2010), the model efficiency for Hysim is not used to calibrate or evaluate the performance of the model; instead the FDC and associated flow descriptor statistics (Q95, MF, Q95(%MF)) are used as a measure of accuracy and therefore an evaluation of the performance of the model with a specific parameter-set. A concern in this approach is highlighted through conference in this study due to the fact that FDCs, unlike model efficiency neglect the temporal aspect of model performance. Additionally, it is possible for an estimated FDC to exactly match a recorded FDC whilst the model efficiency is very poor. However, in studies by Westerberg (2011), which involved the analysis of FDC calibrations in 23 basins, the FDC calibration method was found to have potential for calibration to regionalised FDCs for ungauged basins; reducing the initial model uncertainty by approximately 70% (Westerberg, 2011).
13

Therefore the use of FDC in calibration and as a comparative measure of accuracy is used throughout this report.

1.2.2

ROI as a method of regionalisation

Using the framework of proxy-basin validation in order to evaluate the accuracy of parameter-sets -in accordance with the scheme outlined by Seibert et al (1999b)- requires a method of regionalisation for the application of parameter-sets. The process of transferring information from neighbouring catchments to the catchment of interest is generally referred to as hydrological regionalisation (Blschl and Sivapalan, 1995). It is used to make predictions about hydrological quantities at sites where data are absent or inadequate, frequently for design purposes (Beran, 1990). Three regionalisation methods are used to identify suitable gauged catchments, from which the optimised parameter values are used to estimate flow for the target ungauged catchment: i. The regression method establishes a relationship between the optimised parameter values and catchment climate and physical attributes. Parameter values are then estimated for the ungauged catchment from its attributes and the identified relationship. The spatial proximity method uses parameter values from the geographically closest gauged catchment because neighbouring catchments are expected to behave similarly due to shared physical and climatic characteristics The physical similarity method transfers the entire set of parameter values from a physically similar catchment.

ii.

iii.

Varying the method by which parameters are transferred from the optimal parameter-set of a donor catchment to the target catchment is the source of the full parameter and partial transposition methods that are evaluated for associated uncertainty in this study. Therefore a degree of regionalisation must be factored into the choice of donor and ungauged catchments. The spatial proximity method, where the geographically closest gauged catchment has its parameters transferred to the target catchment would be somewhat adequate for application in Scotland. However, this is unlikely due to the high variation in catchment character across Scotland, owing to underlying geologies and marine landforms for which there are Scottish Water source catchments assigned. Scottish Water utilise ROI as an approach to regionalisation in order to categorise suitable donor catchments and target catchments for parameter transfer. Acreman & Wiltshire, 1987 first suggest this approach with the premise that the technique allows each donor catchment to have a unique set of target catchments, which inclusively constitute the region for that catchment. Thus, there are no boundaries indicating a specific variable and donor catchments within a specific area do not need to have the same target catchments. According to Feaster and Tasker (2002) the ROI is defined as a set containing the n closest stations. The ROI is defined as the set of all stations closer than a distance R (in predictor variable or geographic space) from the site or, if the number of stations in that set is smaller than some minimum allowable number n, the n closest stations. Scottish Water use predictor variables such as: location, SAAR and BFI to identify donor and target catchments. In order to test the validity of Q95(%MF) in such a role, Q95(%MF) is used to help select the catchments for parameter transfer. ROI in application is seen on figure 1.2.4, using gauged values of Q95(%MF). As such, Q95(%MF) ROI will be used as a proxy for regionalisation methods in the allocation of donor and target for the provided SEPA catchments in this study.

14

1.2.2 Modelling flow in ungauged catchments, a calibration scheme compatible with Scottish Water and Scotland
A previous study for Scottish Water by Jacobs Engineering UK Limited (Jacobs, 2010) investigated approaches to Hysim rainfall-runoff modelling and the resultant impact on yield sensitivity. The particular focus of this study was to investigate the sensitivity of Hysim models to the choice of target catchment as well as the impact of using different calibration periods (record length and representativeness). The resulting variable Hysim model outputs were then tested in Aquator water resource system models using different catchment sizes and reservoir storage volume to assess the impact in terms of yield sensitivity and uncertainty. As well as presenting conclusions on model sensitivity and consequential yield sensitivity -the former of which will contribute to the discussion of this paper- the study provides a tailored procedure for flow estimation in ungauged catchments using Hysim for Scottish Water. As this study is interested in isolating the uncertainty associated with parameterisation it is imperative to adhere to a standard method for input data selection and calibration procedure whilst updating these existing processes where improvements can be made without perturbing the uncertainty in parameter choice. The Jacobs (2010) study is therefore used as the reference of a data processing, selection and model calibration procedure that suits Scottish Water. Using this approved calibration procedure as a framework will allow this paper to take advantage of the outcomes of the Jacobs study and further develop the standard calibration method to suit the objectives of this paper. As there is no method for evaluating the choice of parameter-set in ungauged catchments provided by the Jacobs study it would be useful to extend this calibration procedure to formalise a standard method for testing a donor parameter-sets ability to estimate flows in an ungauged catchment. The total process accounted for in the Jacobs (2010) study covered: donor catchment selection, target catchment selection, data acquisition, data quality control, calculation of catchment statistics, calculation of catchment parameters (catchment area, time to peak, rooting depth and interception storage), the calibration of the donor catchment using Hysim, comparison of estimated flow to recorded flow and final calculation of flow estimation descriptors (Jacobs, 2010). The procedures outlined by Jacobs (2010) serve as a foundation for the development of this studys methodology due to the bespoke nature of their outcomes to suit Scottish Water guidelines. Jacobs (2010) aimed to ensure consistency and repeatability in the Hysim calibration procedure by removing the degree of user subjectivity from the process i.e. eliminating the manual adjustment of parameters. It was suggested in the study that an increase in user subjectivity would exist between the calibrations of multiple catchments. Also identified was the trade-off between subjectivity and level of detail, time spent, user experience and quality of the calibration. The calibration process was designed to enable relative differences in resulting yields to be discussed with the same procedure followed in the calibration process. This is important as subjectivity was identified as a key cause of sensitivity in the use of Hysim by Scottish Water (2009). In the calibration methodology for this report it is necessary to achieve the best possible calibration and so manual calibration is essential for some calibrations. However, manual adaptation of parameters beyond the standard calibration procedure must be limited to a number of attempts for best fit between estimated and recorded flow; thus, limiting the subjectivity. In addition, the Jacobs (2010) report identified that the uncertainty associated with different catchment choice appears to be slightly larger than the uncertainty associated with choice of record length and found that an eight year calibration offered the most reliable
15

estimations of flow. Selected catchments for this study therefore have an eight year period of good quality recorded data in order to eliminate the influence of other uncertainties upon the observations of this studys aims. Additionally, Jacobs (2010) suggested that there was an increase in yield sensitivity with a reduced flashiness of catchment. It should be noted that, due to the small sample size involved in the study, these conclusions were considered provisional within the report itself. It would be useful to explore these provisional conclusions in this papers discussion of uncertainties associated with the character of target of catchments chosen. A final remark of the Jacobs (2010) study was the suggestion that collating a library of precalibrated Hysim project files would be an adequate solution to limit uncertainty and reduce the labour involved with detailed calibration for each application. Producing a library of well calibrated Hysim projects, each with a quantified uncertainty and clear construction method would allow Scottish Water or external consultants to use a model where justified. Essentially, this study evaluates the proxy basin methodology for estimating flows in ungauged catchments. If uncertainty is reduced due to the use of single method of parameterisation for hydrologically analogous catchments then this single method can be used to produce a number of calibration parameter-sets that could each be used on a large number of hydrologically analogous catchments with a known level of uncertainty; thus creating a pragmatic and cost effective estimation of flow in ungauged catchments.

1.3 Key questions and summary of methods


In this thesis three key questions are addressed upon completion of the stated objectives of the study: i. Is it possible to use the proxy-basin test framework to quantify the uncertainties associated with parameter transposition? Is there a method of parameter transposition that has a lower uncertainty associated with its application? How can this information be used to create a more pragmatic model application scheme within Scottish Water?

ii.

iii.

In order to support these hypotheses, the objectives of the report were accomplished with the following procedural methodology: i. Selection of 16 gauged catchments that are approved by SEPA and are representative of catchments that are utilised by Scottish Water. This is accomplished through the comparison of hydrological statistics between catchments and mega-zones, and supplemented by discussion with Scottish Water. Use of ROI as a regionalisation method for grouping potential donor and target catchments according to Q95(%MF) flow descriptor. Allocation of donor and target catchments according to availability and quality of data as well referring to existing use within Scottish Water. Update and improve existing Scottish Water catchment calibration models if encountered. Update data used in projects where possible and choose a different time period where beneficial.

ii.

iii.

16

iv.

Development of two parameter transposition method identified as full transposition and partial transposition methods. Evaluation of the performance of these parameter-sets on each group of target catchments using one calibrated donor model according to the proxy-basin test framework; elucidating uncertainty associated with these parameter-sets. Comparison of catchment characteristics in relation to parameter-set performance in order to expand upon conclusions made about uncertainty in target catchment selection mad by previous studies.

v.

17

2 Materials and Methods


2.1 Analogue and target site selection from SEPA catchments

Hydrological representativeness
There are 207 Scottish Water river sources within Scotland, of which 103 feed reservoirs and 104 are standard river sources. The standard river sources are directly applicable for the investigation of runoff and approximation of yield for a water source; therefore, 104 rivers distributed throughout Scotland are suitable candidates for flow estimation studies. In total, 24 catchments referred to as analogue catchments by SEPA- were refined from those selected by SEPAs analogue selection tool and evaluation expertise at Scottish Water. Data for these 24 catchments were obtained from the respective parties and covered the entire recorded period for flow, precipitation and evapotranspiration; the specific derivation of which is covered in later chapters. In order to represent the range of water resource catchments that water authorities in Scotland utilise for water supply in Scotland it was essential to compare the hydrological statistics of catchments to the average statistics of Scottish Water mega-zones that are identified on figure 1.2.1. The hydrological variables that were studied included: A value of catchment area was referenced from the UK Hydrimetric Register (UKHR) delivered by the Centre for Ecology and Hydrology (CEH) (2008). Standard annual average rainfall (SAAR) was referenced from the UKHR. Base flow index (BFI); a value derived from gauged daily flow data. This represents the contribution of the slow flow or groundwater flow in the total measured runoff at the catchment outlet , giving a degree of flashiness i.e. the frequency and rapidity of short term changes in daily runoff values (Deetris & Lital, 2008). This was referenced from the UKHR. Base flow index (BFI HOST SCOT); a base flow value that is derived from Low Flows Enterprise results. 95th percentile flow value as a percentage of the mean flow (Q95(%MF)); a value derived from gauged daily flow data where available,else Low Flow Enterprise modelling was used. This value is a commonly used measure of flashiness and other runoff characteristics; it illustrates the flow that is exceeded 95% of the time as a percentage of mean flow.

It should be noted that Polloch, Skeabost and all mega-zones use the calculated BFI hydrology of soil types Scotland (HOST SCOT) value, which is obtained from LFE results. BFI HOST SCOT is not a substitution for gauged BFI and has been flagged as producing inadequate results in uses by Jacobs (2010); however, this does not directly affect the choices made to exclude specific catchments. Previous studies by Jacobs (2010) critically assessed catchments using factors of: artificial influence, standing water area, record length, Institute of Hydrology grading quality and any further information that would be influential to the suitability of the gauged data for a catchment. This revealed features that could perturb the natural flow of the river and cause error in the evaluation phase of the experiment and were taken into account when selecting catchments for experimentation.

18

Confirmation of ROI grouping


Using ROI for Q95(%MF) catchments were grouped into four Q95(%) groups. The plot of aforementioned catchment statistics were observed in order to interpret the suitability of Q95(%MF) for grouping catchments of hydrological similarity. Each group was then elected a donor catchment, chosen for its reliability as a presently used model and representativeness of typical flow per Q95(%MF) group. The remaining catchments within each group would then be denoted as target catchments. In total there were sixteen catchments identified for use in the study: four donor catchments and twelve catchment analogues; these are displayed on table 2.1.1. The distribution of these chosen catchment analogues across Scotland in relation to their Q95(%MF) group is illustrated on figure 2.1.2. Of the twenty four catchments that refined from SEPA provided analogues there were eight omitted from the study. These eight catchments represented Q95(%MF) groups that were below 5% and above 13%. These catchments were not used for the evaluation of transposition performance; however, they were included in observations of catchment representativeness (see figures 3.1.1 to 3.1.3).
TABLE 2.1.1: DESIGNATION OF DONOR AND TARGET CATCHMENTS. DONOR CATCHMENTS ARE INDICATED IN BOLD, ALL OTHER CATCHMENTS ARE CATCHMENT ANALOGUES.
Group Station Name
Braevallich Glen Strae Polloch Deephope Durkadale

Barsolus Inverlochy Skeabost Luss Candermill Creed Bridge Dargall Lane Lathro Brockhoperig Kinross Whitburn

Area (km) 24.10 36.62 8.05 30.99 19.60 32.83 47.09 80.55 35.47 25.50 44.83 2.07 24.60 38.59 33.60 31.95

SAAR BFI BFI-HOST Q95 (mm) (gauged) (SCOT) (%MF) 2745 0.22 0.22 6.5% 2772 0.26 0.21 5.2% 2650 0.23 0.23 5.5% 1486 0.32 0.26 6.1% 1145 0.28 0.42 8.8% 1150 0.35 0.38 9.0% 2946 0.26 0.24 7.1% 2218 0.26 0.26 7.9% 2296 0.35 0.28 9.4% 1034 0.40 0.31 9.2% 1462 0.25 0.44 9.3% 2439 0.21 0.28 9.8% 1164 0.54 0.43 11.0% 1732 0.37 0.34 11.4% 1266 0.56 0.42 12.1% 1032 0.32 0.30 11.5%

Selection of time periods


It was essential to make sure that each chosen catchment had a period of gauged data that was at least ten years in length and that this was of good quality. Ten years was considered the calibration period length for previous studies by Jacobs (2010). Ten years allowed for two years for model warm-up and the eight years of calibration data. Scottish water recommends a minimum of seven years of data record; therefore this is more than satisfactory. The data to be used was: rainfall, evapotranspiration and flow data; making a total of 48 sets of data that would be subject to scrutiny. The selection of time periods of data was dictated by the availability, quality and representativeness of rainfall, flow and potential evapotranspiration input data. It was chosen that there would be one time period for each Q95(%MF) group, making four time periods in total. Selection was achived by comparing the data between the four catchments in each group then deciding which time period is most complete and which is most respresentative of each individual catchment. It was thought best to keep the time series the same across each four catchments in each Q95(%MF) group in order to similarise climate limits upon inputs across Scotland for the Q95(%MF) group and therefore enable a fair test. If climate

11 - 13%

9 - 11%

7 - 9%

5 - 7%

19

conditions across Scotland differed from the average for a given day, month or year then these trends would impact all catchments in direct comparison with each other. In some instances Scottish Water calibrations existed for catchments that this study had allocated donor catchments. It was seen as useful to improve these models by updating existing data where improved data was available and selecting or extending time periods where possible.

20

FIGURE 2.1.2: CHOSEN CATCHMENTS FROM SEPA PROVIDED CATCHMENTS, AN INDICATION OF Q95(%MF) ROI GROUPING IS PROVIDED. SUPPLIED BY SCOTTISH WATER ON REQUEST.

2.2

The Hysim conceptual rainfall-runoff model


21

The conceptual rainfall-runoff model employed for this study is Hysim, a Hydrological Simulation model developed by the Water Resource Associates (WRA). WRA are a network of consultants in water resources, water quality, hydrology, groundwater hydrology and flooding. Their clients include some of the most important water management bodies in Europe i.e. the European Union, United Kingdom Environmental Agency, National Power, French government, British Waterways and SNIFFER. SNIFFER is a research and development company that works cooperatively with Oxford Scientific Software to develop catchment rainfall-runoff models as well as models for water resource system simulations such as the estimation of yield (Entec UK, 2003). The development of AquatorHysim was undertaken by SNIFFER on behest of Scottish Water for the surface water yield and operational reliability project. The combined program is considered the best practice methodology for estimation of yield (Scottish Water, 2009) and improves previous estimations of yield. Consequently an integration of the Aquator with the rainfall-runoff model in use for England, the Hydrologic Resource Centre reservoir resource Simulator (HEC-ResSim), is currently in development (US Army Corps of Engineers, 2011). Hysim is founded upon two IBM Mathematical Formula Translating System (FORTRAN) subroutines. Initially the model processes both the hydrology and hydraulics of a catchment separately. The hydrology routine is based on seven reservoirs within the catchment. These are illustrated on figure 2.2.1.

FIGURE 2.2.1: FLOW CHART OF HYSIM HYDROLOGY CALCULATION ROUTINE. TAKEN FROM MANLEY (2006).

In the model it is parameters that determine the capacity of and the rate of transfer between each storage as well as the equations that determine transfer processes. Parameters are designated through calibration of the model, they are constant throughout time. Variables in the model describe the volumes in each storage and rates of transfer, they vary with time.

22

Parameters alterations are split into three sections within Hysim, these are data, basic and advanced. In accordance with the Jacobs (2010) standard calibration methodology, which is based upon guidance in the Hysim User Manual (Manley, 2006), the data specific and advanced hydraulic parameters remain at their default values for this study. The basic hydrological parameters that are changed within this report are illustrated at their default values on figure 2.2.2.

FIGURE 2.2.2: BASIC PARAMETER VALUES FOR ALTERATION DURING THE CALIBRATION PROCESS. SCREENSHOT FROM HYSIM V5.00 (MARCH 2010 BUILD).

Data requirements for Hysim are practically obtainable in the field, they are: potential evapotranspiration, potential snowmelt, precipitation, the net value of discharges and abstractions, groundwater abstractions. Input formats are monthly, daily, hourly; though daily is usually sufficient and is the format used for all data in this study. However, it worth noting that data used for input is not required to be in the same time-step format for either hydrological or hydraulic subroutines. Spatially, the data can be distributed amongst user specified sub-catchments. This can be advantageous in reflecting heterogeneity in the hydrology or meteorology across the catchment area; however, this is not required for the study at hand. In the current version of the model: Hysim v5.00 (March 2010 build), there are in excess of six different output data. The most critical of these output data are the daily mean simulated flows, recorded flows as well as the basic statistical analysis and monthly summaries. If sub-catchments or channels are setup there is a generation of flow for every time increment. There is also an output for recorded and simulated daily flow for each reach. Actual evapotranspiration is also available as an output of the model should it be required. A measure of efficiency or accuracy of the simulations is required in order to evaluate the success of a model and adjust parameters with the goal of achieving an optimal parameter-set. Hysim provides a packaged graphics tool for instant hydrograph comparison between the recorded and simulated flow. Hysim also provides a summary of the daily and monthly difference statistics. Additionally Manley (2006) indicates that efficiency, the percentage of explained variance, can be calculated: (2) In this case Qm represents mean daily flow, Qo is observed daily flow, and Qs is simulated daily flow.
23

2.3

Derivation of inputs

Flow rainfall and evapotranspiration data


Historic flow data is provided by SEPA for each of the candidate catchments. Time series data of historical flow measurements is compared to simulated flow in order to evaluate the accuracy of the simulation. The quality of recorded flow data varied between catchments; data could be of a poor standard or missing for days, weeks, months or years in an otherwise complete record. A solution of infilling the missing data was necessary in order to provide a complete record of flow for a small number of the 16 catchments. Two approaches were found to be successful for flow infill. The most accurate infilling of missing data requires the construction, calibration and simulation of a model based on the catchment with the poor flow record. This process follows the same process of model development as outlined in this study. Simulated values may then be substituted for the corresponding missing values in the recorded flow record. This method required good quality and availability of other input data as well as a complete parameter-set for the catchment. This method was therefore only suitable for the donor catchments as these were the only complete, fully calibrated models. In the case of infilling missing data for the analogous catchments, an average of ten years of values was obtained, taking the five years preceding and superseding the missing value in the flow record. Both methods provided successful representations of recorded flow that was otherwise unavailable and allowed more accurate statistical comparisons to be drawn between simulated and recorded flow for catchments. In previous studies by Scottish Water, rainfall was obtained from daily measurement gauges local to the catchment. This procedure was prior to the first Water Resource Plan; it involved the identification of suitable rainfall gauges and infilling gaps or extending records in order to identify rain gauging weights, completed externally from Hysim. Importing this data into Hysim allowed the generation of daily, aerial, weighted rainfall for a catchment. 37.5% of all Hysim models used this method in order to compute rainfall until recent revisions (Scottish Water, 2009). Discussed in Scottish Waters Hydrology Guidance (2010), the Met Office has recently revised their method for providing gridded rainfall data across the United Kingdom. Data is currently available for a 5km2 gridded data set, from the start of 1958 up to the end of 2009. The updated Met Office gridded rainfall is used for input into Hysim for the eight catchments. The derivation of a suitable potential evapotranspiration (PET) series depends upon data availability for specific catchments. Typically, daily PET series can be generated by Hysim for the period 1918 to 1998 using a tool developed as part of the SNIFFER project (SNIFFER, 2001). This method is used for calculations of PET for the eight catchments; however, data will only be available to the end of 1998. Methods for extending PET beyond 1998 exist; Scottish Waters preference for PET extension is the acquisition of MORECS PE weekly 40km2 data grids, which are superimposed onto existing data from 1995 to 1998; providing a prediction beyond 1998. Due to the implicated costs of acquiring and updating MORECS data, this is not be used for PET estimation in this study. The preferred method in this study is the application of an average annual PET value from the calculated Hysim PE data series. A monthly average is applied from the last 10 year time period of study and extended to fill the remaining time period. Where this is not possible it is recommended by Jacobs (2010) to scale yearly averages of near-catchment PET data provided by the Centre of Hydrology according to relationships derived from overlapping periods of data. Daily values of PET were obtained from the average of preceding Hysim data. In some cases this was an essential step due to the lack of available data for internal Hysim PET calculations. Although PET input data was more uncertain, the accuracy
24

of rainfall data is seen as more critical for modelling. This is reflected in the single parameter optimisation method outlined in appendix 8.1. It was found that PET factor could be adjusted in recognition of its unreliability in order to improve estimations.

Parameters
Initial parameters were estimated by referring to hydrological data available for the catchments. These data are illustrated in figure 3.2.1; manually calculated parameters were: catchment area (km2), time to peak (hrs), interception storage (mm) and rooting depth (mm). These parameter values are approximations based upon methods of calculation according to the Hysim User Manual, Manley (2006), which use ordnance survey map observations of area and forest cover. The remaining parameters of the Hysim model were kept at their default values shown on figure 2.2.2. It was important to use these specific parameters for alteration as they are those historically observed in previous evaluations of Hysim performance by Jacobs (2006). Catchment area was obtained from the NRFA Hydrometric register from the Centre of Ecology and Hydrology (CEH) (Centre for Ecology and Hydrology, 2008). Time to peak controls the simulation of the response of minor channels within the catchment; both the Hysim User Manual (Manley, 2006) and the The UK Flood Estimation Handbook (Institute of Hydrology, 1999) give the equation for calculation as: (2) L is the length of the stream, S is slope of the stream in m/km and Tp is time to peak in hours. Interception storage represents the storage of moisture with flora; with moisture being added to this storage from rainfall. It is therefore an approximation of the proportion of the vegetation density of surfaces for a catchment. A value of 2.0mm is normal for grassland and urban areas and up to 10.0mm for woodland (Manley, 2006). Soil rooting depth is also dependent upon studying vegetation coverage, typically it is between 500 and 1000; woodlands may be as high as 5000mm. Time to peak, interception storage and soil rooting depth were calculated through observations of 1:10000, 1:25000 and 1:50000 Ordnance Survey (OS) maps. This was made available by the OS Openspace Application Programming Interface (API) as an overlay for Google Earth (Brock, 2009). Accessing OS maps via Google Earth allowed the plot of river courses and presentation of elevation transects across a rivers course. It also allowed accurate calculations of vegetation coverage areas, where the proportion of grassland/moorland to woodland was required for interception storage and soil rooting depth.

25

2.4

Hysim model calibration

Once satisfied with the input data, as well as the initial input parameters for the proposed modelling scenario, the recommended modelling strategy for Hysim was approached. This strategy has been consistently developed alongside Hysim, for this study it follows the guidance manual by Manley in 2006. Due to the extensive version of history and continual maintenance of Hysim by developers, it was of high priority to check that the current version of Hysim in use was fully up to date before beginning modelling. The strategy suggests selecting discharge data that is representative of the natural flow regime, of good quality and of suitable length. A period of data of ten years length is used in the previous studies by Jacobs (2010). This accounts for two years of at the beginning of the record that represents the model warm-up period and an additional eight years of accurately simulated flow. Model warm-up is a key process of runoff modelling, the two years signify part of the simulation period but not part of the analysis period (Manley, 2006). A period of ten years was therefore chosen for each of the catchments, of which two years would be considered warm-up and was not included in analysis. The input data: flow, evapotranspiration and precipitation are stored in individual files, with a separate file for the parameter-set. Hysim references these individual files using a single project file. Once a project file was created, a standard calibration methodology was applied; this is detailed in appendix 8.1. The goal of calibration is to select an optimum set of parameters that achieve simulation values that are the closest to recorded values. 1. Initial parameters estimated from calculated derivations of catchment area, time to peak, interception storage and rooting depth. All other parameters left as default. 2. Single parameter optimisation for PET 3. Extremes of Error Estimate (EEE) for horizon boundary permeability, base horizon permeability, upper interflow and lower interflow. 4. Manual alteration of parameters according to known relationships between parameter and flow estimations. Single parameter optimisation is an automatic calibration process within Hysim for a single parameter. As potential evapotranspiration (PET) is the most uncertain parameter this is used in the single parameter mode run, so as to provide the best estimation for this parameter. In previous calibrations by Scottish Water this parameter choice for single parameter run was rainfall factor; this does not correspond with the guidance provided by the Hysim User Manual (Manley, 2006). If there is good coverage, with a sufficient density and spread -as with improvements made to the quality of Met Office provided 5km2 gridded rainfall- then the use of PET for single parameter optimisation is suitable. In some cases, using single parameter optimisation for PET produced unrealistic parameter values and dubious flow estimations. In these instances it was more beneficial to use the precipitation factor for single parameter optimisation. EEE is also an automatic calibration process within Hysim for multiple parameter estimation. Hysim simultaneously optimises four parameters; these are: horizon boundary permeability, base horizon permeability, upper interflow and lower interflow. Manual estimation was a necessary step in achieving an optimal parameter-set. This was done according to noted relationships between parameter and flow estimations, which was only possible after considerably experience of using Hysim. As such, this was extensively time consuming. Once
26

enough expertise was acquired, it seemed valuable to make no more than ten attempts of manual calibration, otherwise the model would become over-parameterised; over-parameterisation greatly increases subjectivity, which should be avoided in models. By manually altering parameter values it is possible to combine the requirements of a manual calibration with the advantages of automatic calibration in order to provide a closer fit between simulations and observations in accordance with Boyle et al. (2000).

2.5

Development of parameter transposition methods

It was important to quantify the uncertainties associated with the choice of different parameter-sets. This was accomplished by developing two different methods of parameter transposition. These two methods would contain parameters that were derived from both the donor and one target catchment of a group for the estimation of flow in that specific target catchment. In order to contrast the uncertainty between the two transposition methods it was important to develop one method of transposition that fully derived parameters from the donor catchment and another method of transposition that only partially derived parameters from the donor catchment. Intuitively, these were described as the full parameter transposition method (FTM) and the partial transposition method (PTM). It is noteworthy that, previously, Scottish Water informally identified the FTM as direct transposition and the PTM traditional transposition. The exact operational procedure to replicate these construction methods is listed in appendices 8.2 and 8.3 for the FTM and PTM respectively; the parameter derivation scheme is outlined on table 2.5.1. In exploratory experimentation this study found that catchment area for the target catchment was required, else the results of the simulation had only a climatic relationship to the target catchment and consequently produce unusable results.
TABLE 2.5.1: PARAMETER DERIVATION SCHEME FOR THE FULL TRANSPOSITION (FTM) AND PARTIAL
TRANSPOSITION METHODS (PTM). WHERE: DONOR REPRESENTS TRANSPOSITION FROM THE DONOR CATCHMENT CALIBRATION TO THE TARGET CATCHMENT, TARGET REPRESENTS THAT ARE NOT TRANSPOSED FROM THE DONOR CATCHMENT AND DEFAULT REPRESENTS PARAMETERS THAT ARE KEPT AT THE DEFAULT VALUES.

Parameters
Interception storage (mm) Impermeable Proportion Time to peak (hrs) Rooting Depth (mm) Pore size distribution index Permeability (horizon boundary) (mm/hr) Permeability (base lower) (mm/hr) Interflow (upper) (mm/hr) Interflow (lower) (mm/hr) Groundwater recession (per month) Precipitation factor PET factor Catchment Area (sq km)

FTM
DONOR DONOR DONOR DONOR DONOR DONOR DONOR DONOR DONOR DONOR DEFAULT DEFAULT TARGET

PTM
DONOR TARGET DONOR DONOR TARGET DONOR DONOR DONOR DONOR TARGET DEFAULT DEFAULT TARGET

27

The parameter-set values for the partial transposition method are presented for reference on appendix 8.4. Inherently, this also provides the parameter-set values for the full transposition method following the scheme provided on table 2.5.1. Also provided are the optimal parameter-sets resulting from the calibration of each of the donor catchments, which illustrate the process of parameter transposition to the target catchments parameter-set. The parameter values references in appendix 8.4 are considered within reasonable ranges for computation in Hysim; indicated by the notes given within the programme and also according to the Hysim User Guide (Manley, 2006).

2.6

Evaluating model performance using the proxy-basin test

In-keeping with the method for proxy-basin validation outlined by previous studies, a method was developed for model performance evaluation. In standard proxy-basin methodology a calibrated, optimal parameter-set from a gauged donor catchment is applied to a gauged target catchment. A measure of efficiency of the flow estimation in the target catchment is then made, often delivered by the model itself, providing an interpretation of the quality of the model. This methodology was appropriated as the framework for model evaluation used by this study. A measure of model efficiency is widely backed up in literature as the accepted method of evaluating model performance. However, there was emphasis within this study to evaluate model performance using a visual measure of accuracy between simulated and recorded flow. An FDC was used as an illustrative comparison tool between simulated and recorded flows. Westerberg (2011) indicates that there is good correlation between model efficiency and the accuracy of a simulated FDC. As such, a measure of accuracy in terms of Q95, MF and Q95(%MF) flow descriptors is used throughout this to evaluate the quality of model simulations. Moreover, an Annual Hydrograph of Daily Mean Flows (HDF) will be compared for estimated and recorded flows to ensure there is suitably high model efficiency.

28

3 Results
3.1 Hydrological statistics of mega-zones and SEPA catchments
Comparisons of statistics such as catchment area, SAAR, BFI and Q95(%MF) for every catchment as well as the plots of these statistics for average catchments in Scottish Water mega-zones gave a useful illustration of how representative the selected catchments were of typical Scottish Water catchments. These plots also allowed an illustration of key relationships between hydrological statistics that highlight any erroneous catchments. Mega-zones, as previously discussed (figure1.2.1), are regional groupings of source catchments catchments with a water source for abstraction by Scottish Water. This comparison is invaluable in order to interpret the degree of representativeness of the catchments in this study. Typically, catchments for suitable comparison aimed to be no larger than 50km2 in area; in accordance with comments made by staff at Scottish Water. Exceptional SEPA catchments were included in the selection of catchments due to their specific relevance to Scottish Water operations. It was reported that Low Malzie, with a catchment size of 328.51km2 was used because there were no other suitable gauged records for a Q95(%MF) below 5%. Skeabost, which has a catchment size of 80.55km2, was used due to a need for representation on the Isle of Skye despite its short record. Maidencots, which has a catchment area of 111.50km2, was selected for its availability of data at the Q95(%MF) 13% - 20% range, a factor that was unavailable amongst other catchments. It is clear from these compensations that SEPA approved analogues are considerably limited in their data availability. This is the inherent nature of modelling catchment flow in areas with a limited gauged flow record such as Scotland.

29

FIGURE 3.1.1: COMPARISON BETWEEN THE CATCHMENT CHARACTERISTICS OF BASE-FLOW INDEX (BFI) AND

Q95(%MF) FOR SCOTTISH WATER MEGA-ZONES AND 24 SEPA GAUGED CATCHMENTS. CATCHMENTS THAT WERE ACCEPTED FOR EXPERIMENTATION ARE COLOURED ACCORDING TO Q95(%MF) GROUP ALLOCATION. A plot of BFI and Q95(%MF) (figure 3.1.1) illustrates a good, positive correlation between BFI and Q95(%MF) for both series. A 0.1 increase in BFI is associated with an increase of 4% Q95(%MF) for the plot of the mega-zones. The superimposed catchments are somewhat constrained within the plot of the mega-zones, especially in the 0.20 to 0.45 range for BFI. Damleys Cottage, Kinross and Lathro plot outside of the trending mega-zones and remaining catchments. These three catchments display a BFI that is too large with respect to their Q95(%MF). Critically, there are no mega-zones that are below 7% Q95(%MF); however, nine catchments represent Q95(%MF) below 7%.

FIGURE 3.1.2: COMPARISON BETWEEN THE CATCHMENT CHARACTERISTICS OF BASE-FLOW INDEX (BFI) AND SAAR FOR SCOTTISH WATER MEGA-ZONES AND 24 SEPA GAUGED CATCHMENTS. CATCHMENTS THAT WERE ACCEPTED FOR EXPERIMENTATION ARE COLOURED ACCORDING TO Q95(%MF) GROUP ALLOCATION.

Plotting BFI and SAAR (figure 3.1.2) shows a good, negative correlation between BFI and SAAR for both series. A 0.1 decrease in BFI is associated with an approximate increase 500mm for the plot of the mega-zones. The superimposed catchments are somewhat constrained within the plot of the megazones, especially in the 0.20 to 0.5 range for BFI. Damleys Cottage, Kinross and Lathro once again plot outside the trending mega-zones and catchments. These three catchments display a BFI that is too large with respect to their SAAR value. There are no mega-zones that are exceed a SAAR of 2450mm; however, eight catchments represent SAAR above 2450mm; seven of which also fell below the mega-zone statistics for BFI and Q95(%MF).

30

FIGURE 3.1.3: COMPARISON BETWEEN THE CATCHMENT CHARACTERISTICS OF BASE-FLOW INDEX (BFI) AND CATCHMENT AREA FOR SCOTTISH WATER MEGA-ZONES AND 24 SEPA GAUGED CATCHMENTS. CATCHMENTS THAT WERE ACCEPTED FOR EXPERIMENTATION ARE COLOURED ACCORDING TO Q95(%MF) GROUP ALLOCATION.

A plot of BFI and catchment area, shown on figure 3.1.3, finds no identifiable correlation between BFI and catchment size for either the mega-zones or catchments. The bulk of mega-zones plot within 0.25-0.45 BFI and an area of 20km2.White Laggan, Green Burn, Dargall Lane and Polloch are the only catchments to fall within this plot. A variety of catchment sizes are associated with a variety of catchment areas.

3.2

Calibration quality of donor catchments

A comparative presentation of estimated and recorded flow is essential in order to evaluate the quality of the optimal parameter-set achieved. This is achieved through the representation of the eight years of flow as an Annual Hydrograph of Daily Mean Flows (HDF), Flow Duration Curve (FDC) and flow Q95, MF and Q95(%MF) flow descriptors. The accuracy of these optimal parameter-sets in estimating observed flow divulges the confidence that we can have in the parameter-sets established from these parameters.

Calibration for Q95(%MF) 5-7% group: Braevallich, 1993-2002


For the Q95(%MF) 5-7% group the Braevallich catchment was calibrated for the transposition of parameters. The optimal parameter-set (appendix 8.4) was suitable with no parameters that deviate significantly from typical values expected for such a catchment; they also fell within recommended value ranges (Manley, 2006). Additionally, it should be noted that the PET single parameter optimisation yielded a value of 1; therefore, there is no single parameter optimisation within this calibration.

31

FIGURE 3.2.1: BRAEVALLICH SIMULATED AND RECORDED ANNUAL HYDROGRAPH OF DAILY MEAN FLOWS (HDF), AVERAGE GIVEN FOR THE EIGHT YEAR PERIOD OF STUDY.

Referring to the HDF on figure 3.2.1, there is a close relationship between the timing of fluctuations from high to base flows between the simulated and observed oscillations, there is no significantly consistent lag. The simulation often underestimates base flows and overestimates peak flows, which it does to the same scale throughout the year. Occasionally there are misinterpretations of rapid fluctuations between low and peak flow for small storms, sometimes the opposite peak is observed; see mid-April. There is no clear distinction between seasons, a dry period is rapidly asserted from March to May, which slowly recovers until the wet season is constantly apparent in September. There appears to be no variation in the accuracy of flow estimation related to the seasonality of flow.

FIGURE 3.2.2: BRAEVALLICH SIMULATED AND RECORDED FLOW DURATION CURVE (FDC) COMPARISON OF FLOW FOR THE EIGHT YEAR PERIOD.

32

The simulated FDC fits very well to the recorded FDC; areas of best fit are the 5th to 15th percentile and the 85th to the 95th percentile (figure 3.2.2). Looking lower than the 5th percentile and higher than the 95th percentile the simulated and observed flows become gradually mismatched; with the simulated flow unable to represent the recorded FDC at the 1st and 99th percentile. In the 20th to 80th percentile range there a greatening difference between simulated and recorded FDC until becoming quite a uniform difference towards the 50th percentile.
TABLE 3.2.3: BRAEVALLICH SIMULATED AND RECORDED FLOW DESCRIPTOR NUMERICAL AND PERCENTAGE DIFFERENCES FOR THE EIGHT YEAR PERIOD. BRAEVALLICH IS CONSIDERED AN EXCELLENT CALIBRATION.

Q95 (m3s-1) MF (m3s-1) Q95(%MF)


Simulated flow Recorded flow Numerical difference Percentage difference 0.089 0.088 0.001 1.218% 1.605 1.587 0.017 1.083% 5.534% 5.526% 0.008% 0.137%

The Braevallich simulated flow descriptors are very close to that of those of the recorded flow (table 3.2.3). Q95 of the simulated flow has a numerical difference of 0.001 m3s-1 when compared to the simulated flow; accounting for a percentage difference of 1.20%. The numerical difference between MF for the Braevallich catchment is 0.017m3s-1, translating to a percentage difference of 1.08%. The Braevallich catchment has an extremely close correlation between the simulated 95th percentile as a proportion of mean flow and the recorded value for this statistic, the numerical difference is 0.008%. This is the calibration with the lowest difference between simulated and recorded Q95(%MF) values.

Calibration for Q95(%MF) 7-9% group: Durkadale, 2000-2009


For the Q95(%MF) 7-9% group, the Durkadale catchment was used to as a donor catchment and provided the optimal parameter-set for the group, the optimal parameter-set is illustrated in appendix 8.4. The parameters suit the nature of the catchment and do not exceed the recommended value ranges stated by Manley (2006). The calibration of this model did not follow the single parameter optimisation method outlined by Manley (2006) and relied upon adjustment of the precipitation factor only. Notes that describe the reasoning behind changing the single optimised parameter from PET to precipitation are outlined in appendix 8.5.

FIGURE 3.2.4: DURKADALE SIMULATED AND RECORDED ANNUAL HYDROGRAPH OF DAILY MEAN FLOWS (HDF), AVERAGE GIVEN FOR THE EIGHT YEAR PERIOD OF STUDY.

33

The HDF of both simulated and recorded flow (figure 3.2.4) oscillate at the same time with no consistent lag. The simulation overestimates peak flows and underestimates base flows most significantly for minor storms, where major storms within the same month are very well estimated. There are no misinterpretations of the fluctuation in flow by the simulation, under and overestimations appear to be the overriding factors. There is a clear distinction between seasons, a symmetrical dry period gradually reaches its lowest flows in June and returns to a wet season in September. There appears to be considerable variation in the accuracy of flow estimation related to the current season being estimated. Overestimations of low and peak flows are most prevalent in the transition period from wet to dry season and underestimation is more prevalent from May to August.

FIGURE 3.2.5: DURKADALE SIMULATED AND RECORDED FLOW DURATION CURVE (FDC) COMPARISON OF FLOW FOR THE EIGHT YEAR PERIOD.

The simulated and recorded FDCs (figure 3.2.5) have the best fit for the 10th to 60th percentile, where the FDCs are undistinguishable from each other. This is very useful for improving the similarity of MF between simulated and recorded flow, which is required due to the inconsistent nature of the recorded FDC. The FDCs fit most poorly up to the 10th percentile of flow yet below the 1st percentile appears to be represented well by the simulation, helped by the greater curvature of the simulated FDC at this low percentile. The simulated FDC becomes less suitable toward the 99th percentile as the simulation fails to capture the same curvature as the recorded data. This was a factor of forcing the FDC to match the inconsistent curvature of the recorded flow. The poor match at the 97th to 99th percentile was necessary during modelling for the good estimation of other percentiles.

34

TABLE 3.2.6: DURKADALE SIMULATED AND RECORDED FLOW DESCRIPTOR NUMERICAL AND PERCENTAGE DIFFERENCES FOR THE EIGHT YEAR PERIOD. DURKADALE IS CONSIDERED A GOOD CALIBRATION.

Q95 (m3s-1) MF (m3s-1)


Simulated flow Recorded flow Numerical difference Percentage difference 0.046 0.042 0.004 7.728% 0.489 0.515 0.026 5.082%

Q95(%MF)
9.317% 8.160% 1.157% 12.418%

For the Durkadale calibration the numerical difference of Q95 between simulated and recorded flow is 0.004m3s-1, accounting for a percentage difference of 7.728% (table 3.2.6). The numerical difference of MF was 0.026m3s-1 and MF percentage difference of 5.082%. The numerical difference between simulated and recorded Q95(%MF) for the Durkadale calibration is 1.157%, over two times greater in magnitude Braevallich. The Durkadale calibration has the greatest difference between simulated and recorded Q95(%MF).

Calibration for Q95(%MF) 9-11% group: Luss, 1988-1997


For the Q95(%MF) 9-11% group, the Luss catchment was used to as a donor catchment and provided the optimal parameter-set for the group, which can be seen in appendix 8.4. The parameters are typical of the catchment type and do not exceed the suitable ranges stated by Manley (2006). The calibration of this model followed the standard calibration procedure outlines in the method of this study and required the single parameter optimisation of evapotranspiration and subsequently precipitation factor in order to achieve a suitable fit from which manual adjustment could be made.

FIGURE 3.2.7: LUSS SIMULATED AND RECORDED ANNUAL HYDROGRAPH OF DAILY MEAN FLOWS (HDF) , AVERAGE GIVEN FOR THE EIGHT YEAR PERIOD OF STUDY.

Simulated and recorded flows for the HDF (figure 3.2.7) have a good degree of synchronicity for high and base flow oscillation, there is little lag between plots. The simulation overestimates peak flows and underestimates base flows throughout the year, although base flow underestimation is less frequent. There some misinterpretations of minor storms by the simulation, occasionally illustrating the opposite fluctuation to the recorded flow. There is a clear distinction between seasons, a somewhat symmetrical dry period gradually reaches its lowest flows in June and returns to a wet season in late September. There appears to be no significant variation in the accuracy of flow estimation related to the current season being estimated.

35

FIGURE 3.2.8: LUSS SIMULATED AND RECORDED FLOW DURATION CURVE (FDC) COMPARISON OF FLOW FOR THE EIGHT YEAR PERIOD.

The areas of best fit for the simulated and recorded FDCs (figure 3.2.8) are the 10th to 50th percentile, where the FDCs are undistinguishable from each other. The FDCs fit most poorly up to the 10th percentile of flow yet below the 1st percentile appears to be represented well by the simulation, helped by the greater curvature of the simulated FDC at this low percentile. The simulated FDC underestimates the flows from the 55th percentile to the 95th percentile. The 95th percentile is approximated well by the simulation; however, it becomes less suitable toward the 99th percentile as the simulation fails to match the same curvature as the recorded data.
TABLE 3.2.9: LUSS SIMULATED AND RECORDED FLOW DESCRIPTOR NUMERICAL AND PERCENTAGE DIFFERENCES FOR THE EIGHT YEAR PERIOD. LUSS IS CONSIDERED AN EXCELLENT CALIBRATION.

Q95 (m3s-1) MF (m3s-1) Q95(%MF)


Simulated flow Recorded flow Numerical difference Percentage difference 0.212 0.223 0.011 4.975% 2.596 2.648 0.051 1.934% 8.165% 8.427% 0.261% 3.101%

The calibration of Luss resulted in a numerical difference of Q95 between simulated and recorded flow of 0.011m3s-1, accounting for a percentage difference of 4.98% (table 3.2.9). The numerical difference of MF was 0.051m3s-1 and MF percentage difference of 1.93%. The numerical difference between simulated and recorded Q95(%MF) for the calibration is 0.26%; a greater difference when compared to Braevallich and smaller difference when compared to the Durkadale calibration.

36

Calibration for Q95(%MF) 11-13% group: Lathro, 1992-2001


For the Q95(%MF) 11-13% group, the Lathro catchment was used to as a donor catchment and provided the optimal parameter-set for the group, the optimal parameter-set is illustrated in appendix 8.4. The parameters represent the catchment suitably and are typical of the catchment type and do not exceed the recommended value ranges stated by Manley (2006). The calibration of this model did not follow the standard calibration procedure in that the standard single parameter optimisation method outlined by Maney, 2008. Notes that describe the reasoning behind changing the single optimised parameter from PET to precipitation are outlined in appendix 8.5.

FIGURE 3.2.10: LATHRO SIMULATED AND RECORDED ANNUAL HYDROGRAPH OF DAILY MEAN FLOWS (HDF), AVERAGE GIVEN FOR THE EIGHT YEAR PERIOD OF STUDY.

Looking at the HDF on figure 3.2.10, simulated and recorded flow values fluctuate from low to peak flows synchronously; there is some lag between the plot of some base flows yet this lag results from poor estimation. The simulation estimates peak flows well, with only occasional overestimation; base flows are constantly over and underestimated. There are few misinterpretations of fluctuations by the simulation. There is some distinction between seasons, an asymmetrical dry period gradually reaches its lowest flows in July and rapidly returns to a wet season in late August. There is significant variation in the accuracy of flow estimation related to the current season being estimated. Base flows are overestimated in the wet season and underestimated in the dry season.

37

FIGURE 3.2.11: LATHRO SIMULATED AND RECORDED FLOW DURATION CURVE (FDC) COMPARISON OF FLOW FOR THE EIGHT YEAR PERIOD.

The best fitting areas for the simulated and recorded FDCs (figure 3.2.11) are the 40th to 55th percentile and 85th to 95th percentile, where the FDCs are undistinguishable from each other. The 95th to 99th percentile appear to have a good match as well as the 1st to 5th percentiles. The FDCs fit most poorly in the interim, where the curve of the recorded FDC is unusually distributed. This is the area between the 5th and 35th as well as the 55th to 85th percentile.
TABLE 3.2.12: LATHRO SIMULATED AND RECORDED FLOW DESCRIPTOR NUMERICAL AND PERCENTAGE
DIFFERENCES FOR THE EIGHT YEAR PERIOD. LATHRO IS CONSIDERED A GOOD TO EXCELLENT CALIBRATION.

Q95 (m3s-1) MF (m3s-1)


Simulated flow Recorded flow Numerical difference Percentage difference 0.079 0.078 0.001 1.429% 0.702 0.701 0.001 0.153%

Q95(%MF)
11.271% 11.127% 0.144% 1.277%

The Lathro calibration provided a numerical difference of Q95 between simulated and recorded flow of 0.001m3s-1, accounting for a percentage difference of 1.43% (table 3.2.12). The numerical difference of MF was 0.001m3s-1 and MF percentage difference of 0.15%. The numerical difference between simulated and recorded Q95(%MF) for the calibration is 0.144%, which is a greater difference than Braevallich and smaller difference than the Luss calibration.

38

3.3

Evaluating model performance with transposition method chosen

Percentage differences between recorded and estimated flow were calculated for the flow descriptors Q95, MF and Q95(%MF) for the simulation of each transposition method. Evaluating the accuracy of each transposition method elaborates the uncertainty associated with the use of such methods. As such, percentage differences between the simulated and recorded flow descriptors were used as a quanitifcation of uncertainty for the parameter-set construction methods. The full acount of numerical and percentage differences of flow descriptors for both recorded and simulated flow are located on appendix 8.6.

Evaluating the simulation of Q95 for target catchments


The percentage difference of Q95 for each of the nine target catchments for each four groups is presented on figure 3.3.1.

FIGURE 3.3.1: ACCURACY OF Q95 ESTIMATIONS BY THE FULL TRANSPOSITION METHOD (FTM) AND PARTIAL TRANSPOSITION METHOD (PTM). ACCURACY DEMONSTRATED IN TERMS OF PERCENTAGE DIFFERENCE BETWEEN SIMULATED AND RECORDED DATA.

For the 5-7% group, the FTM simulations of Q95 have the lowest percentage difference for the Deep Hope catchment, at 4.4%, wheras Glen Strae and Polloch catchments have 12.1% and 12.4% respectively. PTM simlations follow a similar trend in estimation with Deep Hope having the least percentage difference at 19.2% and Glen Strae and Polloch having 21.0% and 28.9% respectively. The FTM simulations are consistently lower in percentage difference than those of the PTM, implying less uncertainty in the FTM and a greater uncertainty in the PTM. In the 7-9% group, the FTM simulations of Q95 have the lowest percentage difference for the
39

Barsolus catchment, at 28.6%, wheras Inverlochy and Skeabost catchments have 34.3% and 41.7% respectively. PTM simulations do not follow this trend; the least percentage difference is again Barsolus at 12.2% but with Skeabost the Inverlochy having 30.3% and 42.6% respectively. The PTM simulations are lower in percentage difference than those of the FTM two out of three times. This cannot exclusively imply less for more uncertainty in the use of the FTM or PTM. For the 9-11% group, the simulations of Q95 have the lowest percentage difference for the Candermill catchment, at 20%, wheras Creed Bridge and Dargall Lane catchments have 8.7% and 27.8% respectively. The PTM does not follow a similar trend in estimation, however Deep Hope has a similar value to that of the FTM simulation at 32.4%; Candermill and Creed Bridge have values of 76.8% and 52.6% respectively. The FTM simulations are consistently lower in percentage difference than those of the PTM, implying less uncertainty in FTM and a greater uncertainty in PTM for this group. In the 11-13% group, the FTM simulations of Q95 have the lowest percentage difference for the Whitburn catchment, at 27.5%, wheras Brockhoperig and Kinross catchments have 38.9% and 32.9% respectively. PTM simulations follow a similar trend, with Whitburn having the least percentage difference at 26.6% and Brockhoperig and Kinross having 47.5% and 35.0% respectively. The FTM simulations are consistently lower in percentage difference than those of the PTM with the exception of the Whitburn simulation, which is lower for PTM. As the simulations by the FTM and PTM are only 1% apart in the plots for the Whitburn catchment this indates less uncertainty for the FTM and a greater uncertainty for the PTM in this group.

40

Evaluating the simulation of MF for target catchments


The percentage difference of MF for each of the nine target catchments for each four groups is presented on figure 3.3.2.

FIGURE 3.3.2: ACCURACY OF MF ESTIMATIONS BY THE FULL TRANSPOSITION METHOD (FTM) AND PARTIAL TRANSPOSITION METHOD (PTM). ACCURACY DEMONSTRATED IN TERMS OF PERCENTAGE DIFFERENCE BETWEEN SIMULATED AND RECORDED DATA.

For the 5-7% group, the FTM simulations of MF have the lowest percentage difference for the Deep Hope catchment, wheras Glen Strae and Polloch catchments have a much greater percentage difference when decreasing the natural Q95(%MF). PTM results for Deep Hope, Glen Strae and Polloch are almost exactly the same as those of the FTM and therefore there no way of determining which model could be considered better for estimations in this group. Observing the 7-9% group, the FTM simulations of MF have the lowest percentage difference for the Skeabost catchment, wheras Barsolus and Inverlochy catchments have a considerably greater percentage difference when decreasing the natural Q95(%MF), both plateauing at 14% higher than Skeabost. PTM results for Skeabost, Barsolus and Inverlochy are almost exactly the same as those of the FTM and therefore there no way of determining which model could be considered better for estimations in this group. Looking toward the 9-11% group, the FTM simulations of MF have the lowest percentage difference for the Dargall Lane catchment, wheras Creed Bridge and Candermill catchments have a considerably greater percentage difference when decreasing the natural Q95(%MF). PTM percentage differences are lowest for Dargall Lane, Candermill is the same for the FTM and Creed Bridge shows a marginally smaller percentage difference than the FTM. The FTM simulations have lower
41

percentage differences than the PTM on the whole; however, the PTM is better at estimations for Candermill. It is FTM that appears to have less uncertainty associated with it due to the lower percentage difference for Dargall Lane. For the 11-13% group, the FTM simulations of MF have the lowest percentage difference for both the Kinross and the Shitburn catchments, wheras Brockhoperig is 4% higher. PTM results for Brockhoperig, Kinross and Whitburn are almost exactly the same as those of the FTM with the exception of Kiross, where there is a marginally better simulation than the FTM. However, this is so slight that there is no clarity in determining which model could be considered better for simulations in this group.

Comparing simulations of Q95(%MF) to recorded flow for target catchments


Due to the similarity between FTP and PTM simulations of MF, the expression of Q95 as a percentage of MF would yield the same results as Q95 estimations, which are already described on figure 3.3.1. It is thereforemore pertinent to study the plot of Q95(%MF) in relation to the natural recorded flow of the catchments and the actual values of estimated Q95(%MF), presented on figure 3.3.3.

FIGURE 3.3.3: ESTIMATION OF Q95(%MF) BY THE FULL TRANSPOSITION METHOD (FTM) AND PARTIAL TRANSPOSITION METHOD (PTM). COMPARISON IS MADE WITH THE RECORDED Q95(%MF) VALUE OF THE CATCHMENT.

For the 5-7% group, the FTM simulations of Q95(%MF) are best for the Glen Strae and Deep Hope catchments, almost indistinguishable from their recorded Q95(%MF) values. The FTM overestimates the Polloch catchment by approximately 2%. The PTM overestimates the Polloch and Deep Hope catchments and understimates the Glen Strae catchment. The recorded Q95(%MF) values for the group
42

are constrained tightly in the 5-7% margins, which is indicative of their UKHR derived Q95(%MF) values. The FTM simulations fall within the 5-7% criteria for the group; however, all of the PTM simulations exceed the limits of the group. Looking toward the 7-9% group, the FTM simulations for Q95(%MF) are best for the Barsolus catchment, within 2% of the recorded Q95(%MF) value. The FTM overestimates the Inverlochy and Skeabost catchments significantly. The PTM understimates the Barsolus catchment and overestimates Inverlochy and Skeabost catchments. The PTM produces estimates that are closer to the recorded Q95(%MF) values for the Barsolus and Inverlochy catchments than the FTM. The recorded Q95(%MF) values for the group are constrained tightly in the 7-9% margins, which is indicative of their UKHR derived Q95(%MF) values. The PTM and FTM simulations do not fall within the 7-9% criteria for the group except for the Barsolus catchment. Observing the 9-11% group, the FTM estimation of Q95(%MF) is best for the Creed Bridge catchment, which is indistinguishable from the recorded Q95(%MF) value. The FTM overestimates the Inverlochy and Skeabost catchments significantly. The PTM understimates the Candermill catchment and overestimates the Dargall Lane catchment. The PTM simulations are significantly outside the range of the recorded Q95(%MF) values for all catchments, understimating the recorded values. Dargall lane is the closest to the recorded Q95(%MF). The recorded Q95(%MF) values for the group are broadly distributed from the 8-12% margins, which is somewhat lower than their UKHR derived Q95(%MF) values. As such, the FTM simulations do not fall within the 11-13% criteria for the group but fall close to the 812% range of the recorded Q95(%MF) values. The PTM simulations entirely understimate the recorded Q95(%MF) values and occupy a range that is from 2-6%. In the the 11-13% group, the FTM estimation of Q95(%MF) is best for the Whitburn catchment, which comes within 2% of the recorded Q95(%MF). The FTM has a similar result for the estimation of the Kinross catchment but extremely overestimates the recorded value at Brockhoperig. The PTM produces similar estimates to the FTM, with Whitburn and Kinross plots almost identical between the models. The overestimation of Brockhoperig is 4% greater in the estimation made by the PTM for this catchment. The recorded Q95(%MF) values for the group are tightly distributed from 9-11% margins, which is lower than their UKHR derived Q95(%MF) values, which should lie between 11-13%. As such, the FTM simulations do not fall within the 11-13% criteria either and occupy a range of approximately 8% for the Whitburn and Kinross plots, whilst exceeding 21% for Brockhoperig.

43

3.4

Evaluating model performance with selection of target catchment

In order to evaluate the effect of target catchment selection upon accuracy of the model the percentage difference of Q95(%MF) between estimated and recorded flow was used. Plotting the hydrological statistics shown on table 2.2.1 against the evaluation criteria illustrates the level of uncertainty associated with catchment statistics of the target catchments. This is done for the two methods of parameter-set construction in order to see the relative uncertainty present.

FIGURE 3.4.1: TRANSPOSITION METHOD PERFORMANCE FOR CATCHMENTS RELATED TO THEIR Q95(%MF). PERFORMANCE IS DEMONSTRATED BY ACCURACY: GIVEN AS THE PERCENTAGE DIFFERENCE BETWEEN SIMULATED AND RECORDED VALUES OF Q95(%MF).

A positive correlation between recorded Q95(%MF) of the selected time period and uncertainty appears to exist (figure 3.4.1). This relationship is seen most clearly in the FTM; catchments with a higher Q95(%MF) have a greater uncertainty, with the exception of Creed Bridge, which has a typically low uncertainty for its Q95(%MF). Inverlochy and Skeabost have greater uncertainty for their Q95(%MF) with relation to the general trend of target catchments in this plot.

44

FIGURE 3.4.2: TRANSPOSITION METHOD PERFORMANCE FOR CATCHMENTS RELATED TO THEIR BASE FLOW INDEX (BFI). PERFORMANCE IS DEMONSTRATED BY ACCURACY: GIVEN AS THE PERCENTAGE DIFFERENCE BETWEEN SIMULATED AND RECORDED VALUES OF Q95(%MF).

It is possible to see a relationship between an increase in uncertainty and an increase in BFI; the higher the BFI, the higher the uncertainty (3.4.2). This positive correlation is most clearly seen in the estimations for the FTM, with the exception of Creed Bridge, which has an unusually low uncertainty even though it has the largest BF. The plot of PTM simulations does not show the relationship as clearly, although it is somewhat identifiable if Kinross and Barsolus catchments are not included.

45

FIGURE 3.4.3: TRANSPOSITION METHOD PERFORMANCE FOR CATCHMENTS RELATED TO THEIR AREA. PERFORMANCE IS DEMONSTRATED BY ACCURACY: GIVEN AS THE PERCENTAGE DIFFERENCE BETWEEN SIMULATED AND RECORDED VALUES OF Q95(%MF).

A limited relationship between catchment area and uncertainty can be identified in the plot of FTM and PTM simulations on figure 3.4.3. A variety of uncertainties are associated with the size of the catchment. Skeabost has over double the catchment area of the other target catchments. It would be expected that this would result in a very large uncertainty if there was such a relationship, which is not the case. Most catchments that are below 40km2 have a Q95(%MF) below 30%. Those above 40km2 appear to contrast, with Q95(%MF) above 40%.

46

4 Discussion
4.1 Hydrological statistics of mega-zones and SEPA catchments
Interpreting the representativeness of SEPA catchments provided for the study in relation to typical Scottish Water catchments was important. For any useful contribution to Scottish Waters current Hysim modeling strategy to be made, the catchments used in experimentation must befit typical Scottish Water catchments. It was also necessary to identify those catchments that had hydrological characteristics that would perturb the results of a fair validation test. In comparing the BFI of SEPA approved catchments to that of the Scottish Water mega-zones a number or relationships between BFI and various catchment statistics can be understood. The relationship between BFI and Q95(%MF) is that of a linear relationship. A greater BFI lower flashiness- is associated with a greater Q95(%MF). This is an expected relationship as BFI represents the proportion of low flows for a catchment, which is almost analogous to greater Q95 as a proportion of MF. A lower BFI was found be related to a higher SAAR across the mega-zones and catchments. If there is more water availability, it would be expected that the base flow is a reduced proportion of flow; thus confirming this relationship. The identification of these logical relationships between hydrological statistics helps to identify any erroneous catchments that may require interrogating before including in the experiment. In the plot of these two relationships there were three catchments that failed to represent the trend and showed elevated BFI for the compared hydrological statistic; these were: Damleys Cottage, Kinross and Lathro. As a consequence Damleys Cottage and Kinross were rejected from the experimentation portion of this study. Lathro was continued into the experimentation phase of the study and selected as a donor catchment, which was calibrated as a source of derivation for parameter transposition. This selection was made in order to observe the effects of including an elevated BFI as the calibration of the study and partly due to the poor availability and suitability of alternative donor catchments. The bulk of mega-zones plotted have a BFI of 0.25 to 0.45 and the majority of SEPA catchments did plot within this range; however, with respect to the comparison statistic the SEPA catchments often lied outside that of mega-zone values. When considering the BFI and Q95(%MF) plot, there are no megazones that plot below a Q95(%MF) of 7%. This represents the variety of catchment Q95(%MF) values that are averaged for each Scottish Water mega-zone; elucidating the need for ROI in grouping catchments. Nine catchments had a Q95(%MF) of 7%, in a bid to relate the experiment closer to the character of the mega-zones it was chosen that catchments below 5% would be omitted from the experimentation. This also eliminated uncertainties associated with choosing catchments with a Q95(%MF) close to naught and brought more focus to the investigation of uncertainties associated with the aims of this study. As such, it was also necessary to omit a number of high Q95(%MF) catchments; therefore, catchments exceeding a Q95(%MF) of 13% were not used in experimentation. Eight of the twenty-four SEPA approved analogues were omitted from the experiment due to their lack of hydrological representation in relation to Scottish Water mega-zones. The remaining sixteen catchments were then grouped according to Scottish Waters outline of ROI grouping by Q95(%MF). Four groups were created, with each group representing a range of 2% Q95(%MF). A single donor catchment was then chosen according to their representativeness as well as rate and record of data.

47

4.2 Calibration quality of donor catchments


A high level of experience is necessary for the calibration of hydrological models. The user must be familiar with the model and its software and be aware of its limitations and strengths with regard to the input data being used. This high level of experience was achieved for the purposes of this study through the perseverance in providing high quality calibrations of donor catchments. In some instances the most suitable catchment for calibration was not often possible to calibrate. Most significantly, this occurred when attempting to calibrate the Candermill catchment. After sixty attempts at calibration using the method outlined in this study there was insignificant improvement to the quality of the calibration. In this case, an alternative donor catchment was selected and used for calibration. Jacobs (2010) identified the calibration of Candermill as an illustration of the inadequacies of Hysim. All optimal parameter-sets achieved through calibration are described as satisfactory or better. The manual calibration step of each model was performed to a point where there was a large step in advancement of the quality of the model without the choice of unrealistic parameters that would be evident of over-parameterisation. It is considered that the quality of the donor parameter-sets here is optimal for the experience of the user, as no improvement can be found without detriment to the flow estimations. The accuracy of these optimal parameter-sets in estimating observed flow divulges the confidence that we can have in the parameter-sets established from these parameters. Accuracy in simulation of eight years of flow, represented as: Annual Hydrograph of Daily Mean Flows (HDF), Flow Duration Curve (FDC) and flow Q95, MF and Q95 (%MF) flow descriptors, is the manner in which the quality of a calibration was assessed. An HDF provides the daily mean flows over the eight year period of record, excluding the two year warm-up period. As such, it lacks the comparison of day scale large fluctuations in flow, which often mar the performance comparison of a simulation to recorded flow values. Using an HDF, there is no representation of a change in the annual cycle of flow; therefore this must be checked with the Longterm Daily Hydrograph (LDF) if there are unusual patterns of over or underestimation by the simulation. This was found to be the case for Durkadale, where selection of the flow period included a recent and cyclically variable flow record. An FDC is a cumulative curve that shows the percentage of occurrence that specified discharged were equalled or exceeded during the given time without regard to the sequence of occurrence. In comparing the simulated FDC to the recorded FDC, it relates to the performance of the model calibration in types of flow, and with specific relevance to this investigation: low flows. Observing the difference between simulated and recorded flow descriptors of the catchment i.e. Q95, MF as well as Q95(%MF) it is possible to see the simulations performance in representing various aspects of flow. This is important as a simulation can illustrate high performance in the estimation of one aspect of flow whilst showing poor performance in another. These flow descriptors are used in conjunction with the other methods of representation in order to thoroughly evaluate the quality of the calibration and achieve an optimal parameter-set. The descriptors of flow for a catchment are a direct method of evaluating the methods of parameter transposition as they represent various aspects of flow and are used as such in later discussion of model performance.

48

Q95(%MF) 5-7% group, Braevallich calibration


Simulated, average, daily flow values presented on the HDF have an excellent closeness to the recorded values. There is no lag in the oscillation of the simulated in relation to the recorded values, which indicates that the total soil storage -rooting depth- parameter is adequate. Minor oscillations are not always represented, although it is typical for models to be unable to represent natural small scale fluctuations. The underestimation of base and peak flows does not appear to be restricted to summer or winter periods thus confirming the correct choice of the proportion of soil storage in the upper horizon, which would require lowering if summer storms were underestimated. Whilst there is some underestimation of base and peak flows, it is not significant enough to detract from the overall excellent simulation of daily flows that indicates an excellent calibration. The simulated FDC fits very well to the recorded FDC. Flow estimations are unable to represent the recorded flow at the 1st and 99th percentile; however, this is typical of Hysim simulations, they are oft unable to represent the natural variation of flow. This is reflected in the HDF, for which there are occasional underestimations of low flows. The estimated Q95 differs on the scale of a cubic millimetre, a level of precision that is indicative of an excellent calibration. A percentage difference of 1.20% between estimated and recorded flow confirms the quality of this calibration in estimating Q95. MF differs on the scale of a cubic centimetre, again indicating a high level of precision for the calibrated model. The percentage difference is improved for MF to 1.08%, which again confirms the excellent quality of this calibration. Expressing the percentage differences of Q95 and MF as Q95(%MF), which is the chosen measure of accuracy for simulations in this study, a difference of 0.14% is calculated. From the analysis of the flow estimations provided by the calibration simulations, it is determined that the Braevallich catchment calibration produces an excellent estimation of HDF, FDC, Q95 and MF. This calibration is therefore considered an excellent calibration; the utmost confidence can be had in the derivation of parameters from its constituent optimal parameter-set.

Q95(%MF) 7-9% group, Durkadale calibration


Simulated, average, daily flow values presented on the HDF have a satisfactory fit to the recorded values. There is no lag in the oscillation of the simulated in relation to the recorded values, which indicates that the total soil storage -rooting depth- parameter is adequate. Minor oscillations are often well represented, which is a sign of good calibration quality. The underestimation of base and peak flows does not appear to be restricted to summer or winter periods thus confirming the correct choice of the proportion of soil storage in the upper horizon, which would require lowering if summer storms were underestimated. There are substantial underestimations of base and peak flows that change in behaviour throughout the year, dependent upon season, specifically in season transitions. When conferring these results with the unpublished LDF, this perturbation is thought to be due to a change in the annual cycle of flow half way through the time period; where the wet season has a greater duration. As such, the calibration can only be described as satisfactory in its estimation of recorded flow. The simulated FDC fits well to the recorded FDC. Due to the unusual distribution of data at the 75th percentile, the fit of the simulated FDC required compromise. Therefore percentiles in excess of 96 were not well estimated in favour of more accurately fitting the bulk of the flow. It is assumed that the untraditional distribution of the recorded FDC is a factor of the change in the yearly flow cycle as previously discussed. The estimated Q95 differs on the scale of a cubic millimetre, a level of precision that is indicative of an excellent calibration. A percentage difference of 7.23% between estimated and recorded flow illustrates the efforts in calibrating for the representation of the bulk flow. MF differs on the scale of a
49

cubic centimetre, as with the Braevallich comparison this indicates the high level of precision for the calibrated model. The percentage difference for MF is 5.08%, which proves to be a poorer calibration than Braevallich; however these statistics indicate a good quality calibration nonetheless. Expressing the percentage differences of Q95 and MF as Q95(%MF), which is the chosen measure of accuracy for simulations in this study, a difference of 12.42% is calculated. From the analysis of the flow estimations provided by the calibration simulations, it is determined that the Durkadale catchment calibration produces a satisfactory HDF estimation and an excellent FDC, Q95 and MF estimation. This calibration is therefore considered a good calibration; a good degree of confidence can be had in the derivation of parameters from its constituent optimal parameter-set.

Q95(%MF) 9-11% group, Luss calibration


Simulated, average, daily flow values presented on the HDF have an excellent closeness to the recorded values. There some sense of lag in the oscillation of the simulated in relation to the recorded values, which could indicate that the rooting depth parameter is inadequate. Using similar parameters around this optimal value did not change the proposed lag seen; the lag is therefore identified as modelling inaccuracy that is inherent for Hysim. Minor oscillations are not always represented and are sometimes estimated opposing the nature of recorded flow. It is typical for models to be unable to represent natural small scale fluctuations; however, producing inverted results is not indicative of a high quality simulation. The underestimation of base and peak flow does not appear to be restricted to summer or winter periods thus confirming the correct choice of the proportion of soil storage in the upper horizon, which would require lowering if summer storms were underestimated. Whilst there is some underestimation of base and peak flows, it is not significant enough to detract from the overall excellent simulation of daily flows that indicates an excellent calibration. The simulated FDC fits very well the recorded FDC. Recorded flow is represented well around the 1st percentile, whilst being unable to quite as effectively estimate those around the 99th percentile. This is typical of model simulations, often being unable to represent the short term extremes of flow. The difference between estimated and recorded Q95 and MF is on the scale of a cubic centimetre, a level of precision that is indicative of an excellent calibration. In percentage difference this is expressed as 4.98% and 1.93% respectively, which confirms the high quality of this calibration in estimating Q95 and MF, which reflect two important characteristics of flow. Expressing the percentage differences of Q95 and MF as Q95(%MF), which is the chosen measure of accuracy for simulations in this study, a difference of 3.10% is calculated. From the analysis of the flow estimations provided by the calibration simulations, it is determined that the Luss catchment calibration produces an excellent HDF estimation, a good FDC, and a good estimation of Q95 and MF. This calibration is therefore considered a good to excellent quality calibration; a very high level of confidence can be had in the derivation of parameters from its constituent optimal parameter-set.

Q95(%MF) 11-13% group, Lathro calibration


Simulated, average, daily flow values presented on the HDF have satisfactory fit to the recorded values. There some sense of lag in the oscillation of the simulated in relation to the recorded values, which could indicate that the rooting depth parameter is inadequate; although, using similar parameters around this optimal value did not change the presence of this lag. It is suggested that this lag is a product of the poor estimation of the base flow. Minor oscillations are not always represented and are sometimes estimated opposing the nature of recorded flow. It is typical for models to be unable to represent natural small scale fluctuations; however, producing inverted results is not
50

indicative of a high quality simulation. Peak flows are estimated relatively well when compared to the estimation of base flow. Base flows are overestimated in the wet season and underestimated in the dry season, which questions the parameters pertaining to the proportion of soil storage in the upper horizon. The value of this parameter should be lowered when summer storms are underestimated during calibration in order to find an optimal parameter set. However, in order to maintain a good performance elsewhere and reduce the subjectivity in calibration this was not pursued beyond the present parameter values. The significant underestimation of base flows in particular causes the quality of these estimations to be deemed as satisfactory. The simulated FDC fits very well to the recorded FDC. Recorded flow is represented well around the 1st percentile, whilst effectively estimating those around the 99th percentile; this is exceptional of model simulations, which are often unable to represent the short term extremes of flow. The accuracy of estimations at these flow extremes does come at the cost of incremental inaccuracy between 50th and 1st and 50th and 99th percentiles. The difference between estimated and recorded Q95 and MF is on the scale of a cubic millimetre, a level of precision that indicates an excellent calibration. In percentage difference this is expressed as 1.43% and 0.15% respectively, which confirms the quality of this calibration in estimating Q95 and MF, two important characteristics of flow. Expressing the percentage differences of Q95 and MF as Q95(%MF), which is the chosen measure of accuracy for simulations in this study, a difference of 1.28% is calculated. From the analysis of the flow estimations provided by the calibration simulations, it is determined that the Lathro catchment calibration produces a satisfactory HDF estimation, an excellent FDC, and an excellent estimation of Q95 and MF. This calibration is therefore considered a good to excellent quality calibration; a very high level of confidence can be had in the derivation of parameters from its constituent optimal parameter-set.

Uncertainty identified with transposition method chosen


Using a measure of accuracy in order to quantify the uncertainty associated with the methods of transposition is a key aim of this study. In order to understand influences upon different characteristics of flow, the measure of accuracy was achieved through calculating percentage differences between estimated and recorded flow for Q95 and MF flow descriptors. The full transposition method uses the entire optimal parameter-set from the calibrated donor catchment with the exception of the catchment area parameter. It was presumed that the FTM would be relatively inaccurate at estimating the flow in target catchments due to the full transposition parameter-set being tailored specifically for only the donor catchment. However, observing the performance of the FTM for the estimation of Q95 and Q95(%MF) it is noted that the FTM performs better than the PTM in the majority of catchments. The performance of the FTM appears to be related to the quality of the donor catchments calibration, from which the parameters are mostly derived for the FTM. Donor catchments with the highest quality calibration, such as Braevallich which has a 1% difference for Q95 and MF, produce the greatest accuracy their target catchment application. As such, the 5-7% Q95(%MF) group has the greatest accuracy across target catchments, producing percentage differences below 12%. The poorest quality calibration, the Durkadale calibration has a Q95 and MF, which have percentage difference of 8% and 5% respectively. In the derivation of these parameters for the FTM, the poorest estimations of flows are within that 7-9% Q95(%MF) group. The PTM does not appear to be influenced by the quality of the donor catchment calibration to the same extent as the FTM and often provides exceptionally low accuracy, such as the Candermill and Creed Bridge catchments or relatively high accuracy, as with Barsolus and Skeabost. An adequate
51

explanation for this is that fewer parameters from the donor catchment calibration are used in the PTM and so fails to have the inherent, stable accuracy that the FTM holds. When comparing the FTM estimations within groups, the percentage difference values for the Q95 statistic fall within a less than 20% range, which the PTM only achieves in the 5-7 and 11-13% groups; therefore, there must be some factor in parameterisation that causes sensitivity in the accuracy of the PTM estimations. Because the PTM contains the interception storage, time to peak, rooting depth and catchment area parameters of the target catchment, it is suggested that these parameters are the cause of this occasional success and failure of the PTM producing accurate results. Interception storage, time to peak and rooting depth parameters were estimated through physical catchment properties observed on an OS map and were therefore subjective. It could be the case, that in some circumstances, the parameters chosen were inadequate, which resulted in the dramatic inaccuracies of Candermill and Creed Bridge. However, it would then have to follow that the exceptional estimations of parameters resulted in the relatively high accuracy for the Barsolus and Skeabost catchments. Further, undocumented, experimentation took place in order to investigate the effect of different parameter value choices at different levels of subjectivity. It was found that alternative parameter calculations for interception storage, time to peak and rooting depth did not have a significant effect upon the accuracy of the PTM. A more empirically inferred accountability for this sensitivity is proposed, following the observation that when the calibration quality of the donor catchment became poorer the PTM illustrated a greater accuracy than FTM. It is suggested that a donor parameter-set becomes less suited to the hydrological similarity of the target catchments for a group with a lower quality calibration; therefore, the use of the target specific -interception storage, time to peak, rooting depth- parameters in the PTM makes this a more accurate model than the FTM for groups where there is a less than excellent calibration of the donor catchment. The accuracy in estimating MF was almost identical within 1%- for both the FTM and the PTM for target catchments; the exception being a relatively inaccurate estimation of MF for Dargall Lane using the PTM. From this it is possible to see the robustness of the estimation of MF by Hysim; that even given two methods of parameter selection extremely close estimations of MF can exist. Also illustrating this robustness is the lack of a relationship between accuracy and donor catchment calibration quality, which was so significant in the relative success of the FTM in low flows estimation. Neither the FTM nor the PTM have any significant response to the calibration quality of the donor catchment. The sensitivity of MF estimation accuracy is not as large as that of Q95. All percentage differences are below 18%; however, there are some fluctuations that appear dependent upon the choice of target catchment. It is possible to make this assumption as both FTM and PTM give very similar percentage differences, which is a factor not available in the analysis of Q95. Looking at some of the least accurate estimations: Candermill, Brockhoperig , Whitburn and Kinross have large Q95(%MF) values. Glen Strae, Deep Hope and Polloch also have low Q95(%MF) and are some of the most accurate estimations. There appears to be a definite correlation between an increase in estimation accuracy and increase in Q95(%MF). No other association could be made with any confidence as to the remaining, less accurate catchments, and the hydrological statistics of those catchments. In order to interpret any change in accuracy due to catchment selection, the hydrological statistics of the target catchments were plotted against the measure of accuracy. The excellent accuracy in estimations of MF by the FTM and PTM result in the Q95(%MF) descriptor bearing equivalence to the Q95 flow descriptor previously discussed. A comparison of the recorded
52

Q95(%MF) to the FTM and PTM estimations for each target catchment confirms the same trends identified for the Q95 statistics as one would expect. A factor of poorer quality estimation with increasing Q95(%MF) does appear to have a significant role to play. An observation can be made as to the degree of overestimation and understimation related to the inaccuracy of the estimation. Uniformly amongst the target catchments, an increasing underestimation is associated with increase in recorded Q95(%MF). This is demonstrated best in the 9-11% Q95(%MF) group, where a decrease in recorded Q95(%MF) across the group is matched by increasing undersestimation with both the FTM and PTM. A target catchment with a recorded Q95(%MF) of approximately 10% appears to be the optmimum value for estimations to remain within a reasonable range of estimation for both the FTM and PTM.

4.3 Uncertainty identified with selection of target catchment


Hydrological statistics of each target catchment were compared to the accuracy of Q95(%MF) estimations in order to illustrate the level of uncertainty associated with the choice of target catchment for the estimation of flow. A very strong relationship is thought to exist between recorded Q95(%MF) for the time period and the accuracy of flow estimations; accuracy was seen to decrease significantly with an increase in recorded Q95(%MF). This is a relationship identified in the estimations of both FTM and PTM, but more present in the estimations of FTM due to the relatively low sensitivity of this methods estimation accuracy. Identifying this relationship is important as it illustrates that target catchments with a high Q95(%MF) above 10%- will always have considerable inaccuracy in the estimation of low flows using the methodology outlined in this study. It could be the case that the choice of donor catchment is inadequate for the target analogues, either in hydrological similarity or in scale of Q95(%MF). A target catchment with a higher Q95(%MF) may require a greater similarity to the donor catchment. In order to investigate this, a similar approach for quantifying uncertainty could be approached but with a focus on testing a variety of donor catchments with different recorded Q95(%MF) values and maintaining the same target catchment. It may be that the range of Q95(%MF) in each group in this study is too large for catchments with a larger recorded Q95(%MF). In observing this relationship the need for ROI grouping within Scottish Water is confirmed and proves that a good methodology exists for this study. Had ROI grouping by Q95(%MF) not been followed the evaluation of uncertainty of the parameter transposition methods would have been less clear. A relationship between BFI and the accuracy of estimations was found to exist. A general trend exists, whereby target catchments with a large BFI is have a greater inaccuracy. As BFI is discussed as being inversely proportional to a measure of flashiness, it follows that a target catchment with a high flashiness would be associated with greater accuracy in flow estimations. This corroborates with a conclusion from the Jacobs (2010) report, which found a greater accuracy, ergo a lower uncertainty of yield in relation to the high flashiness of catchments. There is no relationship that can be identified between area and the accuracy of estimations as various levels of uncertainty exist within the same catchment area. It was observed through experimentation that using records with different time periods to that of the donor catchment does not seem to affect uncertainty, Glen Strae uses different and has percentage difference of 0.75% in the estimation of Q95(%MF), which is the lowest. Creed Bridge also uses different and has a difference of 2.13%. This confirms that the procedure of selection of time periods and rendering data outlined by this study is very good. In addition, this approves the suitability of newly recorded data for the Glen Strae catchment, which utilises records from 2010. The target catchment Durkadale has a data record period that was selected according to quality,
53

representativeness and continuity of data across the catchments in the entire group. For some of the catchments in the 7-9% Q95(%MF) data quality and continuity restricted the selection of the time period to 2000-2009. This was not an ideal selection but was thought to be adequate for the sake of maintaining the same time period across the donor and target catchments for the group. This is a factor that may have caused some undue inaccuracy within this group; the availability of a higher quality, representative data time period may have improved the results of this group.

4.4 A more pragmatic methodology for estimations of flow in ungauged catchments


In grouping catchments into four groups according to Q95(%MF) and using a single parameter-set from the calibration of a donor catchment to simulate flows for the remaining catchments in that group, this study has created a library of four Hysim project files that can be used to estimate flows in hydrologically analogous ungauged catchments. This study has evaluated the use of this procedure to estimate flows in ungauged catchments and illustrated the success of its application using various methods of parameterisation. This study has also quantified uncertainties associated with the methods of parameter selection as well as those associated with the selection of target catchments with relation to their hydrological character. Quantifying this uncertainty has allowed this study to suggest means of avoiding such uncertainties; recommendations can therefore be made to improve the efficiency of Hysim application within Scottish Water and suggestions can be made to improve the procedure through further investigation into the uncertainties highlighted by this study. This study recommends the use of a series of Hysim project files in order to estimate the flow of ungauged catchments. Four such files have been developed in the experimentation phase of this study, which can be allocated a parameter-set derived from the desired method chosen with reference to the performances evaluated in this study. It is suggested that the full transposition method has the least uncertainty associated with parameterisation for donor catchment calibrations that are excellent. This could be extended to calibrations that are good, given that there is more stability in simulations using the FTM than the PTM. It could be recommended that for donor catchments that have a poor calibration that the PTM should be used for parameterisation; however, there is limited evidence for this and should form the basis of further, exclusive experimentation into uncertainty associated with the calibration quality of the donor catchment. The FTM is preferential as it only uses the catchment area parameter from the target catchment, thus requiring only a calibration of a donor catchment as well as a reference to the UKHR in order to achieve a parameter-set that can be used in a specific target catchment. Using the Q95(%MF) ROI categorisation for Scottish Waters source catchments it is possible to allocate a single calibration for a number of catchments in a Q95(%MF) range, ensuring hydrological similarity between donor and target catchments. It was found that for target catchments with a Q95(%MF) above 10% that the inaccuracy of flow estimations significantly increases and suggests the revaluation of ROI grouping. There is no other suggestion from the results of this discussion that could eliminate the uncertainties experienced for target catchments of high Q95(%MF); this should be pursued in subsequent studies. The four Hysim project files that have been developed in this study should be extended to include any further ROI grouping for those catchments with a high Q95(%MF). Average catchment statistics for Scottish Water mega-zones show that high Q95(%MF) is quite typical for Scottish Waters source catchments further studies should aim to quantify uncertainty at Q95(%MF) in excess of 13% and limit the inaccuracy therein. In addition, it would be emeritus to quantify the uncertainty of target catchment selection below 5% in order to verify these findings. A further suggestion for experimentation might be to experiment with increasing the range of ROI grouping for catchments with low Q95(%MF), to see if a single Hysim project file and parameter-set may estimate
54

flows for a much greater range of hydrological similarity. This could be achieved by testing the application of a single project file and parameter-set from one group and evaluating its performance on another group. By making these recommendations, the intention is to reduce the extensive labour involved with the process of data selection, model construction, parameterisation and calibration of a donor catchment of the current Scottish Water Hysim methodology. Being able to make accurate and efficient estimations of flow for ungauged catchments in Scotland will allow Scottish Water to provide higher quality yield estimations.

55

Conclusion

This study has evaluated the performance of two methods of parameter-set construction in estimating flows in ungauged catchments in Scotland. In evaluating the performance, a number of uncertainties have been identified and quantified within the modelling framework chosen. Foremost, there was quantification of uncertainty associated with the choice of parameters and the calibration quality of the donor catchment therein. Ultimately, there was quantification of the uncertainty associated with the choice of target catchment, which reflected the hydrological character of the target catchment. Identifying and quantifying these uncertainties allowed recommendations to be made as to the mitigation of inaccuracies within the Hysim modelling scheme. This was achieved by proposing a pragmatic scheme for which quantified uncertainty exists and suggesting further investigations that would expand the application of such a scheme. It has been revealed that it is possible to use alternative methods of parameter transposition to increase the accuracy of flow estimation in ungauged catchments. Uncertainties associated with the use of two different parameter-sets were evaluated through a measure of accuracy in their ability to estimate flows close to the recorded values of a catchment. It was found that a full transposition model that is derived from the excellent calibration of a donor catchment estimates flows with more accuracy than a partial transposition model. The quality of calibration of the donor catchment was suggested to be a key factor in predetermining the overall accuracy of the two parameter transposition methods. If the donor catchment calibration was poor it was suggested that the partial parameter transposition method, with more target catchment specific parameters, would produce more accurate estimations of flow. The uncertainty associated with the choice of a target catchment was determined by observing model performance in relation to the hydrological character of catchments. A key observation was that for catchments with a higher Q95(%MF), flow was estimated with a greater inaccuracy, proving a greater uncertainty when selecting catchments with a larger Q95(%MF). This illustrated the importance of using ROI as a method of regionalisation; grouping eligible donor and target catchments by Q95(%MF) ensured that the evaluation of parameter transposition methods within groups would not be perturbed by this catchment character uncertainty. In order to mitigate the poor performance of modelling results for catchments with high Q95(%MF) it is suggested the range of ROI groups for higher Q95(%MF) catchments should be smaller in order to ensure more similarity between donor and target catchments. In addition, uncertainty was quantified for catchments with a higher BFI (low flashiness), for which there was a low accuracy for catchments with a higher BFI; thus confirming relationships discussed by previous studies by Jacobs (2010). This study also reported that there was a limited relationship found between target catchment area and uncertainty in flow estimations. These relationships do not have direct bearing upon the evaluation of methods of parameter transposition although catchments bearing extremes of these properties should be avoided in the selection of a donor catchment for calibration in future studies. Developing an excellent donor catchment calibration for each Q95(%MF) range and using the full transposition method the transference of catchment size only- reduces the time and effort in producing less uncertainty for estimates of flow in ungauged catchments. This study highly
56

recommends extension of the four donor parameter-sets of this study to create a library of donor parameter-sets that cover the Q95(%MF) groups for all typical Scottish Water catchments. These library models could be called upon to estimate flows with a quantified uncertainty in ungauged catchments with a specific Q95(%MF). As such, this study makes the recommendation of quantifying uncertainties of Q95(%MF) ranges not covered by this study, especially those above 13% in order to further quantify the uncertainty associated with an increase in Q95(%MF).

57

Acknowledgements

This thesis was pursued as partial fulfilment of the requirements of an Msc in Earth Sciences at Uppsala University. After speculatively enquiring about thesis topic opportunities at Scottish Water I was contacted by Paul Rodgers from their Water Resource Planning team. During subsequent meetings at the Edinburgh office Paul and I discussed a brief outline of the numerous objectives Scottish Water had from the investigation of a flow estimation methodology. The professional and courteous manner in which Paul delivered software and data from various sources such as SEPA was essential to the momentum of this project. Frequent discussion with Paul also ensured that I had support directly with this supportive material, which helped the manipulation of data and the very cumbersome Hysim software. Many thanks to the staff at Scottish Water, who trusted this project to me and were able to dedicate their resources to me. I appreciate the cooperation of staff at the Scottish Environmental Protection Agency in the prompt provision of up-to-date data and expertise in catchment selection. I am much obliged to the patience of staff at Uppsala University. Their support of my studentship and thesis whilst living in the United Kingdom was exceptional. Specifically, I would like to thank Roger Herbert, Jan Seibert and Sven Halldin for their consideration, especially at the time of my thesis presentation on my return to Sweden. I would like to extend further gratitude to Jan for his suggestions regarding the written report and his constructive criticism throughout. Special thanks to my brother Tim for his wonderful accommodation and tolerance during the early phase of the project. And thanks to my loving companion Sarah, whose optimism and belief in me gave the strength to finish it.

58

References
Askew, A. J., Greco, F. et al., 1981. Logistics and benefits of using mathematical models of hydrologic and water resource systems: selected papers with summary of discussions from The International Symposium on Logistics and Benefits of Using Mathematical Models of Hydrologic and Water Resource Systems, Pisa, Italy, 24-26 October 1978. Oxford: Pergamon. Beck, M. B., 1987. Water quality modelling: A review of the analysis of uncertainty. Water Resources Research, 23(8), pp.1393-1442. Beran, M. A., International Commission on Surface Water and Yugoslav Association for Hydrology, 1990. Regionalization in hydrology. International Association of Hydrological Sciences, Wallingford, Oxfordshire. Beven, K. J. and Binley, A. M, 1993. The future of distributed models: Model calibration and uncertainty prediction. Hydrological Processes, 6, pp.279298. Blschl, G. and Sivapalan, M., 1995. Scale issues in hydrological modelling: A review. Hydrological Processes, 9, pp.251290. Boak, R., Yleh, C. and Isa, I., 2004. Development of a high-level technical process and detailed participation strategy to support the implementation of water resources management strategies in Scotland and Northern Ireland. Proceedings of the International Commission on Irrigation and Drainage Seminar on Tools for Public Participation, Con ict Resolution and Decision-Making in Water Resources Management, London, pp.2230. Boorman, D. B., 1985. A review of the Flood Studies Report rainfall-runoff model parameter estimation equations. Wallingford: Institute of Hydrology. Booth, P. D. et al., 2008. WFD and catchment restoration, Scoping Project WFD94. Edinburgh: SNIFFER. Boyle, D. P., Gupta, H.V. and Sorooshian, S., 2000. Toward improved calibration of hydrological models: Combining the strengths of manual and automatic methods. Water Resources Research, 36, pp.36633674. British Standards Institution, 1965. BS-3680: Methods of Measurement of Liquid Flow in Open Channels. London: BSI. Brock, 2009. Gavin Brock's Ordnance Survey Map KML UK overlays for Google Earth. [online] Available at <http://www.brock-family.org/gavin/google-earth/osmaps.html> [Accessed on the 21st of February 2011]. Burn, D., 1990. An appraisal of the 'region of influence' approach to flood frequency analysis. Hydrological Science Journal, 35, pp.149165.
59

Centre for Ecology and Hydrology, 2008. The UK Hydrometric Register. Centre for Ecology and Hydrology, Wallingford, Oxon OX10 8BB, UK Copestake, P. and Young, A., 2008. How much water can a river give? Uncertainty and the flow duration curve. In: Sustainable Hydrology for the 21st Century, 10th BHS National Hydrology Symposium. England 2008. Copestake, P.G. et al, 2006. The Water Framework Directive: a monitoring strategy for determining the quantity and dynamics of flow in Scotland. 9th BHS National Hydrology Symposium. England 2006. Croke, B. and Norton, J., 2004. Regionalisation of rainfall-runoff models. 2nd Biennial Meeting of the International Environmental Modelling and Software Society. Germany 2004. Deelstra, J. and lital, A., 2008. The use of the flashiness index as a possible indicator for nutrient loss prediction in agricultural catchments. Boreal Environmental Research, 13, pp.209-221. Eng, K., Tasker, G. and Milly, P., 2005. An Analysis of Region-of-Influence Methods for Flood Regionalization in the Gulf-Atlantic Rolling Plains. Journal of the American Water Resources Association, 41, pp.135143. Entec UK, 2003. SNIFFER Water Framework Directive Project 12: Derivation of a methodology for Groundwater Recharge Assessment in Scotland and Northern Ireland. Penicuik: Enteck UK Ltd. Fennessey, N. and Vogel, R., 1990. Regional Flow-Duration Curves for Ungauged Sites in Massachusetts. Journal of Water Resource Planning and Management, 116(4), pp.530549. Gleick, P.H. ed., 1993. Water in Crisis: A Guide to the World's Fresh Water Resources. New York: Oxford University Press. Gustard, A., Bullock, A. and Dixon, J.M., 1992. Low flow estimation in the United Kingdom. Institute of Hydrology Report No 108. Wallingford: Centre for Ecology and Hydrology. Institute of Hydrology, 1999. The UK Flood Estimation Handbook. Wallingford: Institute of Hydrology. Jacobs, Sutton, R., Price, D. and McGregor, L., 2010. Testing the sensitivity of surface water yields to the derivation of Hysim inflows. Glasgow: Jacobs Engineering U.K. Ltd. Kokkonen, T. S., Jakeman, A. J., Young, P. C. and Koivusalo, H. J., 2003. Predicting daily flows in ungauged catchments: model regionalization from catchment descriptors at the Coweeta Hydrologic Laboratory. Hydrological Processes, 17, pp.22192238. Li, H., Zhang, Y., Chiew, F.H.S. and Xu, S., 2009. Predicting runoff in ungauged catchments by using Xinanjiang model with MODIS leaf area index. Journal of Hydrology, 370, pp.1-4.

60

Littlewood, I.G., 2005. Unit hydrographs and regionalisation of United Kingdom river flows: comments on some estimation uncertainties. Wallingford: Water Resource Associates Ltd. Manley, R.E., 1975. A hydrologic model with physically realistic parameters. UNESCO symposium. Bratislava 1997. Manley, R.E., 1978a. HYSIM - a general purpose hydrologic model. Pisa symposium. Italy 1982. Manley, R.E., 1978b. Simulation of flow in ungauged basins. Hydrological Science Bulletin, 23, pp.13. Manley, R.E., 2006. A Guide to Using Hysim. Wallingford: Water Resource Associates Ltd. Mott Macdonald, 2010. Asset Management Planning Assessment Framework, UK water Industry Research Ltd. [online] Available at <http://www.sepa.org.uk/climate_change/publications.asp> [Accessed on the 21st of February 2011]. Murphy, C., Fealy, R., Charlton, R. and Sweeney, J., 2006. Changing Precipitation Scenarios: Preliminary implications for groundwater flow systems and planning'. 25th Anniversary Conference on Groundwater. Ireland 2006. Neuman, S. P., 2003. S.P.: Maximun likelihood Bayesian averaging of uncertain model predictions. Stochastic Environmental Resources, 17, pp.291-305. Refsgaard, J.C. and Knudsen, J., 1996. Operational validation and intercomparison of different types of hydrologic models. Water Resources Research, 32, pp.21892202. Scottish Water, 2006. SR06 Project (internal acronym meaning: strategic review of charges 20062010). Scottish Water Resource Plan. Dunfermline: Scottish Water. Scottish Water, 2009. Scottish Water Resource Plan. Dunfermline: Scottish Water. Scottish Water, 2011. Scottish Water Hydrology Guidance. Dunfermline: Scottish Water. Seibert, J., 1999. Conceptual runoff models - fiction or representation of reality? Comprehensive Summaries of Uppsala Dissertations from the Faculty of Science and Technology, 436, pp.52. Seibert, J., 1999. Regionalisation of parameters for a conceptual rainfall-runoff model. Agricultural and Forest Meteorology, 98-99, pp.279-293 SEPA, 2009. Change in river flow variability in snow influenced catchments in Scotland. [online] Available at <http://www.sepa.org.uk/climate_change/publications.asp> [Accessed on the 21st of February 2011]. SNIFFER, 2001. SNIFFER Project SR(01)02: A Surface Water Yield Assessment. [online] Available at <http://www.sniffer.org.uk/> [Accessed on the 21st of February 2011].

61

SNIFFER, 2006. SNIFFER Project WFD48: Development of Environmental Standards. [online] Available at <http://www.sniffer.org.uk/> [Accessed on the 21st of February 2011]. Szilagy, J. et al., 1990. Regional estimation of base recharge to ground water using water balance and base-flow index. Ground Water, 41(4), pp.504-513. Uhlenbrook, S., Seibert, J., Leibundgut, Ch. and Rodhe, A., 1999. Prediction uncertainty of conceptual rainfall-runoff models caused by problems to identify model parameters and structure. Journal des Sciences Hydrologiques, 44(5), pp.779-798. US Army Corps of Engineers, 2011. Whats New? Version 3.0 (April 2007) of the Reservoir System Simulation program (HEC-ResSim). [online] Available at <http://www.hec.usace.army.mil/software/hec-ressim/whats_new.html> [Accessed on the 21st of February 2011]. Westerberg, I., 2011. Observational Uncertainties in Water-Resources Modelling in Central America. Ph. D. Uppsala University. Young, A.R., Grew, R., Holmes, M.G.R., 2003. Low flows 2000: A national water resources assessment and decision support tool. Water Science Technology, 48, pp.119-126.

62

8 Appendices
Appendix A: Hysim operational notes
8.1 Calibration: procedural notes
1. Open project file containing donor catchment precipitation, evapotranspiration, recorded flow inputs as well as parameter-set including researched interception storage, time to peak, rooting depth and catchment area parameters. 2. Run single parameter mode optimization, which optimizes the potential evapotranspiration parameter (PET factor). Alternatively precipitation factor is optimized if PET has a negative effect upon estimated flow. 3. Run extremes of error estimate for upper and lower permeability as well as upper and lower interflow parameters. 4. Run model simulation with no optimisation; compare simulated and gauged flow fit. 5. Make changes to parameter values; save parameter changes. 6. Repeat steps 4, 5 and 6 until satisfied with simulated and gauged flow fit. As potential evapotranspiration (PET) is the most uncertain parameter this is used in the single parameter mode run, so as to provide the best estimation for this parameter. In previous calibrations by Scottish Water this parameter choice for single parameter run was rainfall factor; this does not correspond with the guidance provided by the Hysim User Manual (Manley, 2006). If there is good coverage, with a sufficient density and spread -as with improvements made to the quality of Met Office provided 5km2 gridded rainfall- then the use of PET for single parameter optimisation is suitable.

8.2 Full transposition method: procedural notes


Open parameter file for donor catchment donor parameter-set. Change catchment area parameter value to that of catchment analogue. Save parameter file, identifying it as analogue model 1. Open project file for donor catchments donor parameter-set. Open flow file, rainfall file, evapotranspiration file for catchment analogue as well as the previously saved parameter file. 6. Save project file under different project name. 7. Run model with no optimization. 8. Calculate Q95 and MF for recorded and simulated data; calculate percentage difference. 1. 2. 3. 4. 5.

63

8.3 Partial transposition method: procedural notes


1. Open parameter file for donor catchment donor parameter-set. 2. Change interception storage, time to peak, rooting depth and catchment area parameter values to those calculated for the catchment analogue. Set evapotranspiration factor and precipitation factor to 1. 3. Save parameter file, identifying it as analogue model 2. 4. Open project file for donor catchments donor parameter-set. 5. Open flow file, rainfall file, evapotranspiration file for catchment analogue as well as the previously saved parameter file. 6. Save project file under different project name. 7. Run model with no optimization. 8. Calculate Q95 and MF for recorded and simulated data; calculate percentage difference.

Appendix B: Parameter-set references


8.4 Final calibration and partial transposition model parameters
Parameters 10 year period of record Interception storage (mm) Impermeable Proportion Time to peak (hrs) Rooting Depth (mm) Pore size distribution index Permeability (horizon boundary) (mm/hr) Permeability (base lower) (mm/hr) Interflow (upper) (mm/hr) Interflow (lower) (mm/hr) Groundwater recession (per month) Precipitation factor PET factor Catchment Area (sq km) Parameters 10 year period of record Interception storage (mm) Impermeable Proportion Time to peak (hrs) Rooting Depth (mm) Pore size distribution index Permeability (horizon boundary) (mm/hr) Permeability (base lower) (mm/hr) Interflow (upper) (mm/hr) Interflow (lower) (mm/hr) Groundwater recession (per month) Precipitation factor PET factor Catchment Area (sq km)
Default Braevallich Glen Strae Polloch Deep Hope Durkadale Barsolus Inverlochy Skeabost

N/A 2.00 0.02 2.00 1000 0.15 10 10 10 10 0.9 1 1 1


Default

1993-2002 4.00 0.02 3.55 750 0.9 8 3.739 14 31.261 0.9 1 1 23.15
Luss

1993-2002 2.68 0.02 4.06 627.5 0.9 8 3.739 14 31.261 0.9 1 1 36.62
Candermill

1993-2002 3.20 0.02 1.80 1100 0.9 8 3.739 14 31.261 0.9 1 1 8.12

1993-2002 3.35 0.02 4.22 1175 0.9 8 3.739 14 31.261 0.9 1 1 30.99

2000-2009 2.00 0.02 3.62 1250 0.9 0.476 3.753 2.271 2.112 0.9 1.155 1 19.60
Lathro

2000-2009 2.03 0.02 5.10 515 0.9 0.476 3.753 2.271 2.112 0.9 1 1 32.83
Brockhoperig

2000-2009 4.40 0.02 5.09 1700 0.9 0.476 3.753 2.271 2.112 0.9 1 1 47.09
Kinross

2000-2009 2.30 0.02 5.67 650 0.9 0.476 3.753 2.271 2.112 0.9 1 1 80.55
Whitburn

Creed Bridge Dargall Lane

N/A 2 0.02 2.00 1000 0.15 10 10 10 10 0.9 1 1 1

1988-1997 4 0.02 3.00 1000 0.15 0.723 0.995 3.33 1.978 0.9 1.144 0.88 34.81

1988-1997 2.06 0.02 4.14 530 0.15 0.723 0.995 3.33 1.978 0.9 1 1 25.3

1988-1997 2.03 0.02 3.00 515 0.15 0.723 0.995 3.33 1.978 0.9 1 1 44.83

1988-1997 2 0.02 1.45 500 0.15 0.723 0.995 3.33 1.978 0.9 1 1 2.07

1992-2001 2.6 0.02 3.80 800 0.15 8.016 47.647 28.575 45.289 0.9 1 0.8097998 23.60

1992-2001 3.35 0.02 4.95 1175 0.15 8.016 47.647 28.575 45.289 0.9 1 1 38.59

1992-2001 2.42 0.02 4.80 710 0.15 8.016 47.647 28.575 45.289 0.9 1 1 33.75

1992-2001 2.54 0.02 4.90 770 0.15 8.016 47.647 28.575 45.289 0.9 1 1 31.80

8.5 Durkadale parameter-set notes


The calibration of the Durkadale model did not follow standard calibration procedure; the standard single parameter optimisation method outlined by Maney, 2008, was inadequate in representing the FDC of the recorded flow. When estimating the PET factor using single parameter optimisation PET was lowered to 0.51, which is an unsatisfactory parameter value for PET bias. This would indicate some significant inconsistency in the PET data record. The data was checked and confirmed to be acceptable. When studying the simulated FDC against the recorded FDC it was seen that the recorded FDC had a high range of 50th to 70th percentile flow values. Due to the pursuit of Hysim to achieve a high correlation of MF between the simulated and observed flows it was necessary for the single parameter optimisation process to force the PET to an unrepresentatively low value. The resulting
64

standard calibration from this low PET value had a MF percentage difference of 0.002%, proving that for MF simulation this was an important step. However, the shape of the FDC from this calibration was misshapen and did not fit the recorded FDC. This was best proven by the use of the resulting Q95 percentage difference, which was 51.10%. As a consequence of these findings it was decided to use single parameter optimisation of the precipitation factor. A comparison of flow statistics of standard calibration with and without precipitation factor optimised led to the identification of a universal rise in the simulated FDC with an increase in precipitation factor. After considerable unsuccessful attempts to fit the shape of the simulated FDC to shape of the observed FDC without a single parameter optimisation it became apparent that it was only possible to match the shape by having the simulated FDC lie uniformly below the recorded FDC. It was at this point that single parameter optimisation was used on the precipitation factor. The precipitation factor was therefore raised to 1.155. This raised the FDC uniformly and created a close match of simulated catchment statistics to that of recorded catchment statistics. The resulting numerical difference of Q95 was 0.004m3s-1 and MF was 0.026m3s-1. This illustrates the value of using single parameter optimisation to judge precipitation factor and could suggest why the alternative procedure for standard calibration adopted by Scottish Water i.e. the preferential single parameter optimisation of precipitation factor, has been successful in the past. It was therefore recommended when optimising consequent parameter-sets for donor parameter-sets that the effect of changing PET through single parameter optimisation should be studied using FDC and flow statistics before continuing with standard calibration procedure as outlined in the method and Hysim Manual (Manley, 2006).

Appendix C: Results of validation


8.6 Target catchment FDCs from validation of the transposition methods

65

66

67

68

69

70

71

Tidigare utgivna publikationer i serien ISSN 1650-6553 Nr 1 Geomorphological mapping and hazard assessment of alpine areas in Vorarlberg, Austria, Marcus Gustavsson Nr 2 Verification of the Turbulence Index used at SMHI, Stefan Bergman Nr 3 Forecasting the next days maximum and minimum temperature in Vancouver, Canada by using artificial neural network models, Magnus Nilsson Nr 4 The tectonic history of the Skyttorp-Vattholma fault zone, south-central Sweden, Anna Victoria Engstrm Nr 5 Investigation on Surface energy fluxes and their relationship to synoptic weather patterns on Storglaciren, northern Sweden, Yvonne Kramer Nr 217 Mineralogy and Lithogeochemical Signature of a Stratigraphic Profile through the Grnsgruvan Zn-mineralization, Bergslagen, Sweden,ErikBjrklund,April 2011 Nr 218 Belizean Barrier Reef showing signs of recovery twelve years after the 1998 disturbances,Joaquin David Magaa,April 2011 Nr 219 Turbulence Intensity in Complex Environments and its Influence on Small Wind Turbines,Nicole Carpman,May 2011 Nr 220 Observed Ice Supersaturated Layers over Sweden and Implications for Aviation Induced Contrails over the Baltic Sea,Elin Bjrklund,May 2011 Nr 221 Linking forest cover with runoff and sediment yield in the Cir Grande and Trinidad River basins in the Panam Canal Watershed ,Abha Parajulee,May 2011 Nr 222 Re-Processing of Reflection Seismic Data from Line V2 of the HIRE Seismic Reflection Survey in the Suurikuusikko Mining and Exploration Area, Northern Finland,Amir Abdi, August 2011 Nr 223 Re-processing and interpretation of 2D seismic line in the Kristineberg mining area, northern Sweden, Siddique Akhtar Ehsan, August 2011 Nr 224 Verifiering av WRF-modellen ver Svalbard, Anna Waxegrd, September 2011 Nr 225 Constraining Crustal Volatile Release in Magmatic Conduits by Synchrotron X-ray -CT, Sylvia Berg, October 2011 Nr 226 Probing the Toba Super-Eruption: Insights from Oxygen Isotope Geochemistry and Geobarometry, David Budd, November 2011 Nr 227 Uppfljning av projekt Haparandabanans vattenanknutna taganden, naturmiljeffekter och miljml, Hanna strn, November 2011

Vous aimerez peut-être aussi