Vous êtes sur la page 1sur 46

TIME SERIES ANALYSIS USING FUZZY LOGIC

A DISSERTATION

SUBMITTED IN PARTIAL FULFILMENT OF THE REQUIREMENTS FOR THE


AWARD OF THE
DEGREE OF
MASTER OF ENGINEERING
IN
CIVIL ENGINEERING
(WATER RESOURCES ENGINEERING)

PEC UNIVERSITY OF TECHNOLOGY,

CHANDIGARH

SUBMITTED BY
AKSHAY KUMAR
ME-15203015
UNDER THE GUIDANCE & SUPERVISION OF
Dr. KAMAL KUMAR
ASSISTANT PROFESSOR
DEPARTMENT OF CIVIL ENGINEERING

PEC, UNIVERSITY OF TECHNOLOGY


CHANDIGARH
JUNE 2017
CANDIDATE’S DECLARATION
I hereby certify that the work which is being presented in the thesis entitled “TIME SERIES
MODELING USING FUZZY LOGIC” in the partial fulfilment of the requirements for
the award of the Masters of Engineering in Civil Engineering and submitted in the
Department of Civil Engineering of the PEC University of Technology, Chandigarh, is an
authentic record of my own work carried out during the period from June 2016 to June 2017
under the supervision of Dr. Kamal Kumar, Assistant Professor, Department of Civil
Engineering.

I have not submitted the matter presented in this thesis for the award of any other degree of
this or any other University/ Institute.

Date: Akshay Kumar

This is to certify that the above statement made by the candidate is correct to the best of my
knowledge.

Date: Dr. Kamal Kumar


Assistant Professor
Department of Civil Engineering
PEC University of Technology
Chandigarh – 160012, India

ii
ABSTRACT
Runoff estimation from a watershed is of utmost importance for various hydrologic and

hydraulic purposes. Trend analysis of rainfall over a watershed area on various spatial and

time scales, has been a great concern during the past few decades because of global climate

change. Numerous studies have been carried out in modelling the runoff response from a

watershed. However, the modelling may be of distributed /lumped which require different

data parameters /variables. To account fluctuations in rainfall data at time scale requires time

series analysis of data. The aim of the present study is to analyse the temporal and spatial

variation of rainfall using statistical models and fuzzy sets. In this study rainfall estimation

of Gambar sub-watershed (H.P) which is located in Western Himalayas having average

elevation of 1100 m is done using regression models and fuzzy sets. Basin hydrology

features have been extracted using Remote Sensing, Geographical Information System (GIS)

and field observations. Gambar watershed is spread over a total area of approximate

729.51km2. Statistical models like Regression and Analysis of Variance (ANOVA) were

used to study the orographic effect over Gambar watershed. The results shows that there is

no orographic effect in Gambar watershed.

Keywords: Runoff, Time Series, ANOVA, Fuzzy

iii
ACKNOWLEDGEMENTS
It is my great pleasure to express my sincere thanks to all the magnanimous persons who
rendered their full support to my work directly or indirectly.

First I would like to extend my sincerest thanks to my guide Dr. Kamal Kumar,Assistant
Professor, Civil Engineering Department, PEC University of Technology, Chandigarh for
his support throughout my work, perception and buoyant nature that has made working in
PEC truly a pleasure. Without his inspiration and generous guidance, the work would not
have been successful. I have learnt many things from his pertaining to dissertation and I am
sure it would lead to success in my future career.

I express my special gratitude to Dr. Siby John H.O.D., Civil Engineering Department, and
PEC University of Technology for all his possible support in using various facilities of the
department for this work. I also thank the faculty members of the Water Resources
Engineering Department for their valuable support.

I would also like to thank my father for his immense support and love throughout my studies.
The encouragement by the family convinced me that a master’s degree is the best direction
for me to take. Finally, I must thank to all members of PEC, so many of them have provided
me with ideas for successfully completing this research effort.

Akshay Kumar

iv
TABLE OF CONTENTS
CANDIDATE’S DECLARATION…………………………………………………….....ii
ABSTRACT………………………………………………………………………………iii
ACKNOWLEDGEMENTS……………………………………………………………….iv
TABLE OF CONTENTS………………………………………………………………v-vi
LIST OF TABLES & FIGURES…………………………………………………….vii-viii

CHAPTER 1 INTRODUCTION .................................................... 1


1.1 GENERAL ........................................................................................................................... 1
1.2 TIME SERIES IN HYDROLOGY .......................................................................................... 1
1.2.1 Components of Hydrologic Time Series .................................................................................. 2
1.3 RAINFALL-RUNOFF RELATION ........................................................................................ 3
1.3.1 Watershed .................................................................................................................................. 3
1.3.2 Significance of Modelling In Watershed Hydrology ............................................................... 4
1.3.3 Importance of Data in Hydrological Modelling ...................................................................... 5
1.4 FUZZY LOGIC IN HYDROLOGY......................................................................................... 5
1.4.1 Fuzzy Rainfall – Runoff ............................................................................................................ 6
1.5 ADAPTIVE NETWORK BASED FUZZY INFERENCE SYSTEM ............................................ 7
1.6 ORGANIZATION OF THESIS .............................................................................................. 7

CHAPTER 2 LITERATURE REVIEW........................................ 8


2.1 GENERAL ........................................................................................................................... 8

CHAPTER 3 OBJECTIVES & STUDY AREA ......................... 16


3.1 OBJECTIVES..................................................................................................................... 16
3.2 STUDY AREA .................................................................................................................... 16
3.3 SOFTWARE & TOOLS USED ............................................................................................ 17

CHAPTER 4 METHODOLOGY ................................................. 19


4.1 GENERAL ......................................................................................................................... 19
4.2 DATA COLLECTION......................................................................................................... 19
4.2.1 Field Data Procurement .......................................................................................................... 19
4.2.2 Spatial Data Procurement ....................................................................................................... 19
4.3 OROGRAPHIC EFFECT ANALYSIS .................................................................................. 21
4.4 DATA PROCESSING.......................................................................................................... 21

v
4.4.1 Field Data Processing ............................................................................................................... 21
4.4.2 Digital Elevation Model Processing ........................................................................................ 22
4.4.3 Soil Moisture Map .................................................................................................................... 24
4.5 LANDSAT8 DATA PROCESSING .................................................................................... 24
4.5.1 Normalized Difference Vegetation Index ............................................................................... 25
4.5.2 Land Use Land Cover Map ..................................................................................................... 27
4.6 ADAPTIVE - NEURO FUZZY INFERENCE SYSTEM .......................................................... 28
4.6.1 Adaptive Network .................................................................................................................... 28
4.7 FUZZY INFERENCE SYSTEM ........................................................................................... 30

CHAPTER 5 RESULTS & DISCUSSIONS ............................... 32


5.1 GENERAL ......................................................................................................................... 32
5.2 OROGRAPHIC EFFECT ON GAMBHAR WATERSHED ..................................................... 32
5.3 RAINFALL - RUNOFF REGRESSION................................................................................. 33
5.4 ANFIS MODEL ................................................................................................................ 34
5.4.1 Simulation ................................................................................................................................. 36
5.5 ERROR ANALYSIS ............................................................................................................ 36
5.6 DISCUSSIONS.................................................................................................................... 37

vi
LIST OF TABLES & FIGURES

Table4.1 LANDSAT 8 Band Details ................................................................................ 25

Table 5.1 Rainfall Data for ANOVA................................................................................ 32

Table 5.2 ANOVA Table ................................................................................................... 33

Table 5.3 Relative Error Calculation............................................................................... 37

Figure 1.1 A Typical Watershed (Source: Http://Mwmo.Org/Learn/Stormwater-

101/What-Is-A-Watershed/) ............................................................................................... 4

Figure 1.2 Hypothetical Fuzzy Sub- Groups of RR.......................................................... 6

Figure 1.3 Sugeno Rule Operation ..................................................................................... 7

Figure 3.1GAMBHAR WATERSHED, HIMACHAL PRADESH .............................. 16

Figure 4.1Methodology ..................................................................................................... 20

Figure 4.2 Mosaicked Toposheets .................................................................................... 21

Figure 4.3 SRTM DEM with Drainage............................................................................ 22

Figure 4.4 Stream Order and Basin Boundary ............................................................... 22

Figure 4.5 Contour Map (50m Intervel) .......................................................................... 23

Figure 4.6 Slope Map Gambhar Watershed ................................................................... 23

Figure 4.7 Soil Moisture Map of Gambhar Watershed ................................................. 24

Figure 4.9 NDVI Gambhar Watershed ........................................................................... 27

Figure 4.10 Landuse and cover Map ............................................................................... 28

Figure 4.11Adaptive Network .......................................................................................... 29

Figure 4.12 ANFIS Structure. .......................................................................................... 29

Figure 4.13 Fuzzy Inference System ................................................................................ 30

Figure 4.14 FIS Editor ...................................................................................................... 31

Figure 4.15Membership Function Editor for Input and Output .................................. 31

Figure 5.1 Linear Regression Daily Rainfall-Runoff Data ............................................ 33

vii
Figure 5.2 Training Data Regression............................................................................... 34

Figure 5.3 Training FIS .................................................................................................... 35

Figure 5.4 Testing FIS ....................................................................................................... 35

Figure 5.5 Observed vs. Predicted Runoff ...................................................................... 36

viii
CHAPTER 1
INTRODUCTION

Water is the major requirement for the existence of life and it has been man’s
endeavour in history until present to utilise the available sources of water. The
worldwide activities on water resources development have taken rapid advances in
instrumentation, data acquisition and the computer facilities for data analysis have
contributed towards the rapid growth in hydrology. Hydrologic events in nature are very
complex and arise from various uncertainties in forms of vagueness. Complexity and
ambiguity are related, the closer one looks at a real-world problem, the fuzzier becomes
its solution (L. a Zadeh, 1973). Rainfall, which is a product of a number of complex,
processes that use to vary both in space and in time. The portion of rainfall, which
appears in surface streams of either perennial of intermittent nature, is called runoff.
There is a huge contrast between rainfall and runoff is due to the effect of storage of the
surface layers of the earth. The hydrologic behaviour of the watershed in rainfall-runoff
transformation process is a very complicated concept, which is controlled by a large
number of climatic and geographical factors that vary with both time and space.

Time series is a time oriented sequence of observation on a variable of interest.


Most of the forecasting issues involve the use of time series data. The use of statistical
methods for various hydrological problems in earlier days was very much restricted to
surface water problems, mainly to find hydrologic extremes such as floods and droughts.
In hydrology, time series analysis is used for building mathematical models to generate
synthetic hydrologic records, to forecast hydrologic events, to detect trends and shifts in
hydrologic records, and to fill in missing data and extend records (Salas, 1993).
Hydrologic time series can be classified into two basic groups: (a) single time series at
a specified point (univariate), (b) Multiple series of different kinds at one point
(multivariate). Single and multiple time series are further classified on the basis of time
interval used for the modelling. On the basis of time intervals, the following types of
time series are there:

1
 Continuous Time Series (∆t = 0)
 Series with intervals that are fraction of day (∆t = 1hr, 2hr, 6hr etc.)

 Series with intervals that are fraction of year (∆t can be daily, weekly, monthly,
seasonally etc.)

 Annual time series in which integration over the year does not have any cycles.

1.2.1 Components of Hydrologic Time Series

In time series trend defines the long-term movement of the series without
seasonal and irregular effects and shows the reflections of underlying levels. Major
four components of hydrologic time series are:
 Secular Trend: The secular trend in a time series results from the long-term
effect of social, economic and political factors. This trend may show the growth
or decline in a time series over a long period. This is that type of tendency, which
continues to remain for a very long period.

 Seasonal Trend: These are short-term movements occurring in a data due to


seasonal factors. The short term is generally considered as a period in which
changes occur in a time series with variations in weather or festivities. For
example, it is observed that rainfall in monsoons in quite high in comparison to
other seasons.
 Cyclic Movements: These are long term oscillation occurring in a time series.
These oscillations are mostly observed in economics data and the periods of such
oscillations are generally extended from five to twelve years or more. These
cyclic movements can be studied provided a long series of measurements, which
is free from irregular fluctuations.
 Irregular Fluctuations: These are sudden changes occurring in a time series
which are unlikely to be repeated, it is that component of a time series which
cannot be explained by trend, seasonal or cyclic movements. These variations
are accidental in nature, can cause a continual change in the trend, seasonal and
cyclical oscillations during the forthcoming period. Floods, fires, earthquakes,
epidemics and strikes etc., are the root cause of such irregularities.

2
Rainfall which is an end product of a number of complex processes that use to vary
both in space and time. The portion of rainfall which appears in surface streams of either
perennial of intermittent nature is called runoff. There is a huge contrast between rainfall
and runoff is due to the effect of storage of the surface layers of the earth. The hydrologic
behaviour of the watershed in rainfall-runoff transformation process is a very
complicated concept which is controlled by a large number of climatic and geographical
factors that vary with both time and space. In rainfall-runoff relationships, rainfall is
assumed to be distributed over the drainage area. These assumptions are valid for small
areas but when with an increase in the area the validity of this approach can be
questioned (Şen & Wagdani, 2008). Consequently, more uncertainties are included in
the overall rainfall – runoff process. Depending on the antecedent soil and surface
conditions of the drainage area, the portion of the rainfall that appears as direct runoff
will be different even when the peak rainfall amounts are the same. This indicates that
the transformation to runoff is not static, but rather a dynamic process according to the
environmental conditions. For instance, during wet periods, the conditions are different
than during dry spells. It should be noted here that the words wet and dry are
linguistically fuzzy in contents. It is well known that the rainfall–runoff process is
dynamic and non-linear in nature, where proportionality and superposition principles do
not apply (Kundzewicz & Napiórkowski, 1986). The connection of the rainfall–runoff
points in the logical monthly sequence leads to irregular polygons on the coordinate
systems (Kadiolu, Şen, & Gültekin, 2001). Conventionally rainfall-runoff models are of
three categories: deterministic (physical), conceptual (analytic) and parametric
(empirical). These conventional methods/models used to find rainfall-runoff relations
but to describe exact relation the methods become complex.

1.3.1 Watershed

For time series hydrological modelling, it is suitable to have finite surface/ boundary
within which the inherent properties and characteristics of the region is defined.
Subsequently, the local interactions within this area and the external influences are made
and the consequent outcomes are determined and aggregated over the region to arrive at
an output. In hydrology, this region is called a watershed, drainage basin, river basin or

3
catchment and basically defined by the nature of earth’s topography (Figure. 1). A
watershed is the area of land draining into a stream at a given location.

With reference to the hydrologic cycle, the major input to this system is spatially varied
rainfall, while streamflow is the major output concentrated at the watershed outlet.
Inherent characteristics of the watershed include surface area, slope, mainstream length,
shape, orientation, land use and soil types. Watershed is generally regarded as the most
appropriate unit spatial unit for land management.

Figure1.1 A Typical Watershed (Source: Http://Mwmo.Org/Learn/Stormwater-101/What-Is-A-


Watershed/)

1.3.2 Significance of Modelling In Watershed Hydrology

The hydrologic system is a complex phenomenon to understand in detail.


Therefore, a generalisation of a concept is necessary to understand their behaviour.
Indeed, through experience, it is found that abstraction is required for predicting and
understanding the behaviour of any significant part of the environment.

INPUT WATERSHED OUTPUT


(Rainfall, Snow) (Discharge)

4
The watershed hydrologic models have different forms and their development
varies for various reasons. However, these watershed models have been designed for
two primary objectives. The first aim of watershed modelling is to attain a better
knowledge of the hydrologic phenomenon, which operates in a catchment and how this
phenomenon is affected by changes in the catchment. The other purpose of watershed
modelling is to create the artificial sequence of hydrologic data for designing purposes
or forecasting use. In the present study, the fuzzy logic based approach is used for
watershed modelling using a sequence of past events.

1.3.3 Importance of Data in Hydrological Modelling

With the help of improved means to monitor hydrologic data, remote sensing
and GIS techniques are being highly integrated with hydrological models for providing
real-time weather forecasting, flood forecasting, seasonal snowmelt runoff forecasting.
These current techniques are highly useful in ground water potential mapping to support
the consolidated usage of surface water and ground water. It also helps in inventorying
of coastal and marine processes and assessment and destruction caused by floods.

Fuzzy set theory is a soft computing tool like Artificial Neural Networks, Genetic
Algorithm in which a model is trained to have results. In mathematics, Fuzzy sets are
sets whose elements have degrees of membership. Lotfi A. Zadeh and Dieter Klaua
introduced fuzzy sets in 1965 as an extension of the classical notion of set. A fuzzy rule
is a highly sophisticated physical and mathematical approach which requires extreme
effort in data input and handling (L. A. Zadeh, 1965). It works on fuzzy based routines
to simulate the various processes involved in generating runoff from rainfall. Models
based on physical processes generally needs mathematical equations to solve the
problem that requires a high demand of data requirement and sometimes it is necessary
to estimate the input parameters, specifically related to the area being modelled.
Therefore, these parameters are determined subjectively based on the modeller’s
judgement and the effect normally appears in the output of the model.
Crispness and impreciseness are the major terms to define the fuzziness in the problem.
When there is exact value in the output then it is the crisp one but in case there are some
errors in measurements then it’s called vague. In the case of fuzzy sets, the boundary is

5
not crisp but it is vague. The membership function can be described by an arbitrary
curve suitable from the point of view of simplicity, convenience, speed, and efficiency.
A sharp set is a subset of a fuzzy set where the membership function can take only the
values 0 and 1(Lohani, Goel, & Bhatia, 2006). Hydrologists should use objective
information (equations, algorithms and formulation) and subjective knowledge
(linguistic information) for arriving at an optimum solution for solving real life
hydrology problems. Fuzzy logic principles are much suitable for combining objective
information with subjective knowledge. Its principles provide a simple way to draw
definite conclusions from vague, ambiguous, or imprecise information. The fuzzy logic
approach can provide the structure and solution procedure prior to any deterministic
method like mathematics, statistics, or stochastic processes.
 Linguistic Variable: The variables, which can be defined for various
membership functions, are called linguistic variables. For e.g. a set for
temperature, values can take different linguistic variables for its representation
as “VL – Very Low, L – Low, M – Medium, H – High, VH – Very High”.
 Membership Function: A membership function (MF) is a curve that defines
how each point in the input space is mapped to a membership value (or degree
of membership) between 0 and 1.

1.4.1 Fuzzy Rainfall – Runoff

Rainfall and runoff variables are considered in five partial subgroups: “low”
(“L”), “medium low” (“ML”), “medium” (“M”), “medium high” (“MH”), and “high”
(“H”). A small number of fuzzy sub-group selection leads to unrepresentative
predictions whereas large numbers imply unnecessary calculations. Five sub-groups in
each variable imply that there are 5 × 5 = 25 different partial relationship pairs that may
be considered between the rainfall and runoff variables. Because the rainfall–runoff
relationship, in general, has a direct proportionality feature, it is possible to write the
following five rules for the description of fuzzy rainfall–runoff modelling:

R1: “IF rainfall is ‘L’ THEN runoff is ‘L.’”

R2: “IF rainfall is ‘ML’ THEN runoff is ‘ML.’”

R3: “IF rainfall is ‘M’ THEN runoff is ‘M.’”

Figure 1.2 Hypothetical Fuzzy Sub-


6 Groups of RR
R4: “IF rainfall is ‘MH’ THEN runoff is ‘MH.’”

R5: “IF rainfall is ‘H’ THEN runoff is ‘H.’”

ANFIS applies two techniques in updating parameters. For premise parameters that
define membership functions, ANFIS employs gradient descent to fine-tune them. For
consequent parameters that define the coefficients of each output equations, ANFIS uses
the least-squares method to identify them. This approach is thus called hybrid learning
method since it combines gradient descent and the least-squares method(Jang, 1992).
ANFIS is a graphical representation of Sugeno-type fuzzy systems which are endowed
with neural learning capabilities. The Sugeno-type network is comprised of nodes with
specific functions and waves, and are collected in layers with specific functions.

Figure 0.3 Sugeno Rule Operation

The present thesis is organised into six chapters:


The first chapter provides the brief overview of time series, hydrology and fuzziness in
hydrology. The second chapter deals with the review of the literature on fuzzy logic in
hydrological modelling. It deals with studies related to various fuzzy models used in
hydrological operations. Objectives of the study and area are being discussed in chapter
three. It also deals with the softwares and tools used in the study. The fourth chapter
lays emphasis on detailed methodology applied to attain the outputs. A detailed
explanation of results and inferences drawn are discussed in the fifth chapter. The last
chapter deals with the conclusion and limitations of the work carried out.
.

7
CHAPTER 2
LITERATURE REVIEW

The previous chapter gives the brief review of the work which has to be done. To carry
out the whole work following literature has been studied:

Bardossy, Bogardi et al. (1990) presented a general methodology for fuzzy regression
and explained using hydrological case study. In fuzzy regression, several “goodness of
fit” criteria may be used such as the maximum average vagueness criterion and the
prediction vagueness criterion. The author explained by means of a case study involving
the relationship between soil electrical resistivity and hydraulic permeability. The
regression parameters can be calculated by minimizing a vagueness criterion reflecting
the goodness of the fuzzy regression. This relationship was imprecise and based on only
a few data points. The results presented explained that the prediction vagueness criterion
may lead to a more robust fuzzy than the maximum or average vagueness criteria.

Yu & Yang (2000) presented his study of fuzzy multi-objective function (FMOF) to
improve the performance of regular objective functions like root mean square error
(RMSE) and mean absolute percent error (MAPE). The author used daily rainfall and
runoff measurements with monthly evaporation estimates to calibrate and verify rainfall-
runoff model over 4 and 9 years of the time period. Gao-Ping Creek in southern Taiwan
have a drainage area of 3257km2 with 171km mainstream was taken for modeling.
FMOF modeling results were compared with regular objective functions which mainly
focus the simulation of high and low flow periods separately. In this study FMOF allows
various flow stages of interest to be emphasized in model calibration. The method
proposed was found to be appropriate for basins with extremely heterogeneous temporal
flow distributions.

Hundecha, Bardossy et al. (2001) carried out a study on mathematical methods using
semantic variables rather than using numerical variables. The author made an attempt to
develop a fuzzy rule-based routine to simulate the processes involved in the generation
of runoff from rainfall. The routines were implemented within on Hydrologiska Byråns
Vattenbalansavdelning (HBV) model which is a conceptual and semi-distributed model.

8
Application and validation of the model were carried out on river Neckar in southwest
Germany with a net watershed area of 13957km2. Basin was divided into 41 sub-basins
so that each could be modeled separately. Snowmelt, evapotranspiration, runoff and
basin response were four model components formulated for fuzzy logic-based routines.
In this study author concluded that fuzzy logic-based model gave good results for
observed runoffs and model performed well in low and normal flow conditions and there
was no noticeable difference between the HBV model. The fuzzy logic-based model
showed overestimated peak flows.

Bardossy (2005) created various fuzzy sets in rainfall-runoff modeling by doing


measurements and laboratory experiments at a small scale. The main objective of the
study was to define that how uncertainties are related to an inappropriate knowledge of
the model parameters that describe the various physical processes and also about the
considerable uncertainty in mathematical process describing hydrological models. Fuzzy
rules can describe relationships between variables in a linguistic form in accordance to
their case-specific validity and rules can be usefully employed to describe difficult non-
linear relationships so as to have good results in rainfall-runoff modeling.

Tayfur & Singh (2006) applied artificial neural network (ANN) and fuzzy logic (FL)
for rainfall-runoff predictions and tested these models with kinematic wave
approximation (KWA). In this study author used thirty-six event based data sets, twenty-
four laboratory flume data sets and twelve experimental plots were employed for training
and testing the models to predict peak discharge from rainfall events. The author
concluded that artificial neural network and fuzzy logic models were applied on flume
with less area and on a small scale of the watershed. The models needed to be calibrated
with sufficient site data when applied on large watersheds. Also, the author stated ANN
and FL need large historic data for satisfactory results.

Rivas & Roesner (2006) developed a fuzzy rule based system to study the peak flow
rate over a watershed for six different storm based events. The author developed a fuzzy
rule on a watershed located in the Raleigh of North Carolina with watershed area 3.02
mi2. In the study to discretize the watershed author used Arc Hydro (2003), a GIS
extension on 30m Digital Elevation Model (DEM). Storm Water Management Model
(SWMM) simulations were used to train various sets to develop a fuzzy rule-based
system and the resulting fuzzy rule-based system was compared with EPA SWMM5.

9
Study results show that fuzzy rule system performed well in forecasting peak flow rates
but shown absurd results for the highest return periods (50 and 100 years).

McCuen & Knight (2006) used various fuzzy set analyses for the computation of the
distribution of slope-area discharge estimates which was similar to the distribution
assessed using various statistical methods used in hydrology. The author also studied the
effects of errors in channel roughness, channel width, channel side-slopes, and flow
depth on the accuracy of discharges. The slope-area method was widely used to make
discharge estimates at ungauged locations. Instead of actual velocity measurements,
Manning’s equation was applied. Field measurements are used to characterize a cross
section. Confidence intervals are needed for understanding the accuracy of slope-area
methods and to include in risk assessments. The fuzzy set theory provides the means of
assessing the accuracy of slope-area discharge estimates, including the development of
confidence intervals at a specific site. The method requires supplemental information
about the error distributions of the inputs. The author concluded that for low scour rates,
changes that result from the incision or vegetal growth can render a rating curve to be
short lived. Temporally non stationary conditions may lead to underprediction or over
prediction of discharges in as few as 10 years. Thus, rating curves based on slope-area
analyses should be frequently reanalyzed whenever site conditions are unstable.

Ren, Xiang et al. (2013) studied the forecast modeling of monthly runoff with Adaptive
Neural Fuzzy Inference System and Wavelet Analysis. The author took advantage of
localized characteristics of wavelet transform and the approximation function of an
adaptive neural fuzzy inference system (ANFIS), the combined approach of wavelet
transform and ANFIS was used to predict monthly runoff. The ANFIS forecast model
for monthly runoff was established based on wavelet analysis. In his study author studied
Yichang Hydrologic Station of the Yangtze River which is located in Yichang City,
Hubei Province, China, the contributed catchment of which is 1.0055 million km2. In
this research, historical data were collected for 432 months from 1970 to 2005 in Yichang
Hydrological Station. Based on this, a wavelet analysis and forecast model was
constructed to determine model parameters. Runoff data for 24 months from 2006 to
2007 were used in backtracking to examine the prediction accuracy of the model. Based
on the evaluation of simulated and measured values in Yichang Station, it was found the

10
percentage of the pass of relative error was 100% and the effect of prediction was
acceptable. The certainty factor was 0.91 and the prediction level was A.

Chachi, Taheri et al. (2014) prepared a hybrid fuzzy regression model which handle the
large variation issues in fuzzy data by constructing a variable spread multivariate
adaptive regression splines (MARS) fuzzy regression model with crisp parameters
estimation and fuzzy error terms. The author proposed a two-phase procedure which
applies the MARS technique at phase one and an optimization problem at phase two to
estimate the centre and fuzziness of the response variable. This led in sorting out the
problem of large variation issue and the problem of variation spreads in fuzzy
observations. Empirical results demonstrated that the proposed approach was more
efficient and more realistic than some well-known least-squares fuzzy regression
models.

Raje (2014) used Fuzzy Bayesian approach to study changepoint (CP) detection in
hydrological series of Mahanadi River basin. Annual rainfall and stream flow data sets
were used to prepare a fuzzy Bayesian model. Study was carried out in two steps: the
first step consists of a fuzzy clustering of raw time series which transforms the initial
data with arbitrary distribution into data that can be approximated with a beta distribution
and second step uses the Bayesian approach with the Markov Chain Monte Carlo
(MCMC) method for CP detection in the transformed time series. Above methods were
applied to annual maximum and annual average stream flow and sub basin rainfall for
Basantpur and Hirakud stations on the Mahanadi river in India. Both classical and
Bayesian CP detection methods used show that the annual stream flow and the annual
rainfall have decreased significantly in the Mahanadi Basin, with a possible CP for the
Basantpur station between 1975 and 1980 and for the Hirakud station around 1964.

Tayfur & Brocca (2015) considered soil moisture in modeling rainfall-runoff using
fuzzy logic. Coloroso stream a tributary of Niccone stream a sub-catchment of Tiber
River in central Itlay having a catchment area of 13km2 at Pian Di Marte was considered
for the study. The author developed a Mamdani-type fuzzy logic model to simulate daily
discharges as a function of soil moisture at different depths in the catchment. In this study
for each variable of soil moisture, rainfall and discharge 9 fuzzy subsets were employed
and 30 fuzzy rules were made for relating the input variables (soil moisture & rainfall)
with output variable (discharge). A fuzzy model is based on the range and distribution

11
of the input and output data of the related model variables; hence, the model cannot be
employed for extrapolation studies. For different sized watersheds subjected to different
climatic conditions, the parameters have to develop in different practical ways.

Turan & Yurdusev (2016) prepared a fuzzy conceptual hydrological model for water
flow prediction. The processes of GR2M (modele du Génie Rural à 2 paramètres
Mensuel) model were modeled and replaced by fuzzy logic systems and model was
calibrated using genetic algorithm. GR2M is a well-known monthly conceptual
hydrological model. The study area was located in western part of Turkey. The basin
drainage area was 18,000 km2 and mean annual runoff is 1.95 km3. It was concluded that
Fuzzy- GR2M model performed better that conceptual GR2M model. All R2 values were
greater than 10 % for each basin and stage. These values indicate that desired
improvement has been achieved by replacing the internal processes of conceptual model
with fuzzy systems. This study attempted to improve modeling performance of
conceptual hydrological models by integrating fuzzy logic into them.

Yu and Yang (2000) presented his study of fuzzy multi-objective function (FMOF) to
improve the performance of regular objective functions like root mean square error
(RMSE) and mean absolute percent error (MAPE). The author used daily rainfall and
runoff measurements with monthly evaporation estimates to calibrate and verify rainfall-
runoff model over 4 and 9 years of the time. Gao-Ping Creek in southern Taiwan have a
drainage area of 3257km2 with 171km mainstream was taken for modeling. FMOF
modeling results compared with regular objective functions that mainly focus the
simulation of high and low flow periods separately. In this study, FMOF allows various
flow stages of interest be emphasized in model calibration. The method proposed
receives appropriate response for basins with extremely heterogeneous temporal flow
distributions.

Hundecha, Bardossy et al. (2001) carried out a study on mathematical methods using
semantic variables rather than using numerical variables. The author attempted to
develop a fuzzy rule-based routine to simulate the processes involved in the generation
of runoff from rainfall. The routines implemented within on Hydrologiska Byråns
Vattenbalansavdelning (HBV) model, which is a conceptual and semi-distributed model.
Application and validation of the model carried out on river Neckar in southwest
Germany with a net watershed area of 13957km2. Basin divided into 41 sub-basins so

12
that each can be modeled separately. Snowmelt, evapotranspiration, runoff and basin
response were four model components formulated for fuzzy logic-based routines. In this
study author concluded that fuzzy logic-based model gave good results for observed
runoffs and model performed well in low and normal flow conditions and there was no
noticeable difference between the HBV model. The fuzzy logic-based model showed
overestimated peak flows.

Bardossy (2005) created various fuzzy sets in rainfall-runoff modeling by doing


measurements and laboratory experiments at a small scale. The main objective of the
study was to define that how uncertainties are related to an inappropriate knowledge of
the model parameters that describe the various physical processes and about the
considerable uncertainty in mathematical process describing hydrological models. Fuzzy
rules can describe relationships between variables in a linguistic form in accordance to
their case-specific validity and rules can be usefully employed to describe difficult non-
linear relationships to have good results in rainfall-runoff modeling.

Tayfur and Singh (2006) applied artificial neural network (ANN) and fuzzy logic (FL)
for rainfall-runoff predictions and tested these models with kinematic wave
approximation (KWA). In this study author used thirty-six event based data sets, twenty-
four laboratory flume data sets and twelve experimental plots were employed for training
and testing the models to predict peak discharge from rainfall events. The author
concluded that artificial neural network and fuzzy logic models were applied on flume
with less area and on a small scale of the watershed. The models needed to be calibrated
with sufficient site data when applied on large watersheds. Also, the author stated ANN
and FL need large historic data for satisfactory results.

Rivas and Roesner (2006) developed a fuzzy rule based system to study the peak flow
rate over a watershed for six different storm based events. The author developed a fuzzy
rule on a watershed located in the Raleigh of North Carolina with watershed area 3.02
mi2. In the study to discretize the watershed author used Arc Hydro (2003), a GIS
extension on 30m Digital Elevation Model (DEM). Storm Water Management Model
(SWMM) simulations were used to train various sets to develop a fuzzy rule-based
system and the resulting fuzzy rule-based system was compared with EPA SWMM5.
Study results show that fuzzy rule system performed well in forecasting peak flow rates
but shown absurd results for the highest return periods (50 and 100 years).

13
McCuen and Knight (2006) used various fuzzy set analyses for the computation of the
distribution of slope-area discharge estimates that was similar to the distribution assessed
using various statistical methods used in hydrology. The author also studied the effects
of errors in channel roughness, channel width, channel side-slopes, and flow depth on
the accuracy of discharges. The slope-area method was widely used to make discharge
estimates at ungauged locations. Instead of actual velocity measurements, Manning’s
equation was applied. Field measurements are used to characterize a cross section.
Confidence intervals are needed for understanding the accuracy of slope-area methods
and to include in risk assessments. The fuzzy set theory provides the means of assessing
the accuracy of slope-area discharge estimates, including the development of confidence
intervals at a specific site. The method requires supplemental information about the error
distributions of the inputs. The author concluded that for low scour rates, changes that
result from the incision or vegetal growth can render a rating curve to be short lived.
Temporally nonstationary conditions may lead to underprediction or over prediction of
discharges in as few as 10 years. Thus, rating curves based on slope-area analyses should
be frequently reanalyzed whenever site conditions are unstable.

Raje (2014) used Fuzzy Bayesian approach to study changepoint (CP) detection in
hydrological series of Mahanadi River basin. Annual rainfall and stream flow data sets
were used to prepare a fuzzy Bayesian model. Study was carried out in two steps: the
first step consists of a fuzzy clustering of raw time series which transforms the initial
data with arbitrary distribution into data that can be approximated with a beta distribution
and second step uses the Bayesian approach with the Markov Chain Monte Carlo
(MCMC) method for CP detection in the transformed time series. Above methods were
applied to annual maximum and annual average stream flow and sub basin rainfall for
Basantpur and Hirakud stations on the Mahanadi river in India. Both classical and
Bayesian CP detection methods used show that the annual stream flow and the annual
rainfall have decreased significantly in the Mahanadi Basin, with a possible CP for the
Basantpur station between 1975 and 1980 and for the Hirakud station around 1964.

Tayfur and Brocca (2015) considered soil moisture in modeling rainfall-runoff using
fuzzy logic. Coloroso stream a tributary of Niccone stream, which is a sub-catchment of
Tiber River in central Itlay having a catchment area of 13km 2 at Pian Di Marte, was
considered for the study. The author developed a Mamdani-type fuzzy logic model to

14
simulate daily discharges as a function of soil moisture at different depths in the
catchment. In this study for each variable of soil moisture, rainfall and discharge, nine
fuzzy subsets were being employed and 30 fuzzy rules were made for relating the input
variables (soil moisture & rainfall) with output variable (discharge). A fuzzy model is
based on the range and distribution of the input and output data of the related model
variables; hence, the model cannot be employed for extrapolation studies. For differently
sized watersheds subjected to different climatic conditions, the parameters have to
develop in different practical ways.

Turan & Yurdusev (2016) prepared a fuzzy conceptual hydrological model for
water flow prediction. The processes of GR2M (modele du Génie Rural à 2 paramètres
Mensuel) model were modeled and replaced by fuzzy logic systems and model was
calibrated using genetic algorithm. GR2M is a well-known monthly conceptual
hydrological model. The study area was located in western part of Turkey. The basin
drainage area was 18,000 km2 and mean annual runoff is 1.95 km3. It was concluded
that Fuzzy- GR2M model performed better that conceptual GR2M model. All R2 values
were greater than 10 % for each basin and stage. These values indicate that desired
improvement has been achieved by replacing the internal processes of conceptual model
with fuzzy systems. This study attempted to improve modeling performance of
conceptual hydrological models by integrating fuzzy logic into them.

15
CHAPTER 3
OBJECTIVES & STUDY AREA

The previous chapter tells about the numerous studies and research gaps
involved in hydrological modeling using statistical methods and fuzzy sets. Considering
various research gaps in hydrological modeling various objectives have been formed.
The present study is liable for macro watersheds with hilly terrain and major objectives
have been drawn are:
 To evaluate the orographic effect on Rainfall distribution over the watershed
area.
 To estimate runoff using fuzzy logic and regression method.

Gambar river is a sub-basin of River Sutlej which lies in Western Himalayas and
is bounded by latitude 30◦ 52’ N to 31◦ 13’ 45” N and longitude 76◦ 45’ 07” E to 77◦ 00’
09” E. Gambar watershed has an average elevation of 1100m and drains a total area of
approximately 730 km2. The spread of Gambar watershed is mainly in District Solan
and a few parts of District Bilaspur and Shimla. It comprises around 25 ungauged micro-
watersheds.

Figure 3.1GAMBHAR WATERSHED, HIMACHAL PRADESH

16
Gambar watershed mainly contains Chil, Deodar, Ban and Pine trees. Oak
forests are at higher elevations around humid locations. Besides this indigenous
vegetation, there is ornamental and alien plantation too. It consists of silver oak,
jhakranda, bottlebrush, weeping willows, kachnar, grasses etc. The Gambar watershed
area mainly comprises of loamy and sandy loamy soils. Three classes of soils were
available namely, Scantic, Benson, and Buxton.

 Arc GIS: The geographic information system software package ArcGIS was
developed by the Environmental Systems Research Institute (ESRI). ArcGIS is
designed to create, develop, and interact with new and existing geographic data.
It’s designed to be a complete and integrated system for geographic data
processing. The desktop form of ArcGIS is available with three levels of
functionality. The most basic level is called ArcView, which allows for many
map making, visualization and map analysis capabilities. For creating and
editing spatial data that go into these analyses, Arc Editor adds capabilities on
top of ArcView. Finally, more advanced visualization and analysis tools are
available at the Arc Info level. At all levels, users interact with Desktop ArcGIS
through three interface components: ArcMap, Arc Catalog, and Arc Toolbox.
ArcMap is used to perform map and data-based tasks. Arc Catalog is designed
to browse, organize, and document geographic data in a Windows Explorer-like
fashion. Operations such as previewing, copying, moving, renaming, or deleting
can be performed within this module. Arc Toolbox is the data management and
geo-processing module embedded within ArcMap and Arc Catalog. Task
wizards have been created for the most commonly performed geoprocessing
operations. Some of the functionality includes: importing and exporting,
overlays, buffering, and statistical calculations. In this study it has been used for
preparation of vector maps, geo-referencing and mosaicking of toposheets.
Further, it has been used for processing DEM to get desired features of study
area such as slope map, drainage map. Inverse Distance Weightage Tool in Arc
GIS has been used for preparation of soil moisture map using field data.
 ERDAS Imagine: ERDAS
IMAGINE (Earth Resource Development Assessment System): This is an

17
image processing software mainly utilize for study and analysis of satellite
imagery. You can use them for extraction of Digital Number values of the pixels,
Import export raster and vector satellite image, combine various bands of
satellite imageries, to perform detailed analysis of various objects and
information using the pattern recognition technique, Land use/land cover
analysis. In this study it has been used for spatial data processing, supervised
classification to get land use/cover map of the study area.
 MATLAB: Generation of the Fuzzy Logic based model and to carry out training
and simulations in Adaptive Neuro Fuzzy Inference System
 MS Excel 2013 for statistical analysis.

18
CHAPTER 4
METHODOLOGY

The previous chapter explains the topographic details of Gambar watershed. In


brief, the previous chapter deals with shape, area, soil types and vegetation of the
watershed. In this chapter, the complete methodology explained to achieve the
objectives of the study. The objective of the study achieved in four phases as shown in
Figure 4.1.

In the previous chapter, the details of the study area were mentioned. After
finalizing the study area various data sets are required to generate the model. Data
collection has been done in two phases. Phase one deals with the collection of field data
sets and Phase two deals with the procurement of satellite data. Phase three deals with
linear regression analysis on training datasets, and testing datasets of rainfall-runoff. It
also includes setting up of fuzzy inference system rules. Phase four deals with training
of Adaptive-Neuro Fuzzy Inference System and carrying out simulations.

4.2.1 Field Data Procurement

 Procurement of digital toposheets from Survey of India office at


Chandigarh.
 Daily Rainfall-Runoff data for the months June-September (2012-2016)
of the metrological station (BBMB) situated at Kohu village in District
Bilaspur Himachal Pradesh. Daily Rainfall data of two stations Kunihar
and Kandaghat have been procured from NCML Hyderabad.
 Field campaigns to procure soil moisture data across 24 stations situated
in Gambhar watershed using digital TDR and GPS in September 2016.

4.2.2 Spatial Data Procurement

SRTM Digital Elevation Model (DEM) of resolution (30x30) m procured from


https://gdex.cr.usgs.gov/gdex/

19
Landsat 8 OLI sensor’s L1 data product with resolution (30x30) m from
https://earthexplorer.usgs.gov/

• SELECTION OF WATERSHED (GAMBHARKHAD)


• COLLECTION OF TOPOSHEETS (SURVEY OF INDIA)
• GEOREFERENCING & MOSIACING THE TOPOSHEETS
• DELINEATION OF DRAINAGE NETWORK
• HYDROLOGICAL DATA OF WATERSHED
• CHECK ON OROGRAPHIC EFFECT USING ANOVA ON
PHASE-1 RAINFALL DATA OF THREE STATIONS
• ORGANISING FIELD CAMPAIGNS FOR COLLECTION OF
SOIL MOISTURE DATA

• COLLECTION OF SATELITE DATA (LANDSAT 8 OLI


(30m*30m), DEM ASTER (30m*30m)
• EXTRACTION OF BASIN BOUNDARY, SLOPE MAP,
DRAINAGE MAP, TIN FROM DEM USING ArcMap 10.4.1
• PLOTTING SOIL MOISTURE DATA IN ArcGIS & USING
INVERSE DISTANCE WEIGHTAGE TOOL
INTERPOLATION OF SOIL MOISTURE VALUES IN BASIN
PHASE-2 • LANDUSE LANDCOVER MAP FROM LANDSAT 8 OLI
IMAGE USING ERDAS IMAGINE

• SELECTION OF DATA FOR TRAINING,TESTING AND


VALIDATION
• REGRESSION ANALYSIS OF RAINFALL-RUNOFF ON
TRAINING AND TESTING DATA
• SELECTION OF FUZZY MODEL (ADAPTIVE NETWORK
PHASE-3 BASED FUZZY INFERENCE SYSTEM)

• TRAINING OF MODEL USING 40%-60% DATA IN


MATLAB'S NEURO FUZZY DESIGNER TOOL
• GENERATING FIS AND COLLECTION OF OUTPUT USING
DATA FOR SIMULATIONS
• VALIDATION OF OUTPUT WITH GROUND DATA
• COMPARING FIS RESULT AND REGRESSTION RESULT
PHASE-4 WITH OBSERVED DATA

Figure 4.1Methodology

20
The orographic effect is a change in atmospheric conditions caused by a change
in elevation, primarily due to mountains. Gambhar watershed lies on an average
elevation between 450m to 2200m. Orographic effect on rainfall data was done using
ANOVA method. Station Kahu is at a lower altitude than Kunihar and Kandaghat.
Single factor ANOVA test has been carried at the significant limit of 0.05 and 0.01. F-
test shows that there is no significant orographic effect on rainfall distribution in the
watershed. FCal < FCritical which shows that the rainfall distribution is uniform in the
catchment area. The results of F-test are discussed in next chapter.

Data processing has been done on two different set of data using various tools
and software discussed in previous chapters.

4.4.1 Field Data Processing


Digital toposheets numbering H43E16, H43F4, H43K13 and H43L1 with 1:50,000-
scale having 600dpi resolution were geo referred in ArcMap10.4.1 and then mosaicked
as shown in Figure 4.2.
Drainage network is created by digitizing and a shapefile of the same was
generated. With the help of the drainage network, it was easy to identify the study area
on digital elevation model and satellite images.

Figure 4.2 Mosaicked Toposheets

21
4.4.2 Digital Elevation Model Processing
SRTM 30m resolution Digital Elevation Model was processed using Hydrology
toolbox in ArcGIS. Extraction of various features like slope, drainage etc.has been done
using above mentioned tool in different steps. The process starts with DEM as input to
Arc Map followed by filling up of sinks, computation of flow direction and then flow
accumulation raster dataset.

Figure 4.3 SRTM DEM with Drainage

Drainage network and basin boundary was extracted using flow accumulation
raster. Stream order for Gambhar River designated as 1,2,3,4 in ArcGIS as shown in
Figure 4.4.

Figure 4.4 Stream Order and Basin Boundary

22
Contour map, slope map, and Triangular Irregular Network (TIN) map were
generated from the DEM as shown in Figure 4.5 and Figure 4.6 respectively

Figure 4.5 Contour Map (50m Intervel)

Figure 4.6 Slope Map Gambhar Watershed

23
4.4.3 Soil Moisture Map
Volumetric water content is a numerical measure of soil moisture. It is simply the
ratio of water volume to soil volume. Volumetric moisture content (VMC) data was
collected using digital TDR instrument by conducting field visits to 23 different points
in the study area. The TDR 300 has two volumetric water content modes; one for
standard soils and one for higher clay soils. In volumetric water content (VWC) mode,
the meter converts a measured electrical signal into percent soil moisture content using
an equation valid over a wide range of mineral soils. The data was collected with ground
coordinates logged with the help of GPS (Table_) and plotted in ArcGIS to have a point
data shapefile. Soil moisture map was prepared using Inverse Distance Weighted
interpolation (IDW) in ArcGIS as shown in Figure 4.7.

Figure 4.7 Soil Moisture Map of Gambhar Watershed

The Operational Land Imager (OLI) and Thermal Infrared Sensor (TIRS) are
instruments on board the Landsat 8 satellite, which was launched in February of 2013.
The satellite collects images of the Earth with a 16-day repeat cycle, referenced to the

24
Worldwide Reference System-2. The approximate scene size is 170 km north-south by
183 km east-west. LANDSAT 8 is equipped with Operational Land Imager (OLI) and
Thermal Infrared Sensor (TIRS) former contains Band 1 to Band 9 while TIRS is
equipped with two thermal bands as shown in Figure 4.8

Table 4.1 LANDSAT 8 Band Details

4.5.1 Normalized Difference Vegetation Index


Normalized Difference Vegetation Index (NDVI) is a standardized index allows
generating an image displaying greenness (relative biomass). This index takes
advantage of the contrast of the characteristics of two bands from a multispectral raster
dataset, the chlorophyll pigment absorptions in the red band and the high reflectivity of
plant materials in the near infrared (NIR) band. NDVI lies between -1 to 1 with values
close to -1 to 0 relates the presence of snow, water and built up areas while the values
between 0.2 to 0.5 relates the presence of low vegetation, grasslands, pasturies etc. The
values between 0.5-0.9 relates dense forest or highly vegetative land. In present study,
the NDVI lies between 0.30 to 0.52 which shows that Gambhar watershed contains
shrubs, grasslands and few area covering dense forests as shown in Figure 4.8.

25
To calculate NDVI following steps has been followed using Raster Algebra tool
in Arc Toolbox:
 OLI spectral radiance data converted to ToA planetary reflectance using
reflectance-rescaling coefficients provided in the landsat8 OLI metadata file.
Reflectance correction is applied on band 4 and band 5. The following equation
is used to convert DN values to ToA reflectance for OLI image:
 ρλ’ = (M ρQcal + A ρ) where:
 ρλ’ = TOA planetary reflectance, without correction for the solar
angle. ρλ’ does not contain a correction for the sun angle.
 M ρ = Band-specific multiplicative rescaling factor from the
metadata (Reflectance_Mult_Band_x, where x is the band
number)
 A ρ = Band-specific additive rescaling factor from the metadata
(Reflectance_Add_Band_x, where x is the band number)
 QCal = Quantized and calibrated standard product pixel values
(DN)
 In this step, Reflectance with a correction for the sun angle is done using
following formula:
 ρλ = (ρλ‘/CosθSZ) = (ρλ ‘/SinθSE) where:
 ρλ = TOA planetary reflectance
 θSE = Local sun elevation angle. The scene center sun elevation
angle in degrees is given in the metadata (Sun Elevation).
 ΘSZ = Local solar zenith angle; θSZ = 90° – θSE.
 Further, after applying corrections, the formula for calculation of NDVI is
applied.
 NDVI= (NIR-RED) / (NIR+RED)

26
Figure 4.8 NDVI Gambhar Watershed

4.5.2 Land Use Land Cover Map


LANDSAT 8 OLI sensor’s data is used in the present study to prepare LANDUSE
map of the Gambhar watershed. In supervised classification the user or image analyst
“supervises” the pixel classification process. The user specifies the various pixels values
or spectral signatures that should be associated with each class. This is done by selecting
representative sample sites of known cover type called Training Sites or Areas. The
computer algorithm then uses the spectral signatures from these training areas to classify
the whole image. Ideally the classes should not overlap or should only minimally overlap
with other classes. Supervised classification is done in ERDAS IMAGINE by preparing
number of training samples for each class from whole study area. Maximum likelihood
classification method was applied for the classification. Maximum Likelihood
Classification assumes that the statistics for each class in each band are normally
distributed and calculates the probability that a given pixel belongs to a specific class.
Each pixel is assigned to the class that has the highest probability (that is, the maximum
likelihood). This is the default.The study area majorly consists of open grasslands, shrubs
and large forest covers with shallow water bodies as shown in the Figure 4.10.

27
Figure 4.9 Landuse and cover Map

Modify network-based fuzzy inference (ANFIS) is a combination of two soft-


computing methods of ANN and fuzzy logic (Jang, 1992). Fuzzy logic has the ability to
change the qualitative aspects of human knowledge and insights into the process of
precise quantitative analysis. However, it does not have a defined method that can be
used as a guide in the process of transformation and human thought into rule base fuzzy
inference system (FIS), and it also takes quite a long time to adjust the membership
functions (MFs) (Jang, 1992). Unlike ANN, it has a higher capability in the learning
process to adapt to its environment. Therefore, the ANN can be used to automatically
adjust the MFs and reduce the rate of errors in the determination of rules in fuzzy logic.

4.6.1 Adaptive Network


Adaptive network is one example of feed forward neural network with multiple
layers Figure 4.11. In the learning process, these networks often use supervised learning
algorithm. In addition, adaptive network has the architecture characteristics that consists

28
of a number of adaptive nodes interconnected directly without any weight value between
them. Each node in this network has different functions and tasks, and the output
depends on the incoming signals and parameters that are available in the node. A
learning rule that was used can affect the parameters in the node and it can reduce the
occurrence of errors at the output of the adaptive network (Jang, 1992).

Figure 4.10Adaptive Network

In present study the ANFIS structure is based on Takagi–Sugeno Model as


shown in Figure 4.12.

Figure 4.11 ANFIS Structure.

29
FIS was built on the three main components, namely basic rules, where it
consists of the selection of fuzzy logic rules “If-Then;” as a function of the fuzzy set
membership; and reasoning fuzzy inference techniques from basic rules to get the
output. Figure 4.13 shows the detailed structure of the FIS. FIS works when the input
that contains the actual value is converted into fuzzy values using the fuzzification
process through its membership function, where the fuzzy value has a range between 0
and 1. The basic rules and databases are referred to as the knowledge base, where both
are key elements in decision-making. Normally, the database contains definitions such
as information on fuzzy sets parameter with a function that has been defined for every
existing linguistic variable.

Figure 4.12 Fuzzy Inference System

In this study Fuzzy Inference System was developed using Sugeno type network
in mat lab. Gaussian membership functions has been used to develop the fuzzy inference
system. In the present study “If –Then” rules were selected as Fuzzy Rule Base System
on the basis of datasets.
The process of formulation of Fuzzy Model is done in MATLAB’s Fuzzy Logic
Designer toolbox. Figure 4.14 and 4.15 shows the designing of FIS and membership
functions using ANFIS respectively.

30
Figure 4.13 FIS Editor

Figure 4.14Membership Function Editor for Input and Output

31
CHAPTER 5
RESULTS & DISCUSSIONS

This section describes the results related to orographic effect over the catchment
area, rainfall – runoff regression analysis on training datasets and generation of runoff
data by applying regression equation on simulation datasets. The datasets selected were
based on daily Rainfall values. Coefficient of determination (R2) is adopted as a
statistical parameter for comparison of Rainfall – Runoff results obtained by linear
regression analysis and results observed from ANFIS.

The orographic effect on rainfall data has been checked using statistical method
ANOVA. One-way ANOVA has been performed on net monsoon rainfall data from
(June-Sept) 2012 - 2016 of three rain guages situated at three different elevations (Kahu,
Kunihar, and Kandaghat) respectively (Table 5.1).

Table 5.1 Rainfall Data for ANOVA

Rainfall Rainfall Rainfall


Year
(Kahu) (Kunihar) (Kandaghat)
2012 954.8 748.72 823.63
2013 708.2 854.56 801.65
2014 1374.8 1200.47 1228.19
2015 658.5 465.18 436.66
553.7 705.97 714.37
2016

F-test was conducted at 95% confidence interval and the results conclude that
there was no orographic effect in the catchment area as FCritical > FCalculated. Table 5.2
shows the detailed ANOVA results with values FCalculated = 0.52574 which is less than
FCritical = 3.885294.

32
Table 5.2 ANOVA Table

One-Way ANOVA Along Columns

Anova: Single
Factor
SUMMARY
Groups Count Sum Average Variance
KAHU 5 4250 850 107742.8
KUNIHAR 5 3974.9 794.98 71700.68
KANDAGHAT 5 4004.5 800.9 80813.04
ANOVA Results
Source of Variation SS df MS F P-value F crit
Between Groups 9121.761 2 4560.881 0.052574 0.949002 3.885294
Within Groups 1041026 12 86752.18
Total 1050148 14

Linear regression was applied on daily Rainfall-Runoff series of monsoon data


from year 2013 to 2016 in which coefficient of determination (R2) was found to be
0.3871 as shown in Figure 5.1.

Rainfall-Runoff Regression
120

100

80
Runoff (cusec)

60

40

20

0
0.00 10.00 20.00 30.00 40.00 50.00 60.00 70.00 80.00 90.00
Rainfall (mm)

R-R Regression Linear (R-R Regression) R² = 0.3871

Figure 5.1 Linear Regression Daily Rainfall-Runoff Data

33
Figure 5.1 shows the regression line fittings for the determination of Runoff in
each catchment. In case of hydrological series the scatter of rainfall runoff data is
considerable, which suggests instability in the rainfall – runoff relationship. In this study
the application of Sugeno FIS model with constant output membership function is
performed. FIS was trained using 60% of the data with 0.66 as coefficient of
determination as shown in Figure 5.2.

Training Data
120

100

80
Runoff (cumecs)

60

40

20

0
0 10 20 30 40 50 60 70 80 90
Rainfall (mm)

Series1 Linear (Series1) R² = 0.6695

Figure 5.2 Training Data Regression

Hybrid optimization method was adopted in training of FIS with 20 iterations.


The average training error was found to be 8.7344 Figure 5.3. FIS was tested against the
training data as shown in Figure 5.3.

34
Figure 5.3 Training FIS

30% data was used for testing of FIS and relative error was calculated using the same
process. For testing data the relative error was found to be 8.31 between training and
testing datasets as shown in Figure 5.4.

Figure 5.4 Testing FIS

35
5.4.1 Simulation
After the training and testing of datasets the model was simulated against
different rainfall values on Fuzzy Inference System to predict runoff. The same rainfall
values were given as input to the linear regression equation which was calculated in
section 5.3. Both outputs were plotted against the observed runoff values. The
coefficient of determination for FIS predicted runoff was 0.3321 while for runoff
predicted from regression it was 0.2992 as shown in Figure 5.5.

Runoff (Observed) v Runoff (Predicted)


120
100
Runoff Predicted

80
60
40
20
0
0 20 40 60 80 100 120
Observed Runoff
R² = 0.3321 R² = 0.2992
Obs v Fuzzy Obs v Regression
Linear (Obs v Fuzzy) Linear (Obs v Regression)

Figure 5.5 Observed vs. Predicted Runoff

The relative errors for each runoff prediction through the classical regression and
fuzzy models have been presented in the study. It is observed that, invariably, the fuzzy
approach provides better estimates from classical regression rainfall runoff relationship
because in Table 5.3 FL model prediction yields less relative error. The relative error
with FL model was 2.84 while for classical regression it was 3.80. The acceptable error
limit is 10% so both models performed well but FL model was better than classical
regression model.

36
Table 5.3 Relative Error Calculation

Rainfall( Runoff Runoff Runoff


Fuzzy Absolute error F relative Error RR Absolute Error RR Relative error
mm) (Observed) (Fuzzy) (Regression)
4.674316 22.7615 10.0285 16.7566097 -12.733 -0.559409529 -6.0048903 -0.263817863
5.008979 31.70244444 10.2912 17.11871485 -21.41124444 -0.675381499 -14.5837296 -0.460019088
5.65814 29.95495652 10.8309 17.82110783 -19.12405652 -0.638427117 -12.13384869 -0.405069815
5.744628 35.26177778 10.9057 17.91468745 -24.35607778 -0.690721776 -17.34709033 -0.491951666
6.507819 21.77488889 11.5955 18.74046054 -10.17938889 -0.467482931 -3.034428352 -0.139354482
6.862733 31.06141667 11.9336 19.12447696 -19.12781667 -0.615806319 -11.9369397 -0.3843012
7.16284 45.595 12.2276 19.44919268 -33.3674 -0.731821472 -26.14580732 -0.573435844
12.44679 26.62066667 18.1216 25.16642771 -8.499066667 -0.319265733 -1.45423896 -0.054628195
17.1356 9.621708333 23.0396 30.23971852 13.41789167 1.394543588 20.61801018 2.142863769
17.27454 79.83733333 23.1711 30.39005657 -56.66623333 -0.709771118 -49.44727677 -0.619350305
17.51719 21.38155556 23.3994 30.65259711 2.017844444 0.094373136 9.271041554 0.433599956
17.90939 0.368 23.7649 31.07695479 23.3969 63.57853261 30.70895479 83.44824672
18.50894 19.86166667 24.317 31.72567155 4.455333333 0.224318201 11.86400488 0.597331789
20.24493 2.463777778 25.901 33.60401356 23.43722222 9.512717597 31.14023578 12.6392226
23.72867 51.35766667 29.3216 37.3734199 -22.03606667 -0.429070635 -13.98424676 -0.272291318
23.91915 11.27155556 29.5276 37.57952275 18.25604444 1.619656165 26.30796719 2.334013887
24.07713 23.965 29.7004 37.75045005 5.7354 0.239324014 13.78545005 0.575232633
24.38074 30.44411111 30.0373 38.07895989 -0.406811111 -0.013362555 7.634848782 0.25078245
24.84098 53.77644444 30.5605 38.57693969 -23.21594444 -0.43171215 -15.19950476 -0.282642427
25.28874 25.64533333 31.0843 39.06141709 5.438966667 0.21208407 13.41608376 0.523139379
25.73201 11.951 31.6171 39.54103778 19.6661 1.645561041 27.59003778 2.308596584
26.09914 1.671 32.0691 39.9382748 30.3981 18.19156194 38.2672748 22.90082274
29.88088 13.56511111 37.1849 44.03010791 23.61978889 1.741216028 30.4649968 2.245834667
30.05534 6.655333333 37.4349 44.21888122 30.77956667 4.624797155 37.56354788 5.644127199
30.41401 6.916555556 37.9506 44.60695486 31.03404444 4.486921878 37.6903993 5.449301896
34.18531 107.5843333 43.4385 48.68750203 -64.14583333 -0.59623768 -58.8968313 -0.54744803
34.27903 47.77816667 43.5763 48.78890852 -4.201866667 -0.087945331 1.010741854 0.02115489
34.60309 49.40266667 44.0537 49.13954079 -5.348966667 -0.108272833 -0.263125874 -0.005326147
40.80059 84.92233333 54.8715 55.8452411 -30.05083333 -0.353862549 -29.07709223 -0.342396294
43.27681 98.62 61.3061 58.52450428 -37.3139 -0.378360373 -40.09549572 -0.406565562
44.04709 102.556 63.6792 59.35794628 -38.8768 -0.379078747 -43.19805372 -0.4212143
47.47589 26.26133333 76.1914 63.06791328 49.93006667 1.901276909 36.80657995 1.401550313
48.98547 88.61333333 82.2987 50.558 -6.314633333 -0.071260533 -38.05533333 -0.429453807
50.52077 110.35 88.4401 70.668 -21.9099 -0.198549162 -39.682 -0.359601269
53.21895 46.47644444 98.0773 69.28190157 51.60085556 1.110258243 22.80545713 0.490688507
79.1202 83.55655556 102.3123 101.26 18.75574444 0.224467659 17.70344444 0.211873794
2.842939172 3.80998656

In systems modeling and control, there are many difficulties which are
commonly experienced by practicing engineers. For instance, it is generally difficult to
accurately model a complex process by a mathematical model. The methodology of the
fuzzy-logic modeling and control, based on fuzzy set theory and fuzzy logic, appears
promising when the phenomena are too complex for analysis by conventional
quantitative techniques, when the available sources of information are interpreted
qualitatively, inexactly or uncertainly, and/or when qualitative and often conflicting
performance objectives are considered.

37
Present study shows that the fuzzy-logic modeling and control may be viewed
as a step towards a rapprochement between conventional and precise analytical
approach and human-like decision making.

38

Vous aimerez peut-être aussi