Journal of Hydrology 404 (2011) 304322

Contents lists available at ScienceDirect

Journal of Hydrology
journal homepage: www.elsevier.com/locate/jhydrol

Assessment of different precipitation datasets and their impacts on the water balance of the Negro River basin
A.C.V. Getirana a,b,, J.C.V. Espinoza c,d, J. Ronchail e, O.C. Rotunno Filho a
a

Universidade Federal do Rio de Janeiro, Programa de Engenharia Civil, Rio de Janeiro, Brazil Universit de Toulouse, UPS (OMP), IRD, LEGOS, Toulouse, France c Instituto Geofsico del Peru, Lima, Peru d Universidad Agraria La Molina, Lima, Peru e Universit Paris 7, LOCEAN/IPSL, Paris, France
b

a r t i c l e

i n f o

s u m m a r y
With the objective of understanding the potential and limitations of available precipitation products for hydrological studies, this paper compares six daily and sub-daily precipitation datasets and their impacts on the water balance of the Negro River basin in the Amazon basin. The precipitation datasets contain gauge-based data [data derived from the Hybam Observatory Precipitation (HOP) dataset and provided by the Climate Prediction Center (CPC)], satellite-based data [the Global Precipitation Climatology Project (GPCP) one-degree daily and TRMM Multisatellite Precipitation Analysis (TMPA) datasets] and modelbased data [the NCEP-DOE AMIP-II re-analysis (NCEP-2) and 40-year ECMWF Re-Analysis (ERA-40) datasets]. Each dataset has a common set of meteorological forcing data which are used to run the MGB-IPH hydrological model for the period from January 1998 to August 2002. The average precipitation of all the datasets is 2542 mm for the Negro River basin, with a standard deviation of 317 mm. TMPA and NCEP-2 have the lowest (2216 mm/year) and the highest (3065 mm/ year) precipitation rates, respectively. The HOP and CPC datasets agree best with observed discharge. GPCP gives the best results among the ungauged datasets, followed by ERA-40. TMPA and NCEP-2 are found to be the least accurate. TMPA can reproduce the water cycles reasonably well, but underestimates the precipitation elds and discharges over the basin, while NCEP-2 is unable to represent the rainfall quantity and cycles, and the water discharge. Results suggest that gauge-based data are still the most representative of the actual precipitation in the northern Amazon basin. However, some satellite and modelbased can reproduce fairly well the water cycle at the basin scale and monthly time step. 2011 Elsevier B.V. All rights reserved.

Article history: Received 18 May 2010 Received in revised form 24 March 2011 Accepted 30 April 2011 Available online 11 May 2011 This manuscript was handled by K. Georgakakos, Editor-in-Chief, with the assistance of Christa D. Peters-Lidard, Associate Editor Keywords: Water balance Precipitation dataset Hydrological modeling Negro River basin Amazon River basin

1. Introduction In the past several decades, numerous global and quasi-global precipitation datasets have been developed at different time scales using input sources such as ground-based observations, satellite estimates and outputs from general circulation models (Willmott et al., 1994; Kalnay et al., 1996; Sorooshian et al., 2000; New et al., 2000; Adam and Lettenmaier, 2003; Rudolf and Schneider, 2005; Uppala et al., 2005; Huffman et al., 2007; among others). In view of the many precipitation datasets available, several analyses have been carried out to identify their differences at regional (e.g. Gebremichael and Krajewski, 2004; Dinku and Anagnostou, 2005), continental (e.g. Syed et al., 2004; Marengo, 2005; Juarez et al., 2009) and global (e.g. Sapiano et al., 2006) scales. Most studies have found that the datasets typically agree in terms of the
Corresponding author at: Universidade Federal do Rio de Janeiro, Programa de Engenharia Civil, Rio de Janeiro, Brazil. E-mail address: augusto@coc.ufrj.br (A.C.V. Getirana).
0022-1694/$ - see front matter 2011 Elsevier B.V. All rights reserved. doi:10.1016/j.jhydrol.2011.04.037

main temporal precipitation trends and their global spatial distribution but, regionally, they often exhibit marked differences. For example, Costa and Foley (1998) highlighted regional differences for the Amazon basin, and Adler et al. (2001) showed differences among datasets globally. Other comparisons suggest that the largest differences among available precipitation datasets occur in the tropics (Fekete et al., 2004), particularly in the Amazon basin (Rao et al., 2002). Our main objective is to compare the spatio-temporal heterogeneities of six precipitation datasets at the daily time step over the Negro River basin, in the northern Amazon basin (Fig. 1). The comparison of satellite-based and model-based data to gauge-based datasets allows us to identify the strengths and weaknesses of the different precipitation products. Another objective of this paper is to analyze the effects of different precipitation datasets on the simulated water cycle of the Negro River basin. Indeed, comparing simulated and observed discharges can be an efcient way to complement assessments of precipitation datasets. Several studies of this type have been already carried out in recent years (e.g. Yilmaz

A.C.V. Getirana et al. / Journal of Hydrology 404 (2011) 304322

305

Fig. 1. Map of the Negro River basin, including the locations of pluviometric stations with data for the period 19802006, and of six uviometric stations used to evaluate model performance. The additional Colombian and Venezuelan stations are found within the dashed lines.

et al., 2005; Wilk et al., 2006). Here, the MGB-IPH model (Collischonn et al., 2007), forced with different precipitation datasets and using the same atmospheric forcings (i.e. solar radiation, pressure, relative humidity, wind speed and temperature at sea level), has been used to simulate the daily hydrological processes of the Negro River basin for the period from January 1998 to August 2002, for which all datasets are available. Then, sets of simulated runoff, evapotranspiration and soil moisture content have been compared among themselves and water discharge time series have been evaluated with observed data in order to quantify the reliability of each dataset in terms of providing estimates of the water cycle. Evaluating spatio-temporal differences between data sources is particularly useful for identifying how various components of the water cycle (such as runoff, evapotranspiration and soil moisture) are affected by changes in the precipitation eld used in the hydrological models. So, this paper provides important information to both the hydrological and rainfall-retrieval communities. This paper is organized as follows: Section 2 gives a brief description of the Negro River basin and the datasets used in this study. It also provides information about MGB-IPH model, the modeling setup for the study area and the methodology used to evaluate the datasets. In Section 3, the results obtained are presented and discussed. Finally, the conclusions are presented in Section 4. 2. Data sets and methodology 2.1. Study area and available data sets The Negro River basin covers parts of both the northern and southern hemispheres with an area of about 712,000 km2, from 3140 S to 5800 N latitude and from 72570 W to 58160 W longitude, in the northern Amazon basin (Fig. 1). The Negro River is the most important tributary of the Solimes/Amazon River in terms of runoff [4.36 mm/day the mean runoff of the Amazon basin is about 2.9 mm/day (Marengo, 2005)], and the second after the Madeira River in terms of total discharge.

Daily discharge data are available at several gauge stations within the Negro River basin and are freely provided by the Brazilian Water Agency (ANA Agncia Nacional de guas). These stations drain areas varying from 611 km2 to 291,150 km2 [a list of most of gauge stations in service in the Negro River basin can be found in Getirana et al. (2010)]. Five gauge stations (Caracara, Taraqua, Cucu, Curicuriari and Serrinha) representing different regions and hydrological regimes of the basin have been selected to evaluate simulated discharges. The total discharge produced in the basin cannot be evaluated since daily discharges are not available at the basins outlet. However, other hydrological variables such as precipitation, runoff, evapotranspiration and soil moisture are analyzed at the basin scale (represented by the drainage area of the Negro River when it passes by Manaus). The main characteristics of these stations are given in Table 1. Indeed, the Negro River basin is one of the rainiest regions of the Amazon basin (about 3000 mm/year). The Guyana Shield, one of the three cratons of the South American Plate that underlies Guyana, French Guiana, Suriname and parts of Colombia, Venezuela and Brazil, creates a strong gradient in the rainfall spatial distribution over the basin: humid ow from the Atlantic is hindered by

Table 1 Main characteristics of the sub-basins dened by the six gauge stations considered in this study. Station Drainage area (km2) Mean discharge (m3/s) 2903 2755 4940 12,.613 18,.082 35,943 (mm/ year) 731 1972 2207 2092 2211 1591 Mean precipitation (mm/year) 2044 3557 3192 3441 3241 2667 Upstream sub-basin

Caracara Taraqua Cucu Curicuriari Serrinha Manaus

126.085 44.255 71.132 191.787 291.150 712.451

Cucu and Taraqua Curicuriari Caracara and Serrinha

306

A.C.V. Getirana et al. / Journal of Hydrology 404 (2011) 304322

Number of stations 100 80 60 40 20 1980

Data availability (%) 100 90 80 70 60 2006

1993

Fig. 2. Number of precipitation gauge stations and precipitation data availability provided by the Brazilian Water Agency for the Negro River basin for the 19802006 period.

the shield, thereby resulting in less frequent and intense rain events on the North-eastern side of the Negro River basin (Salati et al., 1978; Figueroa and Nobre, 1990; Espinoza Villar et al., 2009). Thus, annual rainfall ranges from less than 1200 mm/year in the driest region (north-eastern basin) to more than 5300 mm/ year in the most humid parts (north-western basin). The Roraima highlands and the Neblina Peak, the highest mountain in Brazil (Pico da Neblina, 2994 m), are located in this region. The hydrological cycle of the basin is characterized by a very wet period between May and August, with precipitation peaks occurring in June and July and no real dry season (Figueroa and Nobre, 1990; Guyot et al., 1993). An increasing interest in the hydrology of the Amazon basin began near the end of the 1970s. This is reected by the growing number of gauge stations operating in the region (Silva et al., 2007). Nevertheless, the basin still has a very low gauge density ($0.12 stations per 1000 km2) (Espinoza Villar et al., 2009), when compared to other large basins in South America, such as the La Plata basin, which had a density of 0.63 to 1.06 stations per 1000 km2 during the period 19982006 (Su et al., 2008). The Negro River basin also has an increasing number of gauge stations over the last three decades. Fig. 2 shows the growth of the Brazilian precipitation gauge network within the basin between 1980 and 2006. The number of stations providing daily precipitation observations increased by about 130% as reported by ANA. In contrast, the data availability, dened as the total number of observations averaged for all stations, remained relatively constant ($85%) over the 27year period, except for a signicant drop in 1991 that coincided with a political crisis in the country. Currently available gauge-based precipitation datasets for the Amazon basin consist mostly in publicly available data, such as those provided by ANA, with limited a number of gauged data from the other countries composing the basin (Rudolf et al., 1994; New et al., 2000; Silva et al., 2007). To acquire a more comprehensive understanding of the water cycle of the Amazon basin, the Hydrology and Geodynamics of the Amazon Basin (Hybam) environmental research observatory (Cochonneau et al., 2006), through international partnerships, created a data base containing hydrological data from most of countries that make up the Amazon basin. Espinoza Villar et al. (2009) rst presented a monthly Hybam Observatory Precipitation (further referred to as HOP) dataset using data from 1446 gauge stations throughout the basin collected between 1964 and 2003 by national agencies from Brazil, Peru, Bolivia, Colombia and Ecuador. This database has been analysed using the regional vector method (RVM) (Hiez, 1977) in order to assess its quality, and 756 stations have been nally retained. 2.2. Precipitation gridded datasets This section provides an overview of the major characteristics of the precipitation datasets considered in this study (see Table 2).

They are (1) the Hybam Observatory Precipitation (HOP) dataset (Espinoza Villar et al., 2009) for the Negro River basin (considered as the reference dataset); (2) the Climate Prediction Center (CPC) dataset (Silva et al., 2007); (3) the Global Precipitation Climatology Project (GPCP) one-degree daily dataset (Huffman et al., 2001); (4) the Tropical Rainfall Measuring Mission (TRMM) Multisatellite Precipitation Analysis (TMPA) dataset (3B42 version 6 V6) (Huffman et al., 2007); (5) the 40-yr ECMWF Re-Analysis (ERA-40) (Uppala et al., 2005) precipitation dataset; and (6) the National Centers for Environmental Prediction Department of Energy (NCEP DOE) Atmospheric Model Intercomparison Project (AMIP-II) reanalysis dataset (referred to as NCEP-2 hereafter) (Kanamitsu et al., 2002). The datasets are available in both daily or sub-daily time steps. They were selected since they have been widely used and evaluated in previous hydrological studies (e.g. Ribeiro Neto et al., 2005; Gonalves et al., 2006; Collischonn et al., 2008; Su and Lettenmaier, 2009; Su et al., 2008; Voisin et al., 2008). It is important to highlight that none of the gauge-based datasets used in this study (e.g. HOP, CPC and CRU the latter one is presented in Section 3) has been corrected for gauge biases, including undercatch errors. The correction of individual records requires detailed local meteorological and station meta-information, which are not readily available (New et al., 2000). 2.2.1. The HOP and CPC gauge-based datasets The HOP Negro River basin precipitation eld was created with data from 115 gauge stations distributed between Brazil (98), Colombia (14) and Venezuela (3) (Fig. 1) for the 19802006 period. These stations are operated by ANA, MARN (Ministerio de Medio Ambiente y de los Recursos Naturales) and IDEAM (Instituto de Hidrologa, Meteorologa y Estudios Ambientales), respectively. Brazilian data are freely available at the daily time step. Colombian and Venezuelan data were acquired at the monthly time step and were homogeneously distributed over the number of days of each month. The well-known inverse squares of distances approach was used to spatially distribute the precipitation data throughout the basin for each computational cell of the hydrological model (more details are found in the Modeling setup section). Only those stations located within a radius of two times the closest distance between each computational cell and stations were considered in the interpolation procedure. HOP has been selected as the reference precipitation dataset because of its relatively dense and more comprehensive network in the Amazon basin when compared to the other gauge-based datasets presented in the literature. This does not mean that HOP (or any other precipitation dataset) represents the truth. All datasets are wrought with error. Gauge-based datasets are affected by both (1) error in gauge observations (systematic bias and random error) and (2) interpolation errors (e.g. station density inhomogeneities and non-representative station locations). The Climate Prediction Center (CPC) of the National Oceanic and Atmospheric Administration (NOAA) provides a South America gridded precipitation dataset from 1978 to 2007 (Silva et al., 2007). The daily gauged data are gridded at a 1.0 1.0 resolution over South America, using a modied Cressman scheme. A previous work suggests that CPC provides the most accurate precipitation estimates over the Amazon basin at a daily time step (Juarez et al., 2009). This dataset comprises the data available in Brazil. However, the absence of a more comprehensive dataset in the other countries within the Amazon basin increases the inaccuracy of the rainfall in these regions. 2.2.2. The GPCP and TMPA satellite-based datasets As part of the Global Energy and Water Cycle Experiment (GEWEX), the GPCP dataset was set up to develop monthly

A.C.V. Getirana et al. / Journal of Hydrology 404 (2011) 304322 Table 2 Precipitation datasets used in the hydrological modeling. Dataset HOP CPC GPCP TMPA ERA-40 NCEP-2
a b c

307

Reference Hybam Observatory Precipitation Climate Prediction Center Global Precipitation Climatology Project Tropical Rainfall Measuring Mission European Center for Medium Range Weather Forecast NCEP DOE AMIP-II

Format/Spatial Resolution Espinoza Villar et al. (2009) Silva et al. (2007) Adler et al. (2003) Huffman et al. (2007) Uppala et al. (2005) Kanamitsu et al. (2002)

Spatial coverage Catchmentsa Grid (1.0 1.0) Grid (1.0 1.0) Grid (0.25 0.25) Grid (1.125 1.125) Grid (1.875 1.875)

Series Span Amazon basin South America Global Quasi-global (180W180E, 50N50S) Global Global

Time-step 01/1980 12/2006 01/1978 12/2007 10/1997present 01/1998present 01/1957 08/2002 01/1979present

Datasource Dailyb Daily Daily 3hourly 6hourly 6hourly Rain-gauge Rain-gauge Satellite + raingauge Satellite + raingauge GCM estimatesc GCM estimatesc

Catchments with areas up to 600 km2. Espinoza Villar et al. (2009) present a monthly time step dataset while a daily version is used herein. These models were run by assimilating key variables.

precipitation data based on remotely sensed data from geostationary and polar-orbiting satellites, plus in situ observations. Several GPCP precipitation datasets are currently available. The Global Precipitation Climatology Project (GPCP) one-degree daily (1dd) dataset (Huffman et al., 2001) has been used in this study. This dataset is available for the 19972006 period and is based on multiple passive microwave, infrared satellite observations and gauge observations. The monthly total GPCP 1dd precipitation dataset matches the monthly values of GPCP version 2, which are based on satellite and are mostly inuenced by gauge observations over land areas (Nijssen et al., 2001). Although its resolution is coarser (1 1) than those of other satellite datasets, GPCP 1dd (further referred to as GPCP) has been selected for this study because of the extensive applications, analyses and validation studies carried out with its monthly and daily versions (Adler et al., 2003; Gebremichael et al., 2003; Voisin et al., 2008; Juarez et al., 2009), and because of substantial temporal overlap with the reanalysis and gauge-based datasets used in our analysis (Fig. 3). The Tropical Rainfall Measuring Mission (TRMM) Multisatellite Precipitation Analysis (TMPA) dataset (Huffman et al., 2007) has also been used in this study. TRMM is a joint mission, initiated in November 1997, between the National Space Development Agency (NASDA) of Japan and the National Aeronautics and Space Administration (NASA) of the United States. The mission was designed to

study tropical rainfall between 35N and 35S. The 3B42 version 6 dataset has been used in this analysis. It has a temporal resolution of 3 h, a spatial resolution of 0.25 0.25 and is available from 1998 to the present. This dataset has been largely used for hydrological studies (e.g. Arvor et al., 2008; Collischonn et al., 2008; Su et al., 2008). 2.2.3. The ERA-40 and NCEP-2 reanalysis-based datasets The ERA-40 dataset results from observations from various sources (land, ship, aircraft, satellite, among others) into a shortterm operational forecast model. Data are provided four times a day, on a 1.125 1.125 grid. The numerous analyses of this dataset in the literature (e.g. Betts et al., 2003a,b, 2005; Su et al., 2006; Fernandes et al., 2008; Voisin et al., 2008) motivated its use in this comparison in the Negro River basin. A second reanalysis precipitation dataset is the NCEP-2 (Kanamitsu et al., 2002), which is an update of the NCEPNational Center for Atmospheric Research (NCAR) reanalysis dataset (Kalnay et al., 1996). NCEP-2 is available from 1979 to the present and incorporates observed data and numerical weather forecast simulations. However, precipitation is mostly simulated. The NCEP-2 dataset is available on an approximately 1.9 1.9 resolution global Gaussian grid at 6-h time steps. As in the ERA-40 dataset, NCEP-2 is a model-based precipitation dataset frequently used for hydrological and climatologic studies (e.g. Roads, 2003; Fekete

4500 TMPA begins ERA-40 ends 2000 ERA-40

3500

P (mm)

2500

1500

500 1980 HOP

1985 CPC

1990 GPCP

1995 TMPA

2005 NCEP-2

Fig. 3. Mean annual precipitation over the Negro River basin between 1980 and 2006. The gure also shows the time periods when the six datasets were available. The lower annual precipitation of ERA-40 in 2002 is due to its end in August 2002.

308

A.C.V. Getirana et al. / Journal of Hydrology 404 (2011) 304322

et al., 2004; Karam and Bras, 2008). More complete descriptions of the datasets can be found in the literature and elsewhere in this paper. 2.3. Precipitation data intercomparison approach The precipitation intercomparison has been performed in two steps: rst, an analysis is carried out in terms of statistical coefcients computed at different time scales, and at both the basin and sub-basin scales; and second, a comparison focusing on an analysis of how the precipitation datasets impact the water balance computed by the MGB-IPH model. The rst step of the analysis is performed by using scatter plots in which the HOP dataset is considered to be the reference dataset. The root mean squared error normalized by the mean of the observed values (Nrmse) and the correlation coefcients (r) are considered in the statistical analysis. They are given by

events, rain event detection coefcients are calculated for precipitation thresholds of 0.5, 1, 5, 10, and 20 mm/day. The second step of the comparison focuses on an analysis of how the precipitation datasets impact the computation of the hydrological components (evapotranspiration, discharge and soil moisture) obtained with the MGB-IPH model. Special attention is given to discharge, which enables a quantitative evaluation based on observed data from ve gauge stations within the Negro River basin (Fig. 1). 2.3.1. The hydrological model The MGB-IPH model has been specially designed for studies of large basins with relatively scarce data, using globally available data as far as possible. The watershed is divided into elements of area (normally square grids or cells) interconnected by channels. In order to reduce the number of model parameters and of required input data, the grouped response unit (GRU) approach (Kouwen et al., 1993) is used to handle sub-grid variability. The MGB-IPH is composed of modules enabling the calculation of soil water balance, evapotranspiration, ow propagation within a cell, and ow routing through the drainage network. Model concepts were initially based on the LARSIM (Krysanova et al., 1998) and VIC (Wood et al., 1992; Liang et al., 1994) models, with some changes in the evapotranspiration, percolation and channel-routing modules. Land use, topography, vegetation cover and soil types are used in the model as inputs to conduct the selection of parameter values. The model has been applied to several studies of basins in South America (Tucci et al., 2005), especially in the Amazon River basin (Ribeiro Neto et al., 2005; Collischonn et al., 2008). Main aspects of those applications are described by Allasia et al. (2006) and a more detailed description of the model can be found in Collischonn et al. (2007) and Getirana et al. (2010). Uncertainty and scale aspects of the MGB-IPH model are not evaluated in this study. Details about these aspects can be found in Nijssen and Lettenmaier (2004) and Hong et al. (2007). Parameter values are associated with soil type, land use and vegetation cover and to the characteristics of relief, using the GRU approach. The soil water balance is computed independently for each GRU of each cell, considering only one soil layer (Fig. 4) using

s Pnt 2 1 t 1 xsimt xobst Nrmse  x obs nt r covxsim ; xobs rxsim rxobs

respectively, where xobs and xsim stand for the observed and simulated time series (in this case, x denes daily precipitation), respectively,  xobs is the mean value of the observed time series, covxsim ; xobs stands for the covariance between observed and simulated variables, and rxobs and rxsim are the standard deviations of xobs and xsim, respectively. t and nt are, respectively, the time index and the total number of time steps. In addition, the intercomparison is carried out by detecting rain events at different precipitation thresholds over the Negro River basin at a daily time step. It is performed by computing the frequency bias index (FBI), the probability of detection (POD), the false alarm ratio (FAR), and the equitable threat score (ETS). These coefcients, summarized in Table 3, are obtained from a 2 2 contingency table made up of four parameters (a, b, c and d), where a is the number of observed rain events correctly detected, b stands for the number of observed rain events not detected, c is the number of false alarms (rainfall events detected but not observed), and d is the sum of cases when neither observed nor detected rain events occurred. FBI indicates whether the dataset tends to underestimate or overestimate rain events, FAR and POD compute the fraction of false alarms and rain occurrences that were correctly detected, respectively, while ETS provides the fraction of rain events (observed and/or detected) which were correctly detected. To quantify the ability of each dataset in predicting light and heavy rainfall

  k k k 1 k k k Wk i;j W i;j P i ET i;j Dsupi;j Dinti;j Dbasi;j Dt

where k, i and j are indexes related to the time step, the cell and the GRU, respectively; dt is the model time step (1 day in most applications); W k i;j [mm] is the water storage in the soil layer at the end of

Table 3 Summary of the rain event detection coefcients. Coefcient FBI POD Full name Frequency bias index Probability of detection False alarm ratio Equitable threat score Equationa FBI a b=a c POD a=a c Range 01 01 Optimal score 1 1

FAR

FAR c=a c

01

ETS

ETS a He =a b c He

1 to 1

He a b a c=N where N is the total number of estimates.

Fig. 4. Schematic of the MGB-IPH soil water budget applied to each grouped response unit (GRU) of a cell (Collischonn et al., 2007).

A.C.V. Getirana et al. / Journal of Hydrology 404 (2011) 304322

1 the kth time step, of the jth GRU of the ith cell; W k [mm] is the i;j same variable at the end of the previous time step; Pk i;j is the precip1 itation that reaches the soil; ETk ] is the evapotranspirai;j [mm dt 1 k tion ux from the soil; Dsupi;j [mm dt ] is the surface runoff, or k quick ow; Dinti;j [mm dt1] is the subsurface ow; and Dbask i;j 1 [mm dt ] is the ow to the groundwater reservoir. The evapotranspiration is calculated based on soil water storage at the start of the time step, according to an explicit scheme in time. The calculation is performed in two steps. Initially, precipitation is assumed to be stored on the surface of the vegetation until a maximum interception storage capacity is reached and only the excess precipitation (throughfall) passes through the canopy to reach the soil surface. The interception storage capacity is determined for each GRU based on the leaf area index (LAI):

309

S max a LAI

where Smax is the maximum interception storage capacity for a GRU; and a is a model parameter assumed to be a constant value of 0.2 mm (Ubarana, 1996). LAI values have been considered constant in time for each GRU (Xue et al., 1991). Intercepted water is then evaporated from the interception storage computed by the PenmanMonteith equation, setting the surface resistance to zero. The remaining energy is used for evapotranspiration of the vegetated soil (soil evaporation plus plant transpiration) which is calculated according to the PenmanMonteith equation weighted by the remaining evaporative demand, as proposed by Wigmosta et al. (1994):

ET fDE

D A qA cp rD a rs D c 1 r a

M k qW

where D [kPa C1] is the gradient of the saturated vapor pressure temperature function, A [MJ m2 s1] the available energy; qA [kg m3] the specic mass of air; qW [kg m3] the specic mass of water; cp [MJ kg1 C1] the specic heat of moist air; D [kPa] the vapor pressure decit; c [kPa C1] the psychrometric constant; rs [s m1] the surface resistance of the land cover; ra [s m1] the aerodynamic resistance; k [MJ kg1] the latent heat of vaporization; and M a constant for unit conversion between m s1 and mm dt1. fDE is given as follows:

subsurface ow and groundwater ow) were manually calibrated and validated using a daily time step from 1997 to 2006 by comparing model results with observed data at 22 gauge stations. The validation procedure was complemented with water levels derived from the ENVISAT radar altimetry satellite from 27 virtual stations (as described in Getirana et al., 2010; 2009c). Another study compared the model outputs with Gravity Recovery and Climate Experiment (GRACE)-derived total water storage (TWS) data from seven sub-basins dened by gauge stations at Caracara, Taraqua, Cucu, Curicuriari, Serrinha, Barcelos and Manaus (Getirana et al., 2009d). The simulated water level in each computational cell was derived from a discharge height relationship generalized to most of the basin, thus permitting a validation of the model against 35-day repeated cycles of ENVISAT data. Monthly TWS values were t derived from the relation dTWS P t Rt ETt where P dt stands for the observed monthly precipitation used to force the model, R and ET stand for monthly values of runoff and evapotranspiration, respectively, provided by the model, and t is the time step. The simulated discharges and evapotranspiration rates both showed satisfactory agreement with observed and satellite-derived data. NashSutcliffe coefcients of modeled discharges varied from 0.47 to 0.94, and correlation coefcients, r, of water heights and altimetric data averaged 0.86 for the entire basin. Correlation coefcients between simulated and GRACE-derived TWS varied from 0.64 to 0.84, depending upon the sub-basin involved. These parameters have been used in this study to simulate the hydrology of the Negro River basin from 1998 to 2002 for the different precipitation datasets. Since MGB-IPH computes vertical water and energy balances using a daily time step, the 3-hourly (TMPA) and 6-hourly (ERA-40 and NCEP-2) datasets have been cumulated in a 24-h time step to be in accordance with the model requirements. The model has been evaluated using the correlation coefcient (r), the NashSutcliffe efciency coefcient (NS) and the relative error (RE) of predicted streamows. The correlation coefcient r is dened by Eq. (2). The NS and RE coefcients are given by

Pnt 2 t 1 Q obs t Q sim t NS 1 P nt 2 t 1 Q obs t Q obs RE%  nt  Q obs t Q sim t  1 X   100  nt t1  Q obs t

fDE

EIP EI EIP

where EI and EIP are, respectively, the evaporation and potential evaporation from the interception storage. Evaporation ux of open-water bodies is calculated considering fDE = 1. Available energy and aerodynamic resistance can be calculated following Shuttleworth (1993). Please refer to Collischonn et al. (2007) for more details about the calculation of the water balance and evapotranspiration. 2.3.2. Modeling setup The Negro River basin has been divided into 1756 catchments (computational cells) having areas of up to 600 km2. Atmospheric forcings (solar radiation, pressure, relative humidity, wind speed and temperature) were provided by the NCEPDOE/AMIP-II reanalysis dataset (Kanamitsu et al., 2002) and passed through the inverse square distance interpolation approach to spatially distribute the data throughout the basin. Land cover and soil types were derived from the JERS-1 image classication (Martinez and Le Toan, 2007) and FAO (FAO, 1995) datasets, respectively. Flow directions and watershed delineations were acquired from a 200m SRTM DEM product after a pre-processing phase as suggested by Getirana et al. (2009a,b). Using the HOP dataset to force the model, parameters (mostly soil hydraulic properties that control water storage, inltration,

where Qobs and Qsim are the observed and simulated daily water discharges, respectively and Q obs is the average observed discharge. 3. Results and discussion 3.1. Precipitation analysis 3.1.1. Spatial and temporal distribution of the annual precipitation As mentioned before, the duration of the study period (1998 2002) was dened by the available precipitation data. TMPA data are available since January 1998, while the ERA-40 dataset ends in August 2002, resulting in a nearly 5-year period (Fig. 3). The mean annual precipitation during the study period, averaged over the Negro River basin, varies signicantly among the datasets, from 2219 (TMPA) to 3065 mm/year (NCEP-2), while HOP presented 2715 mm/year. This means that TMPA and NCEP-2 provide the lowest and highest averaged precipitation estimates, resulting in relative errors of, respectively, 18% and +13% as compared to the reference dataset (2715 mm/year). The overestimated precipitation provided by NCEP-2 may be explained by two consecutive years (1999 and 2000) during which this dataset gives a

310

A.C.V. Getirana et al. / Journal of Hydrology 404 (2011) 304322

particularly exaggerated rainfall, especially 1999 (Fig. 3). However, when all previous years (19801998) are considered, NCEP-2 actually shows much lower precipitation rates (2070 mm/year) as compared to the reference dataset (2530 mm/year) for the same period. ERA-40 underestimates precipitation over the Negro River basin by about 18% during the period from January 1998 to August 2002. This result is similar to those found in previous studies with ERA40 which showed that precipitation over the Amazon basin had a bias of 10% (Fernandes et al., 2008) to 26% (Voisin et al., 2008) compared to gauged-based datasets. Other datasets such as CPC and GPCP give lower errors (7% and 8%, respectively). Fig. 5 presents the spatial distribution of the mean annual precipitation amounts derived from the six datasets. Overall, good agreement exists among the datasets if judged in terms of relative values within the basin. All datasets show lower precipitation rates in the extreme Northeast of the basin than on the western side. However, absolute values vary considerably from one dataset to another. For instance, the northeastnorthwest rainfall gradient observed in the Caracara sub-basin (northeastern part of the Negro River basin), as provided by HOP and CPC, is not reproduced by the other datasets. Visual inspection of the results reveals that somewhat anomalous distributions are given by the NCEP-2 reanalysis-based dataset at the basin-wide scale. The latter one could not properly represent the precipitation gradient that exists between the western and northeastern regions. To improve the intercomparison of the spatial distribution of precipitation elds, four additional rain-gauge datasets have been included in the analysis. Three of these are worldwide datasets: the Climate Research Unit (CRU) dataset (New et al., 2000), the WillmottMatsura (WM) dataset (Willmott and Matsuura, 2009) and the Global Precipitation Climate Center (GPCC) dataset (Rudolf et al., 1994). The fourth one (ANA) is very similar to the HOP dataset, except that it only contains data made available by the Brazilian Water Agency ANA. Precipitation elds obtained with the

raingauge-based datasets show that having less rainfall data generally leads to lower precipitation rates over the western part of the Negro River basin, as shown by the CPC and WM datasets. This is probably due to the use of data from Colombian and Venezuelan gauges that are not located within the Amazon basin but in the less rainy Andes (as shown in Silva et al., 2007). Consequently, they are not representative of the hydrological regime of the basin. The precipitation elds of the Amazon basin presented by Costa and Foley (1998) also appear to have the same limitation. On the other hand, Espinoza Villar et al. (2009) showed that some locations in the Northwest of the Amazon basin are subject to intense precipitation with rates up to 6000 mm/year. Underestimated precipitation rates can signicantly impact the local water balance as described in the hydrological modeling section. The role of the gauge density on the construction of precipitation elds and discharges provided by hydrological models is also addressed by Gebremichael and Krajewski (2004). 3.1.2. Daily precipitation analysis Fig. 6 presents scatter plots between daily Negro River basinaveraged precipitation rates in HOP and in the different datasets. The performance of datasets varies signicantly. The best estimates are obtained with the CPC gauged-based dataset, giving a high coefcient of correlation (r = 0.95) and a low root mean square (Nrmse = 0.24) and relative (RE = 22%) errors. However, CPC tends to underestimate precipitation, with most points situated to the left of the diagonal. The satellite-based GPCP and TMPA datasets also perform relatively well, albeit less than CPC. GPCP performs somewhat better (r = 0.61, Nrmse = 0.67 and RE = 57%) than TMPA (r = 0.52, Nrmse = 0.79 and RE = 59%). In contrast to the results acquired for other large basins in South America, such as the La Plata basin (Su et al., 2008) and the Tapajos River basin (Collischonn et al., 2008), TMPA reveals a slight tendency to underestimate intense rain events (>20 mm/day) in the Negro River basin. ERA-40

Fig. 5. Spatial distribution of mean annual precipitation for the 19982002 period derived from the six datasets considered in this intercomparison (HOP, CPC, GPCP, TMPA, ERA-40 and NCEP-2), and also from ANA, CRU, WM and GPCC datasets. Pmin, Pmax and Pmed represent, respectively, the minimum, maximum and average annual precipitation over the Negro River basin. In order to compensate the early interruption of the ERA-40 time series (August 2002), monthly mean precipitation rates were computed from January to December and then summed, resulting in the mean annual precipitation rate.

A.C.V. Getirana et al. / Journal of Hydrology 404 (2011) 304322

311

60 50

60

Gauged (mm/d)

40 30 20 10 0 0

r=0.52 Nrmse=0.79 RE=59%

50 40 30 20 10 0

r=0.54 Nrmse=0.70 RE=65%

10

20

30

40

50

60

10

20

30

40

50

60

TMPA (mm/d)
60 50
60

ERA-40 (mm/d)
60

Gauged (mm/d)

40 30 20 10 0 0

r=0.95 Nrmse=0.24 RE=22%

50 40 30 20 10 0

r=0.61 Nrmse=0.67 RE=57%

50 40 30 20 10 0

r=0.31 Nrmse=0.91 RE=115%

10

20

30

40

50

60

10

20

30

40

50

60

10

20

30

40

50

60

CPC (mm/d)

GPCP (mm/d)

NCEP-2 (mm/d)

Fig. 6. Scatter plots of daily basin wide-averaged precipitation from the ve datasets (CPC, TMPA, GPCP, ERA-40 and NCEP-2) versus the gauged estimates (HOP) over the Negro River basin for the January 1998August 2006 period.

statistics are comparable to those provided by the satellite-based datasets (r = 0.54 Nrmse = 0.70 and RE = 65%), but they show a tendency to underestimate high precipitation rates (>15 mm/day). Similar results were obtained by Fernandes et al. (2008), who noted that ERA-40 closely reproduces rainfall during the dry and transitional seasons, but it underestimates the wet season rainfall. The worst results are obtained with NCEP-2 (r = 0.31, Nrmse = 0.91 and RE = 115%). The elevated RE value of that dataset is due to the large rainfall overestimation in both the wet and the dry seasons. In order to further elucidate differences between the datasets, a rain event detection analysis over the Negro River basin has also been performed. Fig. 7 shows the results for the frequency bias index (FBI), the false alarm ratio (FAR), the probability of detection (POD), and the equitable threat score (ETS) for each of the ve precipitation datasets for the 19982006 period. Results vary substantially among datasets. The various datasets cannot detect the same frequency of extreme rain events (<0.5 mm and >20 mm) as the reference dataset. For a threshold precipitation <0.5 mm, FBI values are either less than 0.45 (0.44 for CPC and 0.06 for ERA) or higher (2.75 for NCEP-2, 3.81 for TMPA and 5.50 for GPCP). These numbers indicate that CPC and ERA-40 tend to underestimate the number of no rain or very low rain events, whereas NCEP-2, TMPA and GPCP overestimate the number of dry days. Very similar results are obtained for the 0.51.0 mm threshold. These results are probably inuenced by the homogeneous distribution of the monthly precipitation over the days of the months, as performed in the HOP dataset for the Columbian and Venezuelan stations of the Rio Negro basin. This can slightly increase the precipitation rates of the reference dataset in dry events. However, no rain or very low rain events are rare in the basin, corresponding to $2% of all events analyzed, as dened by the sum of coefcients a and b for the respective threshold (Fig. 7). In addition, the number of stations with monthly data (14) is considerably lower than the total num-

ber of stations in the Rio Negro basin (1 1 5). This causes a more signicant local impact than a basin-wide impact. FBI values of 1 0.25 have been obtained for all of the datasets for rain events from 1.0 to 10 mm. All of the datasets tend to underestimate the number of rain events between 10 and 20 mm/day, producing FBI values varying from 0.46 (ERA-40) to 0.89 (NCEP-2). The number of severe rain events (threshold > 20 mm/day) is underestimated by CPC and ERA-40, which have FBI values of 0.36 and 0.41, respectively, and overestimated by NCEP-2 (6.0), TMPA (2.32) and GPCP (1.82). However, according to the HOP dataset, severe rain events represent less than 1% of the total number of events. The CPC dataset produces optimal FAR values for extreme rain events (thresholds <0.5 mm and >20 mm). This means that the dataset does not tend to falsely determine no rain or extreme rainfall events. The gauge-based dataset provides higher FAR values for rain events between 0.5 and 20 mm, with the coefcient varying from 0.11 (threshold 1020 mm) to 0.28 (threshold 0.5 1.0 mm). The satellite and model-based datasets are unable to give such favorable ratios for false alarms. The TMPA, GPCP, ERA-40 and NCEP-2 datasets have a tendency to announce false rain events with thresholds up to 1.0 mm, with FAR values varying from 0.77 to 1.0. FAR values for these same datasets approximately decrease (and hence their performances increase) for thresholds between 1.0 mm and 20 mm, reaching values from 0.41 to 0.58, but then increase again for thresholds >20 mm. Among the non-gauge based datasets, ERA-40 shows the best FAR values, followed by TMPA, GPCP and NCEP-2. GPCP and NCEP-2 provide quite similar tendencies to announce false rain events. Like FBI, rain occurrence is best detected for rain events between 1 and 10 mm. The CPC again produces the best results, with POD values being larger than 0.70 for thresholds 15 mm, 5 10 mm and 1020 mm. Values decrease rapidly (POD < 0.45) for

312

A.C.V. Getirana et al. / Journal of Hydrology 404 (2011) 304322

Threshold (mm/day)

<0.5 0.5-1

5-10 10-20 >20

a
<0.5

Threshold (mm/day)

0.5-1 1-5 5-10

CPC TMPA

10-20 >20 0 200 400 600 0

200 400 600 0 200 400 600 0 600 1200 1800

Fig. 7. Coefcients frequency bias index (FBI), false alarm ratio (FAR), probability of detection (POD) and equitable threat score (ETS), and a, b, c and d for six daily precipitation thresholds over the Negro River basin for the CPC, TMPA, GPCP, ERA-40 and NCEP-2 datasets. The HOP dataset is considered to be the reference precipitation. a, b, c and d represent the number of observed rain events correctly detected, the number of observed rain events not detected, the number of false alarms (rainfall events detected but not observed), and the sum of cases when neither observed nor detected rain events occurred, respectively. FBI, FAR, POD and ETS are described in Table 3.

thresholds of <1 mm and >20 mm. Among the non-gauge-based datasets, TMPA performs better in detecting extreme rain events (POD = 0.5 for threshold <0.5 mm and POD = 0.23 for threshold >20 mm), whereas ERA-40 shows the best results overall. The ability to detect rain events is also evaluated in terms of the equitable threat score (ETS). CPC once again is notably superior in detecting observed rain events, which is easily explained by its rain-gauge basis. ETS values vary from 0.26 to 0.61. The best results for detection is for thresholds between 1 and 20 mm. Rain events of less than 1 mm and above 20 mm show lower ETS values, but are still quite acceptable. Among the non-gauge-based datasets, TMPA shows the best overall ability to detect rainfall events, having higher ETS values for most thresholds than the other datasets, except for >20 mm for which ERA-40 produces the best results (ETS = 0.10). All datasets provide some skill in detecting rain (ETS > 0) for the different thresholds, except ERA-40 and NCEP-2. Both datasets are unable to detect any observed rain (ETS = 0) with thresholds <0.5 mm and >20 mm, respectively. 3.1.3. Monthly precipitation analysis Fig. 8 shows monthly precipitation estimates for the six sub-basins after subtracting a constant value from the whole series, and plots of the correlation coefcients, r, between the different datasets. The monthly precipitation time series have been shifted using their respective mean annual averages, thus showing the seasonal variability of the precipitation dynamics. Most datasets properly represent the annual seasonality in all sub-basins. However, the NCEP-2 dataset is unable to satisfactory represent both the annual cycle and the absolute precipitation values. Correlation coefcients have also been calculated to see how the various datasets relate to each other. Based on these values and considering only the non-gauged based datasets, the best and worst results are obtained with the GPCP and NCEP-2 datasets, respectively. Values for r vary considerably from one sub-basin to

another. The best agreement among the datasets is found for the Caracara sub-basin (as dened by the Caracara station), for which r values between two datasets vary from 0.70 (between TMPA and ERA-40, and TMPA and NCEP-2) to 0.99 (between HOP and CPC). TMPA shows the lowest correlation with the HOP dataset (r = 0.78). NCEP-2 and HOP also are not much correlated (r = 0.82), while the other three datasets (produce) display correlation coefcients higher than 0.90. The mean correlation coefcients ( r) between a dataset and the other ve are listed in Table 4. The HOP and CPC precipitation datasets show the best correlations for the Caracara sub-basin (both with  r 0:90) while TMPA (provides) presents the worst correlations ( r 0:75) for this sub-basin. Correlation coefcients between datasets are lower for the other sub-basins. The lowest correlation coefcients (both for r and  r) are obtained for the Taraqua sub-basin (Fig. 8, see the small size of the graphic). Values for r vary between zero (for ERA-40 and NCEP-2), indicating no correlation at all, and 0.82 (HOP and CPC). TMPA provides overall better correlations (e.g.  r 0:35 for Taraqua and  r 0:66 for Manaus) than NCEP-2 (0.26 and 0.61, respectively). The contrasting results obtained for the Caracara and Taraqua sub-basins suggest that most datasets are highly dependent on gauge observations. Datasets are better correlated in basins where in situ data are available (e.g. Caracara basin), and less correlated in regions where data are unavailable or inexistent (e.g. Taraqua basin). This means that the quality of satellite-based datasets is intrinsic to the gauge network density. A low correlation between rainfall datasets found in the Taraqua sub-basin is also noticed in other nested sub-basins (Curicuriari and Serrinha), as seen in Table 4. For example, mean correlation coefcients vary from 0.46 and 0.42 (both obtained for NCEP-2) to 0.70 and 0.71 for Curicuriari and Serrinha, respectively. Still, correlations between monthly precipitation time series averaged for the entire Negro River basin (represented by the Manaus

GPCP

0.5

1.5

0.2 0.4 0.6 0.8

0.2 0.4 0.6 0.8

0.2

0.4

0.6

0.8

NCEP-2

1-5

ERA-40

Frequency biase index (FBI)

False alarm ratio (FAR)

Probability of detection (POD)

Equitable threat score (ETS)

A.C.V. Getirana et al. / Journal of Hydrology 404 (2011) 304322

313

Caracarai
500
HOP 1.00 0.75 NCEP-2 0.50 0.25 0.00 CPC

ERA-40

GPCP

-500

Taraqua
500

TMPA

HOP 1.00 0.75 NCEP-2 0.50 0.25 CPC

0.00

ERA-40

GPCP

-500
TMPA

Cucui
500

HOP 1.00 0.75 NCEP-2 0.50 0.25 CPC

0.00

ERA-40

GPCP

-500 1/98

1/99

1/00

1/01

1/02

TMPA

HOP

CPC

GPCP

TMPA

ERA-40

NCEP-2

Fig. 8. Monthly precipitation estimates (mm/month) for the 19982006 period after subtracting a constant (monthly mean) value from each series (left) and plots of the correlation coefcient between each of the datasets and the other ve ones at the monthly time step (right). The correlation of a precipitation dataset with itself (which would be 1.0) is not shown in the correlation plots.

station) exhibit overall good similarity between the HOP and CPC datasets (both with  r 0:82). GPCP also provides competitive correlations, with a mean correlation coefcient of  r 0:80. By comparison, ERA-40 ( r 0:73) gives a better mean correlation value than TMPA ( r 0:66) and NCEP-2 ( r 0:61). 3.2. Model outputs A model sensitivity analysis was performed considering three hydrological components: evapotranspiration (ET), runoff (R) and soil moisture content (W). Fig. 9 shows mean daily values of the hydrological components for the 19982002 period averaged for the six sub-basins within the Negro River basin. The hydrological model forced by HOP results in moderate ET spatial heterogeneity among sub-basins, varying from 3.03 mm/day in the Caracara basin to 4.30 mm/day in the Taraqua basin, averaging 3.28 mm/day for the entire basin. Mean ET estimates are in good agreement with

previous studies in the region. For example, Shuttleworth (1988) found a value of 3.76 mm/day for the Reserva Ducke site, Malhi et al. (2002) obtained 3.08 mm/day at Reserva Biolgica do Cuieiras, and Tomasella et al. (2007) estimated ET to be 3.86 mm/day for the Asu catchment. Runoff values (model forced by HOP) vary from 2.52 mm/day (Caracara basin) to 6.64 mm/day (Curicuriari basin), averaging 4.15 mm/day for the entire basin. Mean ET values averaged over the entire Negro River basin (as represented by the Manaus station) vary from 3.13 (TMPA) to 3.44 mm/day (ERA-40), with an average of 3.29 mm/day for all datasets. Larger differences are found in drier areas (Caracara), where ET varies between 2.83 mm/day (TMPA) and 3.40 mm/day (NCEP-2), and averaging 3.15 mm/day. This corresponds to differences of up to 17% among the datasets, which is less than the differences found for precipitation among the same datasets (33%). More humid locations such as Taraqua present ET values between 4.03 (TMPA) and 4.27 mm/day (HOP), with an average of

314

A.C.V. Getirana et al. / Journal of Hydrology 404 (2011) 304322

Curicuriari
500

HOP 1.00 0.75 NCEP-2 0.50 0.25 CPC

0.00

ERA-40

GPCP

-500
TMPA

Serrinha
500

HOP 1.00 0.75 NCEP-2 0.50 0.25 CPC

0.00

ERA-40

GPCP

-500
TMPA

Manaus
500

HOP 1.00 0.75 NCEP-2 0.50 0.25 CPC

0.00

ERA-40

GPCP

-500 1/98

1/99

1/00

1/01

1/02

TMPA

HOP

CPC

GPCP

TMPA

ERA-40
Fig. 8 (continued)

NCEP-2

4.14 mm/day. This represents differences of less than 3% among datasets. Small differences in ET were also found by Voisin et al. (2008) for the Amazon basin using the VIC model. Despite the high spatial heterogeneity of the precipitation elds, as shown in Fig. 3, mean ET values derived from the different datasets rates are very similar. One can notice that a remarkable

Table 4 Mean correlation values ( r ) of the monthly precipitation time series. Simple correlations between each monthly rainfall data set and all the others are averaged in order to obtain the mean correlation between rainfall les, for each dataset and each sub-basin within the Negro River basin. HOP Caracara Taraqua Cucu Curicuriari Serrinha Manaus 0.90 0.54 0.75 0.70 0.69 0.82 CPC 0.90 0.57 0.75 0.71 0.70 0.82 GPCP 0.88 0.53 0.73 0.65 0.65 0.80 TMPA 0.75 0.35 0.57 0.50 0.49 0.66 ERA-40 0.84 0.24 0.63 0.52 0.52 0.73 NCEP-2 0.79 0.26 0.60 0.46 0.42 0.61

gradient of evapotranspiration values is obtained with the six datasets, as shown in Fig. 10. Mean ET values are lower in the eastern part of the basin, but increases signicantly in the western part. Differences between extreme values vary from 73% for CPC (ETmin = 973 mm/year and ETmax = 1681 mm/year) and NCEP-2 (ETmin = 995 mm/year and ETmax = 1654 mm/year) to 123% for HOP (ETmin = 771 mm/year and ETmax = 1718 mm/year). Also note that in all of the datasets the presence of a lower energy demand in the central part of the basin. The agreement of ET spatial distribution derived from the six datasets is an indication that ET is not water limited, which means that water availability (as provided by the precipitation datasets) has a reduced effect on evapotranspiration. This is demonstrated by the fact that expressive differences between rainfall datasets in the western (and more humid) side of the basin result in small variations in ET (see the monthly time series of the precipitation averaged for the Taraqua sub-basin in Fig. 7 where ET derived from the six rainfall datasets has insignicant differences). The exception is the driest region located in the north-eastern part of the basin where ET has more accentuated differences among datasets. This is probably due to the fact that ET

A.C.V. Getirana et al. / Journal of Hydrology 404 (2011) 304322

315

12 9 6 3 0 HOP CPC

Caracarai

600 400 200 0

12 9 6 3 0 HOP CPC

Taraqua

600 400 200 0

GPCP

TMPA ERA-40 NCEP-2

GPCP

TMPA ERA-40 NCEP-2

12

Cucui

600 400

12 9 6

Curicuriari

600

R,ET (mm/day)

9 6 3 0 HOP CPC GPCP TMPA ERA-40 NCEP-2

W (mm)

400 200 0

200 0

3 0 HOP CPC GPCP TMPA ERA-40 NCEP-2

12 9 6 3 0 HOP CPC

Serrinha

600 400 200 0

12 9 6 3 0 HOP CPC

Manaus

600 400 200 0

GPCP

TMPA ERA-40 NCEP-2

GPCP

TMPA ERA-40 NCEP-2

Evapotranspiration

Runoff

Soil Moisture

Fig. 9. Mean daily evapotranspiration (ET), runoff (R) and soil moisture (W) at the selected stations for the 19982002 period.

values in drier basins are more sensitive to precipitation than in wetter basins. According to the monthly time series of the evapotranspiration averaged for the Caracara basin, more signicant differences of ET occur during the dry seasons. Note also that dry seasons are practically not detected by the evapotranspiration derived from NCEP-2 due to the overestimated precipitation during these periods. Basin-wide precipitation derived from the HOP dataset is only 8% higher than that obtained with CPC. However, as ET values are about the same, differences between runoff values derived from the same datasets increase to 26%. The differences are more signicant in the Taraqua sub-basin where precipitation and runoff rates between HOP and CPC differ by 22% and 125%, respectively. In general, the averaged basin-wide runoff varies from 2.70 mm/day (ERA-4) to 4.83 mm/day (NCEP-2). Runoff values derived from the satellite-based datasets are underestimated throughout the basin, although GPCP presents overall better results since values are relatively close to the reference dataset. For example, GPCP and TMPA provide 3.40 and 2.82 mm/day, respectively, for the Negro River basin scale. The spatial distribution (discretized by sub-basins) of the soil moisture content W shows a high sensitivity to the precipitation datasets, but the basin-wide values are relatively constant among the datasets, with variations of up to 10% between 301 mm (ERA-40) to 334 mm (NCEP-2) averaged over the study period. The simulation with the reference precipitation (HOP dataset) provides W = 333 mm, while the mean W value for all datasets is 320 mm. Extreme W values are found for the dry Caracara and the wet Taraqua sub-basins, where variations among the datasets are also signicant. The mean soil moisture content of the Cara-

cara sub-basin varies between 168 mm (TMPA) and 248 mm (NCEP-2), while HOP produces a value of 210 mm. NCEP-2 also provides the highest mean soil moisture content for the wet Taraqua sub-basin (523 mm) while the lowest value is given by CPC (360 mm). HOP has a mean W of 465 mm. The above results show that disparities among the precipitation datasets of the western Negro River basin are reected in similar disparities among the model output variables, notably runoff and soil moisture content. For this reason, modeled and observed discharges are compared at ve gauge stations located within the basin. Fig. 11 shows the mean annual cycles of P, ET, R and W averaged over the Negro River basin. The mean precipitation peaks occur in May while the ET and W peaks arise 1 month later. Evapotranspiration rates are the highest between July and September, with peaks occurring in July. The exception is NCEP-2, which shows its precipitation peak 1 month sooner. This dataset also overestimates the dry period from November to March. Mean R and W values, however, show the same temporal variability as the other variables. Lower precipitation peak values lead to lower amplitude in R and W peaks. The ERA-40 dataset shows the lowest precipitation peak among all datasets, but is able to properly capture the hydrological seasonality of the Negro River basin. 3.3. Water discharge analysis Fig. 12a and b presents the daily hydrographs derived from the MGB-IPH model forced with the six precipitation datasets at ve gauge stations (Caracara, Taraqua, Cucu, Curicuriari and Serrinha stations). Modeled discharges show a general agreement with

316

A.C.V. Getirana et al. / Journal of Hydrology 404 (2011) 304322

6 4 2 0 6

ET(mm/day)

Caracarai
HOP

CPC

Taraqua

GPCP

4
TMPA

2 5
4 3 2 1/98 1/99

Manaus

ERA-40

NCEP-2

1/00

1/01

1/02

Fig. 10. Top: spatial distribution of mean annual evapotranspiration for the 19982002 period derived from the MGB-IPH model forced with the six datasets considered in this intercomparison (HOP, CPC, GPCP, TMPA, ERA-40 and NCEP-2). ETmin, ETmax and ETmed represent, respectively, the minimum, maximum and average annual evapotranspiration over the Negro River basin. Bottom: monthly evapotranspiration time series (in mm/day) averaged for three sub-basins (Caracara, Taraqua and Manaus).

gauged data when forced with HOP. CPC leads to good results in the Caracara basin but not in the western Negro basin because of the lack of adequate Colombian and Venezuelan rainfall. Qualitatively, most datasets (TMPA, GPCP and ERA-40) considerably underestimate discharges while NCEP-2 overestimates them. Except for NCEP-2, all datasets give positive NashSutcliffe (NS) coefcients (Table 5) at the Caracara station. Positive NS values indicate that the modeled discharges are able to represent gauged data more accurately than the mean observed discharge. HOP, CPC and GPCP provide low relative errors (up to 17.2% Table 6) at the Caracara station, whereas TMPA, ERA-40 and NCEP-2 produce higher relative errors. Correlation values r (Table 7), vary between 0.78 (ERA-40) and 0.95 (CPC). The relatively high r values are due to the same seasonality presented in all of the datasets. CPC shows the best NS and RE coefcients at the Caracara station, exceeding those values provided by the HOP dataset. On the other hand, HOP is the only dataset to show a positive NS coefcient (0.69) at the Taraqua station, while the other precipitation datasets have negative values. This means that the observed mean is better than model outputs. The best relative error at Taraqua is given by HOP (4.8%), followed by NCEP-2 (8.6%). The other datasets result in poorer RE coefcients (less than 44%). Discharges provided by the MGB-IPH model forced with HOP performed well, with NS coefcients varying from 0.80 to 0.86 and RE values between 2.5% (Serrinha) and 13.7% (Curicuriari). Following HOP, GPCP generally produces better performance coefcients for the simu-

lated discharges as reected by the NS values of 0.65 at Cucu and 0.42 at Curicuriari. The GPCP relative error for streamows at the Serrinha station was 27.7%. CPC also provides good results, followed by TMPA and ERA-40. NCEP-2 has the worst results. Fig. 13 shows the mean monthly observed and simulated discharges at ve gauge stations. In general, NCEP-2 overestimates discharges all over the study period at all stations and gives a weaker annual amplitude than observed. Other datasets provide fairly consistent discharges during the wet seasons, but peaks are mostly underestimated. At Caracara, NCEP-2 substantially overestimates observations from July to December, while also being too high during the dry season. HOP and CPC at this station both produce the best discharge estimates. The NCEP-2 also produces overestimates of the discharge at the Cucu station. Similarly to HOP, GPCP provides the closest mean monthly discharge time series at that station. Because of their upstream location, the runoff computed for the Taraqua and Cucu sub-basins affects the mean monthly discharges at Curicuriari and Serrinha stations. Most datasets, except for HOP and GPCP, signicantly underestimate observed discharges in the wet season at these four locations.

4. Conclusions This study focuses on two main objectives: the evaluation of the ability of six datasets to reproduce rainfall in the Rio Negro basin

A.C.V. Getirana et al. / Journal of Hydrology 404 (2011) 304322

317

400 300 200

HOP

TMPA

ERA-40

P,Q,ET,W (mm)

100 0
400 300 200 100 0 J F M A M J J A S O N D
Precipitation (P)

CPC

GPCP

NCEP-2

J F M A M J J A S O N D
Runoff (R)

J F M A M J J A S O N D
Soil Moisture (W)

Evapotranspiration (ET)

Fig. 11. Monthly mean precipitation (P), evapotranspiration (ET), runoff (R) and soil moisture content (W) averaged over the Negro River basin for the 19982002 period as derived from the hydrological modeling forced with the different precipitation databases.

during the 19982002 period, as given by the Hybam Observatory Precipitation dataset (HOP); and the simulation of the water balance by forcing the MGB-IPH hydrological model. The HOP dataset has been chosen as a reference as it is gauge-based and includes a large number of stations. The other ve rainfall datasets are gaugebased (CPC), satellite-based (GPCP and TMPA) and reanalysisbased (ERA40 and NCEP-2). The rainfall datasets have been compared at the annual, monthly and daily temporal scales and also at the basin-wide and catchment scales. Mean annual rainfall over the basin varied substantially from a dataset to another, and, in particular ERA-40 and satellite products had the largest underestimations. The EastWest gradient of annual rainfall was not correctly reproduced by some datasets. For instance, the CPC datasets underestimated rainfall in the western most part of the basin because rainfall interpolation was performed using data from relatively dry Colombian stations that are located in the Andes, outside of the Amazon basin (Silva et al., 2007). This has been conrmed by the comparison of monthly rainfall in six sub-basins of the Rio Negro: the best agreement between datasets has been found in the north-eastern Caracara basin, while more signicant differences have been observed in the western Taraqua basin where comprehensive observed data are not found in all of the datasets. This explains why CPC achieves better scores in this region when compared to HOP. Moreover, the analysis of daily events through various indicators [frequency bias index (FBI), probability of detection (POD), false alarm ratio (FAR) and equitable threat score (ETS)] has conrmed the notable superiority of the CPC dataset which is explained by its rain gauge basis. Among the non-gauge-based datasets, TMPA has provided the best results for most rainfall thresholds. Finally, the reanalysis datasets exhibited substantial deciencies in representing the temporal and spatial distribution of rain events in the Rio Negro basin resulting in overestimated rainfall (as shown in Figs. 5 and 6) and inconsistent rain events (as shown in Figs. 7 and 8) over the entire the basin. In conclusion, limited data availability and the use of unrepresentative rain-gauge stations located outside the Amazon basin in the interpolation process are the most limiting factors for

constructing precipitation datasets over poorly gauged and ungauged regions. This problem has been reduced considerably by using the HOP dataset as it includes stations from the north-western and western parts of the Negro River basin that are missing in other datasets. For that reason, the reference parameter set used to run the MGB-IPH model has been chosen as the one obtained by calibrating the model with the HOP dataset. This procedure certainly biased the water balance towards the HOP dataset, but it is argued that this rainfall dataset is likely the most complete dataset comprised of rain gauges which has ever been used over this region. The analysis of the water balance simulated by the MGB-IPH model forced by the different data sets showed that the mean evapotranspiration (ET) is in good agreement among the datasets. The similarity among the mean and the spatially distributed ET derived from the different datasets reveals an energy limited evapotranspiration in most parts of the basin. The exception is the Caracara basin which is the driest region of the study area. In term of spatial distribution, soil moisture content and runoff are more sensible to rainfall datasets. While soil moisture content tends to be relatively constant among the datasets, though it varies in space, water discharge varies considerably in space and among datasets. The HOP and CPC gauge-based datasets provided the most consistent runoff at the six gauge stations considered in the study. They best reproduced the mean values, the annual cycle and interannual variability. This has been shown for the Caracara basin where all datasets generally behave reasonably well. CPC is not as good as HOP in the western part of the Rio Negro basin, where most of the datasets provided bad results. The GPCP satellite-based dataset has provided good results despite its relatively coarse resolution. However, as expected considering the rainfall biases found in the rst part of the analysis, TMPA and ERA-40 underestimate water discharges. Dening the best precipitation dataset to be used in general situations is a difcult or even impossible task. Assessments must be performed to identify the best datasets for specic regions. This intercomparison identied limitations of datasets that incorpo-

318

A.C.V. Getirana et al. / Journal of Hydrology 404 (2011) 304322

Q (m3/s)
15,000

Caracarai
10,000

5,000

0 7,500

Taraqua

5,000

2,500

0 15,000

Cucui
10,000

5,000

0 30,000

Curicuriari

20,000

10,000

0 39,000

Serrinha
26,000

13,000

0 1-1-98

1-1-99
Obs

1-1-00
HOP

1-1-01
TMPA ERA-40

1-1-02

Fig. 12a. Comparison of observed and simulated discharge values obtained with the MGB-IPH model using a daily time step and forced by the HOP, TMPA and ERA-40 precipitation datasets at ve gauge stations (Caracara, Taraqua, Cucu, Curicuriari and Serrinha).

rated information from gauged data, satellite observations or large scale atmospheric model outputs for the hydrological modeling of the Negro River basin. Improved understanding of rainfall variabil-

ity during long time periods will also potentially help atmospheric models to improve estimates of precipitation, thus improving prediction of future rainfall variability.

A.C.V. Getirana et al. / Journal of Hydrology 404 (2011) 304322

319

Q (m3/s)
15,000

Caracarai

10,000

5,000

0 7,500

Taraqua

5,000

2,500

0 15,000

Cucui

10,000

5,000

0 30,000

Curicuriari

20,000

10,000

0 39,000

Serrinha

26,000

13,000

0 1-1-98

1-1-99
Obs

1-1-00
CPC

1-1-01
GPCP NCEP-2

1-1-02

Fig. 12b. Comparison of observed and simulated discharges obtained with the MGB-IPH model using a daily time step and forced by the CPC, GPCP and NCEP-2 precipitation datasets at ve gauge stations (Caracara, Taraqua, Cucu, Curicuriari and Serrinha).

Other satellite-based datasets that could be included in further analyses are the Climate Prediction Center Morphing technique precipitation dataset, CMORPH (Joyce et al., 2004) and the Precip-

itation Estimation from Remotely Sensed Information using Articial Neural Networks, PERSIANN (Sorooshian et al., 2000). Like TMPA and GPCP, these datasets are also based on passive micro-

320

A.C.V. Getirana et al. / Journal of Hydrology 404 (2011) 304322

Table 5 Values of the NashSutcliffe (NS) efciency coefcient for predicted streamows. Station Caracara Taraqua Cucu Curicuriari Serrinha HOP 0.84 0.69 0.80 0.80 0.86 CPC 0.87 1.02 0.42 0.13 0.18 GPCP 0.79 0.24 0.65 0.42 0.19 TMPA 0.61 0.16 0.10 0.02 0.38 ERA-40 0.52 1.26 0.20 0.32 0.81 NCEP-2 0.21 0.48 0.54 0.62 0.47

Table 6 Values of the relative error (RE) for predicted streamow volumes. Station Caracara Taraqua Cucu Curicuriari Serrinha HOP 17.2 4.8 2.7 13.7 2.5 CPC 17.1 65.4 21.9 32.8 36.9 GPCP 16.4 44.8 13.8 23.5 27.7 TMPA 31.5 44.1 37.3 37.2 40.8 ERA-40 24.3 66.6 22.9 39.4 45.7 NCEP-2 63.0 8.6 41.6 30.5 12.3

are global efforts with the objective of obtaining more accurate precipitation estimates at high spatio-temporal resolutions. The availability of such new data will require a better understanding of the potential and limitations of the available satellite precipitation products for hydrological studies. International initiatives such as those proposed by the International Precipitation Working Group (IPWG), the Program to Evaluate High Resolution Precipitation Products (PEHRPP) and the Distributed Model Intercomparison Project (DMIP) promote the assessment and understanding of precipitation elds and their impacts on the water cycle at the meso, regional and global scales via hydrological modeling. These global collaborative efforts should support more comprehensive studies of present and future water resources availability and provide guidelines for future development and water management.

Acknowledgments The rst author would like to thank CNPq and CAPES (Projeto 516/05) for nancial support. The study benetted from data made available by Agncia Nacional de guas (ANA, Brazil), Ministerio de Medio Ambiente y de los Recursos Naturales (MARN, Venezuela), Instituto de Hidrologa, Meteorologa y Estudios Ambientales (IDEAM, Colombia) and Observatoire de Recherche en Environnement Hybam (INSU). Grateful acknowledgements are also due to G. Cochonneau (IRD), B. Collischonn (ANA), J.-L. Guyot (IRD) and A. Laraque (IRD) for their help in data acquisition and processing, to M. Th. van Genuchten (UFRJ), A. Boone (CNRM/Mto France) and three anonymous reviewers for their suggestions and revision of the manuscript, and to W. Collischonn for providing the MGBIPH model. The authors also would like to acknowledge GSFC/ DAAC, NASA for providing TMPA data and V.B.S. Silva (NOAA) for providing the CPC dataset for South America. The GPCP-1dd dataset was obtained from the NOAA National Climatic Data Center (http://www.ncdc.noaa.gov/), the NCEP/NCAR Reanalysis dataset from the NOAA Climate Diagnostics Center (http://www.cdc.noaa.gov/), and the ECMWF ERA-40 dataset from the ECMWF Data Server (http://www.ecmwf.int/).

Table 7 Correlation coefcients (r) between observed and predicted streamow using different rainfall products. Station Caracara Taraqua Cucu Curicuriari Serrinha HOP 0.94 0.83 0.91 0.95 0.95 CPC 0.95 0.76 0.85 0.87 0.87 GPCP 0.91 0.66 0.85 0.83 0.80 TMPA 0.90 0.77 0.87 0.86 0.84 ERA-40 0.78 0.41 0.61 0.58 0.61 NCEP-2 0.80 0.14 0.63 0.46 0.41

wave and infrared data and constitute the latest generation of high spatiotemporal resolution precipitation data derived from satellite observations. Future spatial missions such as the Global Precipitation Measurement (GPM) and the Franco-Indian Megha Tropiques mission

12000 Caracara 9000 6000 3000 0 J F MAMJ J A SOND

6000 Taraqua 4500 3000 1500 0 J F MA MJ J A SOND

14000 Cucu 10500 7000 3500 0 J F MAMJ J A S OND

Q(m3/s)

30000 Curicuriari 22500 15000 7500 0 J F MAMJ J A SOND

HOP CPC GPCP

36000 Serrinha 27000 18000 9000 0 J F MA MJ J A SOND

TMPA ERA-40 NCEP-2 OBS

Fig. 13. Mean monthly discharges at ve gauge stations (Caracara, Taraqua, Cucu, Curicuriari and Serrinha).

Q(m3/s)

A.C.V. Getirana et al. / Journal of Hydrology 404 (2011) 304322

321

References
Adam, J.C., Lettenmaier, D.P., 2003. Adjustment of global gridded precipitation for systematic bias. J. Geophys. Res. 108, 114. Adler, R.F., Huffman, G.J., Chang, A., Ferraro, R., Xie, P.-P., Janowiak, J., Rudolf, B., Schneider, U., Curtis, S., Bolvin, D., Gruber, A., Susskind, J., Arkin, P., Nelkin, E., 2003. The Version-2 Global Precipitation Climatology Project (GPCP) Monthly Precipitation analysis (1979-present). J. Hydrometeorol. 4, 11471167. Adler, R.F., Kidd, C., Petty, G., Morissey, M., Goodman, H.M., 2001. Intercomparison of global precipitation products: the third precipitation intercomparison project (PIP-3). Bull. Am. Meteorol. Soc. 82 (7), 13771396. Allasia, D.G., Silva, B., Collischonn, W., Tucci, C.E.M., 2006. Large basin simulation experience in South America. IAHS Publication n. 303, pp. 360370. Arvor, D., Dubreuil, V., Ronchail, J., Meirelles, M., 2008. Apport des donnes TRMM 3B42 ltude des prcipitations au Mato Grosso. Climatologie 5, 4970. Betts, A.K., Ball, J.H., Bosilovich, M., Viterbo, P., Zhang, Y.-C., Rossow, W.B., 2003a. Intercomparison of water and energy budgets for ve Mississippi subbasins between ECMWF reanalysis (ERA-40) and NASA Data Assimilation Ofce fvGCM for 19901999. J. Geophys. Res. 108, 8618. doi:10.1029/2002JD003127. Betts, A.K., Ball, J.H., Viterbo, P., 2003. Evaluation of the ERA-40 surface water budget and surface temperature for the Mackenzie River basin. J. Hydrometeor. 4, 11941211. Betts, A.K., Ball, J.H., Viterbo, P., Dai, A., Marengo, J., 2005. Hydrometeorology of the Amazon in ERA-40. J. Hydrometeor. 6, 764774. Cochonneau, G., Sondag, F., Guyot, J.L., Boaventura, G., Filizola, N., Fraizy, P., Laraque, A., Magat, P., Martinez, J.M., Noriega, L., Oliveira, E., Ordonez, J.J., Pombosa, R., Seyler, F., Sidgwick, J., Vauchel, P., 2006. Lobservatoire de recherche en environnement ORE HYBAM sur les grands euves amazoniens. In: Demuth, S., Gustard, A., Planos, E., Scatena, F., Servat, E. (Eds.), 5th FRIEND World Conference Climate Variability and Change, Hydrological Impacts. La Havane, Cuba, 11/2006, IAHS, 308 4450. Collischonn, W., Allasia, D., Silva, B.C., Tucci, C.E.M., 2007. The MGB-IPH Model for Large-Scale Rainfall-Runoff Modelling. Hydrol. Sci. J. 52 (5), 878895. Collischonn, B., Collischonn, W., Tucci, C.E.M., 2008. Daily hydrological modeling in the Amazon basin using TRMM rainfall estimates. J. Hydrol.. doi:10.1016/ j.jhydrol.2008.07.032. Costa, M.H., Foley, J.A., 1998. A comparison of precipitation datasets for the Amazon basin. Geophys. Res. Lett. 25 (2), 155158. Dinku, T., Anagnostou, E.N., 2005. Regional Differences in Overland Rainfall Estimation from PR-Calibrated TMI Algorithm. J. Appl. Meteorol. 44, 189205. Espinoza Villar, J.C., Ronchail, J., Guyot, J.-L., Cochonneau, G., Filizola, N., Lavado, W., Noriega, L., de Oliveira, E., Pombosa, R., Romero, H., Vauchel, P., 2009. Spatiotemporal rainfall variability in the Amazon basin countries (Brazil, Peru, Bolivia, Colombia, and Ecuador). Int. J. Climatol. 29 (11), 15741594. doi:10.1002/ joc.1791. Fekete, B.M., Vrsmarty, C.J., Roads, J.O., Willmott, C.J., 2004. Uncertainties in precipitation and their impacts on runoff estimates. J. Clim. 17, 294304. Fernandes, K., Fu, R., Betts, A.K., 2008. How well does the ERA40 surface water budget compare to observations in the Amazon River basin? J. Geophys. Res. 113, D11117. doi:10.1029/2007JD009220. Figueroa, S.N., Nobre, C.A., 1990. Precipitation distribution over central and western tropical South America. Climanalise 6, 3640. Gebremichael, M., Krajewski, W.F., 2004. Assessment of the statistical characterization of small-scale rainfall variability from radar: analysis of TRMM ground validation datasets. J. Appl. Meteor. 43, 11801199. Gebremichael, M., Krajewski, W.F., Morrissey, M., Langerud, D., Huffman, G.J., Adler, R., 2003. Error uncertainty analysis of GPCP monthly rainfall products: a databased simulation study. J. Appl. Meteor. 42, 18371848. Getirana, A.C.V., Bonnet, M.-P., Martinez, J.-M., 2009a. Evaluating parameter effects in a DEM burning process based on land cover data. Hydrol. Process. 23 (16), 23162325. doi:10.1002/hyp.7303. Getirana, A.C.V., Bonnet, M.-P., Rotunno Filho, O.C., Guyot, J.-L., Seyler, F., Mansur, W.J., 2010. Hydrological modelling and water balance of the Negro River basin: evaluation based on in situ and spatial altimetry data 24, 32193236. doi:10.1002/hyp.7747. Getirana, A.C.V., Bonnet, M.-P., Rotunno Filho, O.C., Mansur, W.J., 2009b. Improving hydrological information acquisition from DEM processing in oodplains. Hydrol. Process. 23 (3), 502514. doi:10.1002/hyp.7167. Getirana, A.C.V., Bonnet, M.-P., Roux, E., Calmant, S., Rotunno Filho, O.C., Mansur, W.J., 2009c. Hydrological monitoring of large poorly gauged basins: a new approach based on spatial altimetry and distributed rainfall-runoff model. J. Hydrol. 1, 45. doi:10.1016/j.jhydrol.2009.09.049. Getirana, A.C.V., Frappart, F., Ramillien, G., Bonnet, M.P., Rotunno Filho, O.C., Mansur, W.J., 2009d. Negro River basin water balance: comparison between rainfall-runoff model outputs and GRACE-derived products. Joint International Convention of 8th IAHS Scientic Assembly and 37th IAH Congress Water: A vital resource under stress How Science can help, September 6-12, 2009, Hyderabad, India. Gonalves, L.G.G., Shuttleworth, W.J., Nijssen, B., Burke, E.J., Marengo, J.A., Chou, S.C., Houser, P., Toll, D.L., 2006. Evaluation of model-derived and remotely sensed precipitation products for continental South America. J. Geophys. Res. 111, D16113.1D16113.13. doi:10.1029/2005JD006276. Guyot, J.-L., Molinier, M., Guimares, V., Cudo, K.J., de Oliveira, E., 1993. Balano Hidrico da bacia do rio Negro. In: X Simpsio Brasileiro de Recursos Hdricos, Gramado, RS.

Hiez, G. 1977. Lhomognit des donnes pluviomtriques. Cahier ORSTOM, srie Hydrologie, 14, 129172. Hong, Y., Adler, R., Hossain, F., Curtis, S., 2007. A rst approach to global runoff simulation using satellite rainfall estimation. Water Resour. Res. 43, W08502. doi:10.1029/2006WR005739. Huffman, G.J., Adler, R.F., Morrissey, M., Bolvin, D.T., Curtis, S., Joyce, R., McGavock, B., Susskind, J., 2001. Global precipitation at one-degree daily resolution from multi-satellite observations. J. Hydrometeorol. 2, 3650. Huffman, G., Adler, R., Bolvin, D., Gu, G., Nelkin, E., Bowman, K., Hong, Y., Stocker, E., Wolff, D., 2007. The TRMM multisatellite precipitation analysis (TCMA): quasiglobal, multiyear, combined-sensor precipitation estimates at ne scales. J. Hydrometeorol. 8, 3855. Joyce, R.J., Janowiak, J.E., Arkin, P.A., Xie, P., 2004. CMORPH: A method that produces global precipitation estimates from passive microwave and infrared data at high spatial and temporal resolution. J. Hydrometeorol. 5, 487503. Juarez, R.I.N., Li, W., Fu, R., Fernandes, K., Cardoso, A.O., 2009. Comparison of precipitation datasets over the tropical South American and African continents. J. Hydrometeorol. 10, 289299. doi:10.1175/2008JHM1023.1. Kalnay, E., Kanamitsu, M., Kister, R., Collins, W., Deaven, D., Gandin, L., Iredell, M., Saha, S., White, G., Woollen, J., Zhu, Y., Leetmaa, A., Reynolds, B., Chelliah, M., Ebisuzaki, W., Higgins, W., Janowiak, J., Mo, K.C., Ropelewski, C., Wang, J., Jenne, R., Joseph, J., 1996. The NCEP/NCAR 40-years reanalyses project. Bull. Am. Meteorol. Soc. 77, 437471. Kanamitsu, M., Ebisuzaki, W., Woolen, J., Yang, S.-K., Hnilo, J.J., Fiorino, M., Potter, G.L., 2002. NCEP-DOE AMIP-II Reanalysis (R-2). Bull. Amer. Meteor. Soc. 83, 16311643. Karam, H.N., Bras, R.L., 2008. Estimates of net atmospheric moisture ux convergence over the Amazon basin: a comparison of reanalysis products. J. Hydrometeorol. 9, 10351047. doi:10.1175/2008JHM887.1. Kouwen, N., Soulis, E.D., Pietroniro, A., Donald, J., Harrington, R.A., 1993. Grouped response units for distributed hydrologic modelling. J. Water Resourc. Plan. Manage. 119 (3), 289305. Krysanova, V., Mller-Wohlfeil, D-I., Becker, A., 1998. Development and test of a spatially distributed hydrological/water quality model for mesoscale watersheds. Ecol. Model. 106, 261289. Liang, X., Lettenmaier, D.P., Wood, E.F., Burges, S.J., 1994. A simple hydrologically based model of land surface water and energy uxes for general circulation models. J. Geophys. Res. 99 (7), 1441514428. Malhi, Y., Pegoraro, E., Nobre, A.D., Pereira, M.G.P., Grace, J., Culf, A.D., Clement, R., 2002. Energy and water dynamics of a central Amazonian rainforest. J. Geophys. Res., 107, D20): 8061. Marengo, J.A., 2005. Characteristics and spatio-temporal variability of the Amazon river basin water budget. Climate Dynamics 24, 1122. doi:10.1007/s00382004-0461-6. Martinez, J.-M., Le Toan, T., 2007. Mapping of ood dynamics and spatial distribution of vegetation in the Amazon oodplain using multitemporal SAR data. Remote Sens. Environ. 108 (3), 209223. doi:10.1016/j.rse.2006.11012. New, M., Hume, M., Jones, P., 2000. Representing twentieth century spacetime climate variability: II. Development of 19011996 monthly grids of terrestrial surface. J. Climatol. 13, 22172238. Nijssen, B., Lettenmaier, D.P., 2004. Effect of precipitation sampling error on simulated hydrological uxes and states: Anticipating the Global Precipitation Measurement satellites. J. Geophys. Res. 109, D02103. Nijssen, B., ODonnell, G., Lettenmaier, D.P., Lohmann, D., Wood, E.F., 2001. Predicting the discharge of global rivers. J. Clim. 14, 33073323. Rao, V.B., Santo, C.E., Franchito, S.H., 2002. A diagnosis of rainfall over South America during the 1997/98 El Nio event. Part I: validation of NCEPNCAR reanalysis rainfall data. J. Clim. 15, 502511. Ribeiro Neto, A., Vieira da Silva, R., Collischonn, W., Tucci, C.E.M., 2005. Hydrological Modelling in Amazonia - Use of the MGB-IPH Model and Alternative Data Base. Proceedings of VII IAHS Scientic Assembly, Foz do Iguau, Brasil. Roads, J.O., 2003. The NCEPNCAR, NCEPDOE, and TRMM tropical atmosphere hydrologic cycles. J. Hydrometeor. 4, 826840. Rudolf, B., Hauschild, H., Rueth, W., Schneider, U., 1994. Terrestrial precipitation analysis: operational method and required density of point measurements. In: Desbois, M., Desalmand, F. (Eds.), Global Precipitation and Climate Change, NATO ASI Series, Springer Verlag, 1(26), 173186. Rudolf, B., Schneider, U., 2005. Calculation of gridded precipitation data for the global land-surface using in-situ gauge observations. Proc. 2nd Workshop of the Int. Precipitation Working Group IPWG, Monterey October 2004, EUMETSAT EUM P44, 231-247. Salati, E., Marquez, J., Molion, L.C., 1978. Origem e distribuio das chuvas na Amaznia. Interciencia 3, 200205. Sapiano, M.R.P., Stephenson, D.B., Grubb, H.J., Arkin, P.A., 2006. Diagnosis of variability and trends in a global precipitation dataset using a physically motivated statistical model. J. Clim. 19 (17), 41544166. doi:10.1175/ JCLI3849.1. Shuttleworth, W.J., 1993. Evaporation. In: Maidment, D., (Ed.), Handbook of Hydrology. McGraw-Hill, New York, USA, pp. 4.14.53. Shuttleworth, W.J., 1988. Evaporation from Amazonian forest. Proc. Roy. Soc. 233, 321346. Silva, V.B.S., Kousky, V.E., Shi, W., Higgins, R.W., 2007. An improved gridded historical daily precipitation analysis for Brazil. J. Hydrometeorol. 8, 847861. Sorooshian, S., Hsu, K., Gao, X., Gupta, H.V., Imam, B., Braithwaite, D., 2000. Evaluation of PERSIANN system satellite- based estimates of tropical rainfall. Bull. Am. Meteor. Soc. 81, 20352046.

322

A.C.V. Getirana et al. / Journal of Hydrology 404 (2011) 304322 Isaksen, L., Janssen, P.A.E.M., Jenne, R., McNally, A.P., Mahfouf, J.F., Morcrette, J.J., Rayner, N.A., Saunders, R.W., Simon, P., Sterl, A., Trenberth, K.E., Untch, A., Vasiljevic, D., Viterbo, P., Woollen, J., 2005. The ERA-40 re-analysis. Quart. J. R. Meteorol. Soc. 131, 2961. Voisin, N., Wood, A.W., Lettenmaier, D.P., 2008. Evaluation of precipitation products for global hydrological prediction. J. Hydrometeorol. 9, 388407. doi:10.1175/ 2007JHM938.1. Wigmosta, M.S., Vail, L.W., Lettenmaier, D.P., 1994. A distributed hydrologyvegetation model for complex terrain. Water Resour. Res. 30 (6), 16651679. Wilk, J., Kniveton, D., Andersson, L., Layberry, R., Todd, M.C., Hughes, D., Ringrose, S., Vanderpost, C., 2006. Estimating rainfall and water balance over the Okavango River Basin for hydrological applications. J. Hydrol. 331 (12), 1829. doi:10.1016/j.jhydrol.2006.04.049. Willmott, C.J., Robeson, S.M., Feddema, J., 1994. Estimating continental and terrestrial precipitation averages from rain-gauge networks. Int. Climatol. 14, 403414. Willmott, C.J., Matsuura, K., 2009. Terrestrial precipitation: monthly time series (19002006) Version 1.01. <http://climate.geog.udel.edu/$climate/ html_pages/download.html>. Wood, E.F., Lettenmaier, D.P., Zartarian, V.G., 1992. A land surface hydrology parameterization with subgrid variability for general circulation models. J. Geophys. Res. 97 (3), 27172728. Xue, Y., Sellers, P.J., Kinter, J.L., Shukla, J., 1991. A simplied biosphere model for global climate studies. J. Clim. 4, 345364. Yilmaz, K.K., Hogue, T.S., Hsu, K.-L., Sorooshian, S., Gupta, H.V., Wagener, T., 2005. Intercomparison of rain gauge, radar, and statellite-based precipitation estimates with emphasis on hydrologic forecasting. J. Hydrometeorol. 6, 497 517.

Su, F., Lettenmaier, D.P., 2009. Estimation of the surface water budget of the La Plata basin. J. Hydrometeorol. 10, 981998. doi:10.1175/2009JHM1100.1. Su, F., Hong, Y., Lettenmaier, D.P., 2008. Evaluation of TRMM multisatellite precipitation analysis (TMPA) and its utility in hydrologic prediction in the La Plata basin. J. Hydrometeorol. 9, 622640. doi:10.1175/2007JHM944.1. Su, F., Adam, J.C., Trenberth, K.E., Lettenmaier, D.P., 2006. Evaluation of surface water uxes of the pan-Arctic land region with a land surface model and ERA40 reanalysis. J. Geophys. Res. 111, D05110. doi:10.1029/2005JD006387. Syed, T.H., Lakshmi, V., Paleologos, E., Lohmann, D., Mitchell, K., Famiglietti, J., 2004. Analysis of process controls in land surface hydrological cycle over the continental United States. J. Geophys. Res. 109, D22105. doi:10.1029/ 2004JD004640. Tomasella, J., Hodnett, M.G., Cuartas, L.A., Nobre, A.D., Oliveira, M.J., Waterloo, S.M., 2007. The water balance of an Amazonian micro-catchment: the effect of interannual variability of rainfall on hydrological behaviour. Hydrol. Process.. doi:10.1002/hyp.6813. Tucci, C.E.M., Marengo, J.A., Silva Dias, P.L., Collischonn, W., Silva, B.C., Clarke, R.T., Cardoso, A.O., Juarez, R.N., Sampaio, G., Chan, C.S., Tomasella, J., 2005. Streamow Forecasting in So Francisco River Basin Based in the Climatic Forecasting. Technical Report ANEEL/WMO/98/00. Porto Alegre. Ubarana, V.N., 1996. Observation and modelling of rainfall interception loss in two experimental sites in Amazonian forest. In: Gash, J.H.C., Nobre, C.A., Roberts, J.M., Victoria, R.L. (Eds.), Amazonian Deforestation and Climate. John Wiley & Sons, Chichester, UK, pp. 51162. Uppala, SM., Kllberg, PW., Simmons, A.J., Andrae, U., da Costa Bechtold, V., Fiorino, M., Gibson, J.K., Haseler, J., Hernandez, A., Kelly, G.A., Li, X., Onogi, K., Saarinen, S., Sokka, N., Allan, R.P., Andersson, E., Arpe, K., Balmaseda, M.A., Beljaars, A.C.M., van de Berg, L., Bidlot, J., Bormann, N., Caires, S., Chevallier, F., Dethof, A., Dragosavac, M., Fisher, M., Fuentes, M., Hagemann, S., Hlm, E., Hoskins, BJ.,