Get 20M+ Full-Text Papers For Less Than $1.50/day. Start a 14-Day Trial for You or Your Team.

Learn More →

Dynamical bias correction procedure to improve global gridded daily streamflow data for local application in the Upper Blue Nile basin

Dynamical bias correction procedure to improve global gridded daily streamflow data for local... J. Hydrol. Hydromech., 69, 2021, 1, 41–48 ©2021. This is an open access article distributed DOI: 10.2478/johh-2020-0040 under the Creative Commons Attribution ISSN 1338-4333 NonCommercial-NoDerivatives 4.0 License Dynamical bias correction procedure to improve global gridded daily streamflow data for local application in the Upper Blue Nile basin 1* 2 Haileyesus Belay Lakew , Semu Ayalew Moges Ethiopian Space Science and Technology Institute, Addis Ababa, Ethiopia. Civil and Environmental Engineering, University of Connecticut, Storrs, CT, USA. E-mail: semu.moges.2000@gmail.com Corresponding author. Tel.: +251-912-17-94-83. E-mail: haileyesusbelay@gmail.com Abstract: Recently water resources reanalysis (WRR) global streamflow products are emerging from high- resolution global models as a means to provide long and consistent global streamflow products for assessment of global challenge such as climate change. Like any other products, the newly developed global streamflow products have limitations accurately represent the dynamics of local streamflow hydrographs. There is a need to locally evaluate and apply correction factors for better representation and make use of the data. This research focuses on the evaluation and correction of the bias embedded in the global streamflow product (WRR, 0.25°) developed by WaterGAP3 hydrological model in the upper Blue Nile basin part of Ethiopia. Three spatiotemporal dynamical bias correction schemes (temporal- spatial variable, temporal-spatial constant and spatial variable) tested in twelve watersheds of the basin. The temporal- spatial variable dynamical bias correction scheme significantly improves the streamflow estimation. The Nash-Sutcliffe coefficient (NSCE) improves by 30% and bias decreases by 19% for the twelve streamflow gauging stations applying leave one out cross-validation approach in turn. Therefore, the temporal-spatial variable scheme is applicable and can use as one method for the bias correction to use the global data for local applications in the upper Blue Nile basin. Keywords: Blue Nile; WaterGAP3; Bias Correction; Water Resource Reanalysis. INTRODUCTION al., 2019). The first evaluation of the global WRR products is carried out in our previous paper in the same basin of the upper Development of accurate hydrologic model predictions for Blue Nile for three case studies of small, medium and large current and potential future climatic conditions requires reliable spatial scales (Lakew et al., 2019). The evaluation was for spatially and temporally distributed climatic information (Hay various WRR1 (7 products) and WRR2 (4 products) for daily, and Clark, 2003; Jasper et al., 2002). However, the availability monthly as well as wet season temporal scales. The evaluation and quality of climate data vary from region to region, and result highlighted that the recently released WaterGAP3 prod- observations are sparse and irregularly distributed in many uct exhibits consistently better performance statistics than the areas (El-Sadek et al., 2011; Haberlandt and Kite, 1998; earlier coarser-resolution WRR1 and the rest of the WRR2 Kanamaru and Kanamitsu, 2007). Especially in Africa in which products at all ranges of temporal and spatial scales (Lakew et the gauging stations are sparse and irregularly distributed, ac- al., 2019). Even though the WaterGAP3 product exhibits con- quiring accurate and consistent hydro-meteorological data is a sistently better performance, it still needs bias correction to challenging task. Given this high-resolution satellite and rea- capture the gauged streamflow in the upper Blue Nile basin. nalysis precipitation estimation are investigated for more than Therefore, there is a gap in adjusting the mean annual stream- two decades (Guetter et al., 1996; Hossain and Anagnostou, flow on the newly released WRR products to correct the bias 2004; Lakew et al., 2017, 2020; Mei et al., 2016; Nikolopoulos embedded using bias correction approaches for improvements et al., 2010). for various local applications. Along with precipitation, to improve the spatiotemporal Based on this, it is vital to narrow the gap, and the main aim availability of streamflow data and to avail a consistent set of of this study is to evaluate and propose a bias correction proce- data for various global and regional water resources analysis, dure that would enable to downscale the global streamflow eartH2Observe (http://www.earth2observe.eu) developed glob- product WaterGAP3 to a regional scale. The bias correction al streamflow WRR products (Schellekens et al., 2017). The schemes were carried out at daily temporal scales using twelve products with the preceding version 1 (WRR1, 0.5°) and the observed streamflow gauging stations in the upper Blue Nile recent version 2 (WRR2, 0.25°) using the global hydrological basin. The spatiotemporal dynamical bias correction procedures models of WaterGAP, ORCHIDEE, AWRA-L, TESSEL, (temporal-spatial variable, temporal-spatial constant and spatial LISFLOOD, SURFEX and PCR-GLOBWB (Lakew et al., variable) were applied and evaluated for the dynamical bias 2019). Such products have a significant role in the assessment correction scheme. The leave-one-out cross-validation (LOOCV) of global water balance and water management applications, approach implemented for each streamflow stations in turn. especially in data-scarce regions of the world. As such global products usually not correctly represent the DATA AND METHOD local dynamics of the hydrological cycle and also have the Study area limitations to estimate the gauged streamflow data. These glob- al products should be evaluated and corrected before they use The Blue Nile River (known as the Abbay River in Ethiopia) for water resources applications for local case studies (Lakew et rises in the Ethiopian highlands in the region of West Gojam Haileyesus Belay Lakew, Semu Ayalew Moges and flows northward into Lake Tana, located at an elevation of were used and shown in Figure 1. The updated streamflow just under 1,800 m. The basin stretches from 34°24'E–39°48'E gauged data is available for a period between 1980 and 2010 for the twelve gauging stations and the mean streamflow with and 7°42'N–12°30'N and covers a total area of 199,812 km the corresponding basin area shown in Table 1. (Figure 1). The basin annual mean streamflow is about 247mm measured at Sudan border. It is the most vital basin in the coun- Table 1. The upstream basin area and the mean streamflow of the try by most criteria as it contributes about 45% of the country 12 streamflow gauging stations. surface water resources, accommodates 25% of the population, 20% of the landmass, 40% of the nation’s agricultural produc- 2 3 No. Station Area (km ) Mean flow (m /s) tion and most of the hydropower and a significant portion of 1 Eldiem 199,812 1,505.7 irrigation potential of the country (Erkossa et al., 2009). The 2 Kessie 65,784 529.7 basin characterized by a highly seasonal rainfall pattern with 3 Gilgel Abbay 1,656 54.6 most of the rainfall falling in four months (June to September, 4 Main Beles 3,431 58.6 JJAS) with a peak in July or August. The mean annual rainfall 5 Chacha 418 4.1 for the 1961–1990 period amounts to a little over 1,200 mm, of 6 Guder 524 12.3 which more than 70% fall during those four months. More than 7 Neshi 322 7.6 80% of the annual flow in the Blue Nile results from the sum- 8 Dedissa (Dembi) 1,806 40.5 mer monsoon and concentrated between June and September 9 Ribb 1,592 14.2 (Elshamy et al., 2009; Setegn et al., 2010). 10 Abbay (Bahir Dar) 15,321 116.1 11 Gumera 1,394 34.3 12 Birr 978 16 WaterGAP3 reanalysis global streamflow product The WaterGAP3 (Water Global Assessment and Prognosis) is a global model of water availability as well as water use has been developed to assess the current water resources situation and to estimate the impact of global change on the problem of water scarcity (Müller Schmied et al., 2014; Schellekens et al., 2017). The WaterGAP3 model was developed at the Centre for Environmental Systems Research of the University of Kassel, Germany, in cooperation with the National Institute of Public Health and the Environment of the Netherlands (RIVM). The goals of the model are: • to enable a comparison of the “freshwater situation” in different parts of the world, i.e. the uses and availability of freshwater to meet various objectives related to the requirements of society and aquatic ecosystems; • to provide a long term perspective (at least a few decades) on changes in global water resources. The WaterGAP3 model consists of two main components a Global Water Use model and a Global Hydrology model. The Water Use model takes into account basic socio-economic factors that lead to domestic, industrial and agricultural water use, while the Hydrology model incorporates physical and climate factors that lead to runoff and groundwater recharge (Eisner, 2016). Two multi-model WRR streamflow product of WaterGAP2 is available at 0.5° (WRR1) and WaterGAP3 0.25° Fig. 1. Location of streamflow stations in the upper Blue Nile basin (WRR2) grid resolutions, recently produced in the framework with the global streamflow data of WaterGAP3 (0.25°). of a European Union project (eartH2Observe). In the current study, the high-resolution (0.25°) of WaterGAP3 considered for Local gauge streamflow data the bias correction analysis due to its consistent and better performance statistics than from the WRR products at all rang- The observed streamflow datasets obtained from the Ethio- es of temporal and spatial scales evaluated (Lakew et al., 2019). pian Ministry of Water, Irrigation and Energy used to evaluate the global grid-based reanalysis streamflow data and serve as a Schemes for bias correction reference to estimate the bias factor of the corresponding pixel of the gridded global dataset. The gauging station network in Both global precipitation products of satellite and reanalysis the upper Blue Nile basin is sparse as well as the stations dis- rainfall estimates exhibit large systematic and random errors tributed unevenly over the basin area. From the whole set of (Bitew and Gebremichael, 2011; Habib et al., 2014; Lakew et streamflow gauging stations, twelve gauging stations that al., 2020). Hence, WRR global products consist of errors cas- would cover the spatial coverage of the basin with a better caded from precipitation forcing, global hydrological model quality of long records from 1980 to 2010 selected for bias formulation and processing simplification. Thus, the bias in correction and validation procedures. The streamflow gauging global streamflow products should be evaluated and corrected stations Eldiem, Kessie, Gilgel Abbay, Beles, Chacha, Guder, before the global streamflow products use for various applica- Neshi, Dedissa, Ribb, Abbay (Bahir Dar), Gumera and Birr tions for local case studies. 42 Dynamical bias correction procedure to improve global gridded daily streamflow data We used a multiplicative daily bias reduction estimator (Bias where n is the total number of gauges within the spatial cover- Factor, BF) to correct the bias in the WaterGAP3 model prod- age of the study, and T is the full duration of the study period. uct. The correction for the WaterGAP3 bias, the pixel-based The bias correction, in this case, is applied by dividing the daily bias factors were calculated for some gauging stations and WaterGAP3 estimates by the bias factor, BF , to result in a TSC spatially interpolated with the same WaterGAP3 spatial 0.25° new set of WaterGAP3 estimates that are bias-corrected in a grid resolution. For the case of streamflow estimates, we used spatially and temporally-lumped scheme. the nearest neighbor method of spatial interpolation as well as it is applicable for the insertion of the bias factor values at each III. Spatial Variable, BFSV pixel for the entire basin with 0.25° spatial resolution. For the In this formulation, the BF is temporally lumped over the en- evaluation and analysis, three dynamical bias correction tire domain but, it is still estimated for each streamflow gauge schemes were applied and evaluated. Temporal-spatial variable, stations Equation (3): temporal-spatial constant and spatial variable bias factors were estimated to correct WaterGAP3 estimates. The new set of (3) spatiotemporal bias-corrected of WaterGAP3 streamflow grid- ded data produced by dividing the uncorrected global gridded The bias correction, in this case, is applied by dividing each data of WaterGAP3 by the spatially interpolated gridded bias WaterGAP3 field by the bias factor, BF , to result in a new set SV factor for the respective time window that is computed by the of WaterGAP3 estimates that are bias-corrected in a temporally three methods discussed below. lumped but spatially varying scheme. The bias factor at a given pixel for the three schemes was I. Temporal Spatial Variable, BFTSV computed by dividing the uncorrected WaterGAP3 to the ob- In this study, we estimated and corrected the bias in Wa- served (gauged) streamflow. This computation might lead to terGAP3 estimates as follows. The multiplicative daily bias unexpected extremely large or small (near to zero) bias factor factor (BF) of a given streamflow gauging station with the values. To avoid the propagation of large and small magnitudes WaterGAP3 grid global streamflow data formulated as follows. of the bias factor to the neighbor pixels during interpolation, we For each streamflow observation station, 365 bias factors de- restricted the bias factor value between 0.5 and 2. The prelimi- termined. The same with the daily temporal scale, the monthly nary evaluation of the global WaterGAP3 product for the most bias factor was computed using Equation (1). For monthly tem- streamflow gauging stations bias value was between 0.5 and 2. poral scale, 12 bias factors generated for each streamflow station. If the BF value of the streamflow station of a given pixel is greater than 2 or less than 0.5, the BF will be replaced automat- (1) ically by 2 or 0.5 respectively before the interpolation. The bias factor of the streamflow stations was interpolated where G and WaterGAP3 represent daily (monthly) gauge and using the nearest neighbor method as well the interpolated map global streamflow product, respectively, i refers to gauge loca- of the bias factor values shown in Figure 2. The map shows the tion, and t refers to a Julian day or month number. The interpolated map of the BF scheme for temporally and spa- TSV subscript “TSV” stands for “Temporal Spatial Variable” since tially varied schemes for randomly selected date of the first July the bias factor in this formulation estimated for a specific pixel of 2000 (peak runoff season). with 0.25° spatial resolution and a particular day or month. Each pixel has the corresponding bias factor value for a specific day (month). The bias factor is dynamic spatially and tempo- rally. This scheme helps in adjusting the bias at a pixel-based at 0.25° spatial and at a daily or monthly temporal scales (i.e., time and space varying), and based on using the BF factor TSV estimated from Equation (1). The interpolation yields a spatial and temporally varying field of BFs over the entire study area and over the whole time window. The spatially gridded Wa- terGAP3 daily or monthly streamflow fields were then divided by the spatially gridded and interpolated BFTSV bias fields for the respective time windows. This result a new set of bias- corrected WaterGAP3 estimates that as such are bias-corrected in a temporally and spatially varying scheme. II. Temporal Spatial Constant, BF TSC Temporal and spatial constant (TSC) bias correction: in this formulation, the bias factor estimated using Equation (2). The average daily WaterGAP3 and the average observed streamflow over 31 years (1980–2010) for the twelve stations computed and obtained by dividing the average WaterGAP3 product with the average observed streamflow values. Then the bias factor of the eleven stations computed and interpolated for the 0.25° grid resolution to get temporally and spatially lumped value. (2) Fig. 2. The interpolated bias factor values of the upper Blue Nile basin using the nearest neighbor method. Haileyesus Belay Lakew, Semu Ayalew Moges From the total stations, three stations of Chacha, Guder and equal to the sample size (12). The advantage is that the proce- Neshi have less drainage area from the grid cell of the global dure delivers the same results every time because all possible streamflow data (Table 1). This might have an impact on the options evaluated. rest stations that have large drainage area during the BF inter- polation. However, the BF values restricted between the mini- RESULT mum and maximum bounds of 0.5 and 2. Furthermore, the Evaluation of bias correction schemes three stations of Chacha, Guder and Neshi have independent BF values shown in Figure 2 for temporally and spatially varied BF To evaluate the three bias correction schemes that were ap- values. The three stations considered in this study due to the plied for the gridded based global WaterGAP3 streamflow shortage of streamflow stations. product. For the demonstration, we selected three streamflow gauging stations from a total of twelve stations. The three sta- Performance of bias corrected reanalysis streamflow tions represent a wide range of different spatial scales are small 2 2 (WaterGAP3) (Gilgel Abbay, 1,656 km ), medium (Kessie, 65,784 km ) and large scale (Eldiem, 199,812 km ). The recent and the available The new set of bias-corrected streamflow product computed updated observed streamflow data from 2000 to 2010 taken for by the three schemes evaluated using statistical metrics between the comparison of the three schemes. the bias-corrected and observed streamflow at a daily and The results of the three statistical metrics show that the bias- monthly temporal scales. The three statistical metrics are de- corrected data, applying the bias factor of the temporal-spatial scribed below. variable (BFTSV) outperforms the other method of bias correc- First, for statistical goodness of fit of simulated streamflow, tion for the three different spatial scales as shown in Table 2. In we utilized the Nash-Sutcliffe coefficient of efficiency (Nash the case of the Eldiem (large basin), the NSCE performance and Sutcliffe, 1970). shows significant improvement from 0.80 to 0.95 for the uncor- rected and corrected WaterGAP3 product respectively applying BFTSV bias correction scheme. The bias also reveals a high , , (4) reduction from 13.9% to 1.2% as the WaterGAP3 is corrected using BF . Similarly, in the case of the medium-scale of TSV Kessie shows a high improvement of NSCE and reduction of where is the observed streamflow of the ith day; is the , , bias applying BF scheme. The NSCE improves from 0.66 to TSV global streamflow (WaterGAP3) of ith day; and is the aver- 0.93 as well as the bias reduces from –12% to 7.2%. The NSCE age of all the daily observed streamflow values. If NSCE ≤ 0, performance result from Gilgel Abbay of small basin scale then the model provides no skill in relation to using the ob- makes sure that the scheme of bias correction using BFTSV gives served mean as a predictor and values greater than zero indicat- the best performance from among the methods used for the ing better agreement. three case studies of small, medium and large basin scales. The Second, the Pearson correlation coefficient (CC) is used to bias efficiency metric shows that the result from BFsv scores assess the agreement between simulated and observed stream- minimum bias value for medium and large basin scales. How- flow as follows: ever, the bias value of Gilgel Abbay shows highest from the other schemes. The result of bias of Gilgel Abbay and the per- , , formance of NSCE of the three case studies shows that BF is TSV (5) applicable for the streamflow bias correction in the upper Blue , , Nile basin. The BF and BF manifest poor performance TSC SV even from the uncorrected WaterGAP3 products, that is due to where is the average of all daily simulated streamflow (Wa- the bias values are highly varied temporally and spatially in the terGAP3) values. upper Blue Nile basin and should not be lumped in both as- Third, relative bias ratio assesses the systematic bias of the pects. Lumping the bias factors temporally or spatially might simulated discharge: decrease the performance even from the uncorrected WaterGAP3 product. , , (6) , Table 2. Different schemes of bias correction for the three-different sized basin scales from 2000 to 2010. The best skill occurs with NSCE = 1, CC = 1, and Bias = 0%. Uncorrected Correction Schemes for Station WaterGAP3 WaterGAP3 Cross-validation BFTSV BFTSC BFSV Eldiem NSCE 0.80 0.95 0.76 0.83 In cross-validation (CV), some parts of the datasets are held Bias 13.9 1.2 21.3 0.1 back when the model is fitted, and used for evaluating the error (100%) of the fitted model. The hold-out is called the test, the other part CC 0.91 0.98 0.91 0.91 of the training data set. Cross-validation approaches differ in Kessie NSCE 0.66 0.93 0.60 0.58 how many of the data are allocated to test/training, and how Bias –12 –7.2 –3.9 0.38 often the CV itself is repeated (Lachenbruch and Mickey, 1968; (100%) Stone, 1977). In our case, we used the leave-one-out cross- CC 0.83 0.97 0.81 0.81 validation (LOOCV) method for the analysis. In LOOCV, a Gilgel NSCE 0.66 0.76 0.67 0.61 single observation from the sample used as test data and the Abbay Bias –21.7 –5.3 –16.7 –31.3 remaining observations as training data. It leaves out only 1 (100%) data point and does that for each data point in turn (Badr et al., CC 0.83 0.87 0.83 0.83 2014). This is done repeatedly N times, with N has chosen 44 Dynamical bias correction procedure to improve global gridded daily streamflow data Evaluation of the selected bias correction scheme and due to the large upstream area errors average out more as com- streamflow pared to streamflow from smaller basins. Typical, large-scale hydrological models perform better for downstream stations The result from the evaluation of the three bias correction than for upstream stations and the rest show relatively weak schemes of the recent period from 2000 to 2010 reveals the performance below 0.6 (except Gilgel Abbay). Especially, BFTSV is applicable for the upper Blue Nile basin. Considering Abbay (Bair Dar) shows the lowest NSCE performance of –1.56 this, we implemented this bias correction scheme applying that is due to the effect of lake Tana. This result indicates that twelve gauging stations of Eldiem, Kessie, Gilgel Abbay, the uncorrected product of WaterGAP3 manifests poor perfor- Beles, Chacha, Guder, Neshi, Dedissa, Ribb, Abbay (Bahir mance, especially it has less efficiency for the stations that have Dar), Gumera and Birr in the upper Blue Nile basin The sta- a reservoir effect at the upstream of the streamflow gauging tions are with a long period of record from 1980 to 2010 at stations. daily temporal scale. There are 365 BF values for each of the The new set of bias-corrected WaterGAP3 streamflow prod- TSV eleven stations values interpolated using the nearest neighbor uct shows high NSCE efficiency improvement. From the total method and the remaining one station used for validation apply- 12 stations, 33% of the stations score above 0.8, and 58% of the ing leave-one-out cross-validation approach. This procedure stations score above 0.7 of NSCE evaluation metric in the cor- carried out for each twelve streamflow stations that were used rected WaterGAP3 new set of product. for bias correction in the upper Blue Nile basin, leaving one Based on bias performance evaluation of the corrected grid- observation out of the calibration data set, recalibrating the ded WaterGAP3 streamflow product, more than half of the model, and doing that for each data point in turn. stations score less than 10% of bias, such as Gumera and Ribb The result reveals that the bias correction applying BFTSV for score the minimum bias value of –1.9% and 3.8% respectively. the gridded streamflow data WaterGAP3 shows significant im- More than 80% of the stations score below 25% bias value for provement for long-term data for the case studies. Nash-Sutcliffe the new set of bias-corrected streamflow gridded data. coefficient (NSCE) improves by 30% and bias decreases by 19% The overall NSCE results manifest that except Guder (0.87) for the twelve streamflow gauging stations (Table 3). the stations with small watershed area show relatively weak The NSCE performance of the uncorrected WaterGAP3 performance compared to the large watersheds for the corrected product from 1980 to 2010 shows that only Eldiem scores WaterGAP3 product. The small watersheds with the weak higher than 0.70. Eldiem is the most downstream station, where performance of NSCE are Chacha (0.41), Neshi (0.52) and Table 3. Performance of daily and monthly WaterGAP3 streamflow with and without bias-correction using BFTSV (1980–2010). Daily Monthly No. Station WaterGAP3 Bias corrected WaterGAP3 Bias corrected WaterGAP3 WaterGAP3 1 Eldiem NSCE 0.73 0.85 0.79 0.85 (199,812 km ) Bias (100%) 24.2 6.1 24.2 6.1 CC 0.90 0.94 0.93 0.95 2 Kessie NSCE 0.37 0.73 0.43 0.74 (65,784 km ) Bias (100%) 7 –5.8 6.6 –0.58 CC 0.66 0.91 0.71 0.92 3 Gilgel Abbay NSCE 0.67 0.86 0.84 0.87 (1,656 km ) Bias (100%) –18 –8.6 –17 –8.4 CC 0.83 0.93 0.93 0.95 4 Main Beles NSCE 0.44 0.78 0.71 0.79 (3,431 km ) Bias (100%) –17.3 –12.5 –10.9 –6.7 CC 0.67 0.91 0.85 0.97 5 Chacha NSCE 0.12 0.41 0.45 0.82 (418 km ) Bias (100%) 63.4 25.1 63.4 25.3 CC 0.62 0.73 0.85 0.91 6 Guder NSCE 0.48 0.87 0.79 0.91 (524 km ) Bias (100%) 21.7 7.1 21.7 7.1 CC 0.77 0.92 0.91 0.94 7 Neshi NSCE 0.19 0.52 0.22 0.55 (322 km ) Bias (100%) –53.2 –38.1 –53 –38 CC 0.64 0.83 0.77 0.90 8 Dedissa (1,806 km ) NSCE 0.30 0.68 0.48 0.76 Bias (100%) –38.3 –23.2 –38 –20 CC 0.62 0.86 0.87 0.94 9 Ribb (1,592 km ) NSCE 0.54 0.84 0.75 0.88 Bias (100%) 8.6 –3.8 7.9 –29 CC 0.80 0.92 0.89 0.92 10 Abbay NSCE –1.56 0.64 –1.5 0.67 (Bahir Dar) Bias (100%) 107.7 41 106 40.5 (15,321 km ) CC 0.61 0.84 0.63 0.87 11 Gumera NSCE 0.47 0.70 0.71 0.87 (1,394 km ) Bias (100%) 15.4 1.9 15.3 1.9 CC 0.78 0.84 0.88 0.93 12 Birr NSCE 0.34 0.46 0.34 0.68 (978 km ) Bias (100%) –36.4 –25.6 –36.3 –25.2 CC 0.63 0.71 0.85 0.88 Haileyesus Belay Lakew, Semu Ayalew Moges Birr (0.46) as shown in Table 3. This indicates that large water- and Gumera shows that the corrected WaterGAP3 manifests sheds with a large magnitude of streamflow have high perfor- significant improvement from the uncorrected WaterGAP3 to mance to estimate the observed streamflow for the corrected capture the shape and the magnitude of the hydrograph for both global product of WaterGAP3. This is because of that the Wa- Gilgel Abbay and Gumera stations. This result indicates that the terGAP3 product has better performance of minimum bias for corrected WaterGAP3 product using BF bias correction TSV the larger basin than smaller basin to estimate the streamflow scheme is applicable for local applications for the data-scares magnitude. region of the upper Blue Nile. The daily WaterGAP3 and observed streamflow were aggre- As a final step of this dynamical bias correction, the analysis gated to the monthly mean flow to investigate the effectiveness extended to find out how many of the twelve observational time of the temporal-spatial variable scheme for the monthly tem- series are required to improve performance applying BFTSV bias poral scale and shown in Table 3. The bias correction of the correction scheme at a daily time scale. The analysis was car- monthly BFTSV was carried out as the same procedure as the ried out by taking a different number of stations using leave- daily temporal scale. The monthly temporal scale bias-corrected one-out cross-validation (LOOCV) technique for each trial and new product of WaterGAP3 improves by much and performs each station and shown in Table 4. The average values of the better even from the daily temporal scale for the whole stream- evaluation metrics of NSCE, Bias and CC for each analysis flow stations shown in Table 3. These all results show that the were computed and tabulated below. The result shows that for bias-adjusted new set of high resolution (0.25°) gridded stream- the upper Blue Nile basin that has a drainage area of 199,812 flow data shows high performance to estimate the observed km , the BF bias correction procedure is applicable for more TSV streamflow applying BF scheme. This new set of bias- than nine streamflow stations to improve the performance. TSV corrected data can be in use for the data-scarce region of the Taking a lesser number of stations for the analysis might make upper Blue Nile basin, Ethiopia. the performance worse even from the uncorrected WaterGAP3. For the randomly selected the year of 2000, the corrected Mainly, if the stations are less than six, it would deteriorate the and uncorrected WaterGAP3 daily products plotted using hy- original global streamflow data shown in Table 4. This is due to drograph for Gilgel Abbay and Gumera stations. From the the stations’ bias factor (BF) are spatially varied and taking an twelve case studies that used for bias correction scheme (Gilgel insignificant number of stations of less than six would make the Abbay and Gumera) stations were selected randomly to plot the performance worse even from the original global streamflow hydrograph (Figure 3). The hydrograph of both Gilgel Abbay data (uncorrected) of WaterGAP3 model. Fig. 3. Hydrograph of the corrected, uncorrected WaterGAP3 and observed streamflow for (a) Gilgel Abbay and (b) Gumera watersheds for the year of 2000. Table 4. Performance for different number of stations for the bias correction procedure in the upper Blue Nile basin. Number of stations taken for the bias correction Un corrected 12 11 10 9 6 3 2 1 NSCE 0.70 0.67 0.63 0.56 0.34 0.28 0.27 0.21 0.36 Bias (100%) 16.6 18.2 19.8 22.1 34.9 38.9 40.1 43.6 34.3 CC 0.86 0.85 0.83 0.81 0.69 0.64 0.64 0.54 0.71 46 Dynamical bias correction procedure to improve global gridded daily streamflow data DISCUSSION AND CONCLUSIONS bias correction was applied only using twelve streamflow sta- tions in the upper Blue Nile basin. We recommend further This study contributes to the improvement of the perfor- investigation based on the analysis of a reasonably large num- mance of global WaterGAP3 product by implement bias correc- ber of stations with high-quality and recently updated stream- tion schemes to develop a new set of dynamically bias- flow data. corrected gridded streamflow product for the data-scarce region Taking more streamflow stations (>9) for the procedure of the upper Blue Nile basin. The new bias-corrected Wa- makes the bias correction BF scheme applicable and im- TSV terGAP3 daily data is compared with the uncorrected and also proves the performance. On the other side, taking a lesser num- with previous study findings. Previously, (Lakew et al., 2019) ber of stations (<6), will make the performance worse even the first evaluation of the global WRR products (including the from the uncorrected WaterGAP3 model and not recommended WaterGAP3) was carried out at a range of spatial and temporal to apply for the bias correction procedure. scales in the nested watersheds (Gilgel Abbay, Kessie and Eldiem) for the upper Blue Nile basin. The results revealed that Acknowledgements. The authors would like to thank the Ethio- the WaterGAP3 product manifests better and consistent per- pian Ministry of Water, Irrigation and Energy for the updated formance from the other products. However, the findings have observed streamflow data. highlighted that the accuracy and reliability of global hydrolog- ical products vary greatly as a function of the model selected Conflicts of interest. The authors declare no conflict of interest. and spatial/temporal requirements and thus uncertainty evalua- tion and bias correction of those products should precede their REFERENCES use in practice. Therefore, this study aims to implement bias Badr, H.S., Zaitchik, B.F., Guikema, S.D., 2014. Application of correction to improve the performance of the WaterGAP3 Statistical Models to the Prediction of Seasonal Rainfall product using dynamical bias correction schemes. Anomalies over the Sahel. J. Appl. Meteorol. Climatol., 53, In the previous studies and findings, the results revealed that 614–636. https://doi.org/10.1175/JAMC-D-13-0181.1 the reanalysis streamflow cannot capture the gauged stream- Bitew, M.M., Gebremichael, M., 2011. Assessment of satellite flow for local applications. The daily reanalysis streamflow rainfall products for streamflow simulation in medium product of the uncorrected WaterGAP3 has high bias range watersheds of the Ethiopian highlands. Hydrol. Earth Syst. from 7% to 100.7%, and the NSCE performance is poor and Sci., 15, 1147–1155. https://doi.org/10.5194/hess-15-1147- between –1.56 to 0.7 for the selected streamflow gauging sta- tions in the upper Blue Nile basin. This indicates that the uncor- Eisner, S., 2016. Comprehensive evaluation of the WaterGAP3 rected global WaterGAP3 streamflow product has an error in model across climatic, physiographic, and anthropogenic streamflow generation mechanisms to estimate the observed gradients. PhD thesis. University of Kassel, Kassel, streamflow for the local basin river network and needs bias Germany. Available at http://nbn- correction or adjustment for various applications at the regional resolving.de/urn:nbn:de:hebis:34-2016031450014 scale. However, not a lot has been done to demonstrate how El-Sadek, A., Bleiweiss, M., Shukla, M., Guldan, S., Fernald, these products can be ready for various local application by A., 2011. Alternative climate data sources for distributed reducing their errors. hydrological modelling on a daily time step. Hydrol. This research focuses on improving the quality of global Process., 25, 1542–1557. https://doi.org/10.1002/hyp.7917 streamflow data by bias-correction to make the product ready Elshamy, M.E., Seierstad, I.A., Sorteberg, A., 2009. Impacts of for diverse applications in the upper Blue Nile basin, Ethiopia. climate change on Blue Nile flows using bias-corrected Three spatiotemporal dynamical bias correction schemes (tem- GCM scenarios. Hydrol. Earth Syst. Sci., 13, 551–565. poral-spatial variable, temporal-spatial constant and spatial https://doi.org/10.5194/hess-13-551-2009 variable) tested in the twelve watersheds (streamflow gauging Erkossa, T., Awlachew, S., Haileslassie, A., Denekew Yilma, stations) of the basin. For the bias correction scheme and analy- A., 2009. Impacts of improving water management of sis, from the whole observed streamflow stations in the basin, smallholder agriculture in the Upper Blue Nile Basin. twelve stations that have a better quality of records selected for https://doi.org/DOI: 10.22004/ag.econ.212433 bias correction procedure using long term set of observed data Guetter, A.K., Georgakakos, K.P., Tsonis, A.A., 1996. from 1980 to 2010 at a daily temporal scale. Hydrologic applications of satellite data: 2. Flow simulation The bias correction applying BFTSV manifests significant and soil water estimates. J. Geophys. Res. Atmos., 101, improvement compared to the other schemes for the global 26527–26538. https://doi.org/10.1029/96JD01655 streamflow dataset. The NSCE improves by 30% and bias Haberlandt, U., Kite, G.W., 1998. Estimation of daily space– decreases by 19% for streamflow gauging stations that used for time precipitation series for macroscale hydrological the bias correction in the upper Blue Nile basin. From the total modelling. Hydrol. Process., 12, 1419–1432. stations, 33% of the stations score above 0.8, and 58% of the https://doi.org/10.1002/(SICI)1099- stations score above 0.7 of NSCE. Based on bias evaluation 1085(199807)12:9<1419::AID-HYP645>3.0.CO;2-A metrics, more than half of the stations score less than 5% of Habib, E., Haile, A., Sazib, N., Zhang, Y., Rientjes, T., 2014. bias in the corrected WaterGAP3 new set of daily gridded Effect of bias correction of satellite-rainfall estimates on streamflow product. The leave-one-out cross-validation runoff simulations at the source of the Upper Blue Nile. (LOOCV) approach applied for the bias correction and valida- Remote Sens., 6, Article Number: 6688. tion procedures. https://doi.org/doi.org/10.3390/rs6076688 These results reveal that the temporal-spatial variable Hay, L.E., Clark, M.P., 2003. Use of statistically and scheme is applicable and can use as one method for the bias dynamically downscaled atmospheric model output for correction to use the global data for local applications. The hydrologic simulations in three mountainous basins in the bias-adjusted new set of high resolution (0.25°) gridded stream- western United States. J. Hydrol., 282, 56–75. flow data applying BFTSV can be in use as one source of data for https://doi.org/10.1016/s0022-1694(03)00252-x various applications in the upper Blue Nile basin, Ethiopia. The Haileyesus Belay Lakew, Semu Ayalew Moges Hossain, F., Anagnostou, E.N., 2004. Assessment of current Remote Sens., 8, Article Number: 293. passive-microwave- and infrared-based satellite rainfall https://doi.org/https://doi.org/10.3390/rs8040293 remote sensing for flood prediction. J. Geophys. Res. Müller Schmied, H., Eisner, S., Franz, D., Wattenbach, M., Atmos., 109, Article Number: D07102. Portmann, F.T., Flörke, M., Döll, P., 2014. Sensitivity of https://doi.org/10.1029/2003JD003986 simulated global-scale freshwater fluxes and storages to Jasper, K., Gurtz, J., Lang, H., 2002. Advanced flood input data, hydrological model structure, human water use forecasting in Alpine watersheds by coupling meteorological and calibration. Hydrol. Earth Syst. Sci., 18, 3511–3538. observations and forecasts with a distributed hydrological https://doi.org/10.5194/hess-18-3511-2014 model. J. Hydrol., 267, 40–52. Nash, J.E., Sutcliffe, J. V., 1970. River flow forecasting through https://doi.org/https://doi.org/10.1016/S0022-1694(02)00138-5 conceptual models part I - A discussion of principles. J. Kanamaru, H., Kanamitsu, M., 2007. Fifty-seven-year Hydrol., 10, 282–290. https://doi.org/10.1016/0022- California reanalysis downscaling at 10 km (CaRD10). Part II: 1694(70)90255-6 Comparison with North American regional reanalysis. J. Nikolopoulos, E.I., Anagnostou, E.N., Hossain, F., Clim., 20, 5572–5592. https://doi.org/10.1175/2007jcli1522.1 Gebremichael, M., Borga, M., 2010. Understanding the scale Lachenbruch, P.A., Mickey, M.R., 1968. Estimation of error relationships of uncertainty propagation of satellite rainfall rates in discriminant analysis. Technometrics, 10, 1–11. through a distributed hydrologic model. J. Hydrometeorol., https://doi.org/10.1080/00401706.1968.10490530 11, 520–532. https://doi.org/10.1175/2009jhm1169.1 Lakew, H.B., Moges, S.A., Asfaw, D.H., 2017. Hydrological Schellekens, J., Dutra, E., Martínez-de la Torre, A., Balsamo, evaluation of satellite and reanalysis precipitation products G., van Dijk, A., Sperna Weiland, F., Minvielle, M., Calvet, in the Upper Blue Nile basin: A case study of Gilgel Abbay. J.-C., Decharme, B., Eisner, S., Fink, G., Flörke, M., Hydrology, 4, Article Number: 39. Peßenteiner, S., van Beek, R., Polcher, J., Beck, H., Orth, R., https://doi.org/https://doi.org/10.3390/hydrology4030039 Calton, B., Burke, S., Dorigo, W., Weedon, G.P., 2017. A Lakew, H.B., Moges, S.A., Anagnostou, E.N., Nikolopoulos, global water resources ensemble of hydrological models: the E.I., Asfaw, D.H., 2019. Evaluation of global water eartH2Observe Tier-1 dataset. Earth Syst. Sci. Data, 9, 389– resources reanalysis runoff products for local water 413. https://doi.org/10.5194/essd-9-389-2017 resources applications: Case study - Upper Blue Nile basin Setegn, S.G., Srinivasan, R., Melesse, A.M., Dargahi, B., 2010. of Ethiopia. Water Resour. Manag., SWAT model application and prediction uncertainty https://doi.org/10.1007/s11269-019-2190-y analysis in the Lake Tana Basin, Ethiopia. Hydrol. Process., Lakew, H.B., Moges, S.A., Asfaw, D.H., 2020. Hydrological 24, 357–367. https://doi.org/10.1002/hyp.7457 performance evaluation of multiple satellite precipitation Stone, M., 1977. An asymptotic equivalence of choice of model products in the upper Blue Nile basin, Ethiopia. J. Hydrol. by cross-validation and Akaike’s criterion. J. R. Stat. Soc. Reg. Stud., 27, Article Number: 100664. Ser. B, 39, 44–47. https://doi.org/10.1111/j.2517- https://doi.org/https://doi.org/10.1016/j.ejrh.2020.100664 6161.1977.tb01603.x Mei, Y., Nikolopoulos, E., Anagnostou, E., Zoccatelli, D., Borga, M., 2016. Error analysis of satellite precipitation- Received 31 January 2020 driven modeling of flood events in complex Alpine terrain. Accepted 25 September 2020 http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Journal of Hydrology and Hydromechanics de Gruyter

Dynamical bias correction procedure to improve global gridded daily streamflow data for local application in the Upper Blue Nile basin

Download PDF
8 pages

Loading next page...
 
/lp/de-gruyter/dynamical-bias-correction-procedure-to-improve-global-gridded-daily-G8CzFezv3W

References

References for this paper are not available at this time. We will be adding them shortly, thank you for your patience.

Publisher
de Gruyter
Copyright
© 2021 Haileyesus Belay Lakew et al., published by Sciendo
ISSN
0042-790X
eISSN
0042-790X
DOI
10.2478/johh-2020-0040
Publisher site
See Article on Publisher Site

Abstract

J. Hydrol. Hydromech., 69, 2021, 1, 41–48 ©2021. This is an open access article distributed DOI: 10.2478/johh-2020-0040 under the Creative Commons Attribution ISSN 1338-4333 NonCommercial-NoDerivatives 4.0 License Dynamical bias correction procedure to improve global gridded daily streamflow data for local application in the Upper Blue Nile basin 1* 2 Haileyesus Belay Lakew , Semu Ayalew Moges Ethiopian Space Science and Technology Institute, Addis Ababa, Ethiopia. Civil and Environmental Engineering, University of Connecticut, Storrs, CT, USA. E-mail: semu.moges.2000@gmail.com Corresponding author. Tel.: +251-912-17-94-83. E-mail: haileyesusbelay@gmail.com Abstract: Recently water resources reanalysis (WRR) global streamflow products are emerging from high- resolution global models as a means to provide long and consistent global streamflow products for assessment of global challenge such as climate change. Like any other products, the newly developed global streamflow products have limitations accurately represent the dynamics of local streamflow hydrographs. There is a need to locally evaluate and apply correction factors for better representation and make use of the data. This research focuses on the evaluation and correction of the bias embedded in the global streamflow product (WRR, 0.25°) developed by WaterGAP3 hydrological model in the upper Blue Nile basin part of Ethiopia. Three spatiotemporal dynamical bias correction schemes (temporal- spatial variable, temporal-spatial constant and spatial variable) tested in twelve watersheds of the basin. The temporal- spatial variable dynamical bias correction scheme significantly improves the streamflow estimation. The Nash-Sutcliffe coefficient (NSCE) improves by 30% and bias decreases by 19% for the twelve streamflow gauging stations applying leave one out cross-validation approach in turn. Therefore, the temporal-spatial variable scheme is applicable and can use as one method for the bias correction to use the global data for local applications in the upper Blue Nile basin. Keywords: Blue Nile; WaterGAP3; Bias Correction; Water Resource Reanalysis. INTRODUCTION al., 2019). The first evaluation of the global WRR products is carried out in our previous paper in the same basin of the upper Development of accurate hydrologic model predictions for Blue Nile for three case studies of small, medium and large current and potential future climatic conditions requires reliable spatial scales (Lakew et al., 2019). The evaluation was for spatially and temporally distributed climatic information (Hay various WRR1 (7 products) and WRR2 (4 products) for daily, and Clark, 2003; Jasper et al., 2002). However, the availability monthly as well as wet season temporal scales. The evaluation and quality of climate data vary from region to region, and result highlighted that the recently released WaterGAP3 prod- observations are sparse and irregularly distributed in many uct exhibits consistently better performance statistics than the areas (El-Sadek et al., 2011; Haberlandt and Kite, 1998; earlier coarser-resolution WRR1 and the rest of the WRR2 Kanamaru and Kanamitsu, 2007). Especially in Africa in which products at all ranges of temporal and spatial scales (Lakew et the gauging stations are sparse and irregularly distributed, ac- al., 2019). Even though the WaterGAP3 product exhibits con- quiring accurate and consistent hydro-meteorological data is a sistently better performance, it still needs bias correction to challenging task. Given this high-resolution satellite and rea- capture the gauged streamflow in the upper Blue Nile basin. nalysis precipitation estimation are investigated for more than Therefore, there is a gap in adjusting the mean annual stream- two decades (Guetter et al., 1996; Hossain and Anagnostou, flow on the newly released WRR products to correct the bias 2004; Lakew et al., 2017, 2020; Mei et al., 2016; Nikolopoulos embedded using bias correction approaches for improvements et al., 2010). for various local applications. Along with precipitation, to improve the spatiotemporal Based on this, it is vital to narrow the gap, and the main aim availability of streamflow data and to avail a consistent set of of this study is to evaluate and propose a bias correction proce- data for various global and regional water resources analysis, dure that would enable to downscale the global streamflow eartH2Observe (http://www.earth2observe.eu) developed glob- product WaterGAP3 to a regional scale. The bias correction al streamflow WRR products (Schellekens et al., 2017). The schemes were carried out at daily temporal scales using twelve products with the preceding version 1 (WRR1, 0.5°) and the observed streamflow gauging stations in the upper Blue Nile recent version 2 (WRR2, 0.25°) using the global hydrological basin. The spatiotemporal dynamical bias correction procedures models of WaterGAP, ORCHIDEE, AWRA-L, TESSEL, (temporal-spatial variable, temporal-spatial constant and spatial LISFLOOD, SURFEX and PCR-GLOBWB (Lakew et al., variable) were applied and evaluated for the dynamical bias 2019). Such products have a significant role in the assessment correction scheme. The leave-one-out cross-validation (LOOCV) of global water balance and water management applications, approach implemented for each streamflow stations in turn. especially in data-scarce regions of the world. As such global products usually not correctly represent the DATA AND METHOD local dynamics of the hydrological cycle and also have the Study area limitations to estimate the gauged streamflow data. These glob- al products should be evaluated and corrected before they use The Blue Nile River (known as the Abbay River in Ethiopia) for water resources applications for local case studies (Lakew et rises in the Ethiopian highlands in the region of West Gojam Haileyesus Belay Lakew, Semu Ayalew Moges and flows northward into Lake Tana, located at an elevation of were used and shown in Figure 1. The updated streamflow just under 1,800 m. The basin stretches from 34°24'E–39°48'E gauged data is available for a period between 1980 and 2010 for the twelve gauging stations and the mean streamflow with and 7°42'N–12°30'N and covers a total area of 199,812 km the corresponding basin area shown in Table 1. (Figure 1). The basin annual mean streamflow is about 247mm measured at Sudan border. It is the most vital basin in the coun- Table 1. The upstream basin area and the mean streamflow of the try by most criteria as it contributes about 45% of the country 12 streamflow gauging stations. surface water resources, accommodates 25% of the population, 20% of the landmass, 40% of the nation’s agricultural produc- 2 3 No. Station Area (km ) Mean flow (m /s) tion and most of the hydropower and a significant portion of 1 Eldiem 199,812 1,505.7 irrigation potential of the country (Erkossa et al., 2009). The 2 Kessie 65,784 529.7 basin characterized by a highly seasonal rainfall pattern with 3 Gilgel Abbay 1,656 54.6 most of the rainfall falling in four months (June to September, 4 Main Beles 3,431 58.6 JJAS) with a peak in July or August. The mean annual rainfall 5 Chacha 418 4.1 for the 1961–1990 period amounts to a little over 1,200 mm, of 6 Guder 524 12.3 which more than 70% fall during those four months. More than 7 Neshi 322 7.6 80% of the annual flow in the Blue Nile results from the sum- 8 Dedissa (Dembi) 1,806 40.5 mer monsoon and concentrated between June and September 9 Ribb 1,592 14.2 (Elshamy et al., 2009; Setegn et al., 2010). 10 Abbay (Bahir Dar) 15,321 116.1 11 Gumera 1,394 34.3 12 Birr 978 16 WaterGAP3 reanalysis global streamflow product The WaterGAP3 (Water Global Assessment and Prognosis) is a global model of water availability as well as water use has been developed to assess the current water resources situation and to estimate the impact of global change on the problem of water scarcity (Müller Schmied et al., 2014; Schellekens et al., 2017). The WaterGAP3 model was developed at the Centre for Environmental Systems Research of the University of Kassel, Germany, in cooperation with the National Institute of Public Health and the Environment of the Netherlands (RIVM). The goals of the model are: • to enable a comparison of the “freshwater situation” in different parts of the world, i.e. the uses and availability of freshwater to meet various objectives related to the requirements of society and aquatic ecosystems; • to provide a long term perspective (at least a few decades) on changes in global water resources. The WaterGAP3 model consists of two main components a Global Water Use model and a Global Hydrology model. The Water Use model takes into account basic socio-economic factors that lead to domestic, industrial and agricultural water use, while the Hydrology model incorporates physical and climate factors that lead to runoff and groundwater recharge (Eisner, 2016). Two multi-model WRR streamflow product of WaterGAP2 is available at 0.5° (WRR1) and WaterGAP3 0.25° Fig. 1. Location of streamflow stations in the upper Blue Nile basin (WRR2) grid resolutions, recently produced in the framework with the global streamflow data of WaterGAP3 (0.25°). of a European Union project (eartH2Observe). In the current study, the high-resolution (0.25°) of WaterGAP3 considered for Local gauge streamflow data the bias correction analysis due to its consistent and better performance statistics than from the WRR products at all rang- The observed streamflow datasets obtained from the Ethio- es of temporal and spatial scales evaluated (Lakew et al., 2019). pian Ministry of Water, Irrigation and Energy used to evaluate the global grid-based reanalysis streamflow data and serve as a Schemes for bias correction reference to estimate the bias factor of the corresponding pixel of the gridded global dataset. The gauging station network in Both global precipitation products of satellite and reanalysis the upper Blue Nile basin is sparse as well as the stations dis- rainfall estimates exhibit large systematic and random errors tributed unevenly over the basin area. From the whole set of (Bitew and Gebremichael, 2011; Habib et al., 2014; Lakew et streamflow gauging stations, twelve gauging stations that al., 2020). Hence, WRR global products consist of errors cas- would cover the spatial coverage of the basin with a better caded from precipitation forcing, global hydrological model quality of long records from 1980 to 2010 selected for bias formulation and processing simplification. Thus, the bias in correction and validation procedures. The streamflow gauging global streamflow products should be evaluated and corrected stations Eldiem, Kessie, Gilgel Abbay, Beles, Chacha, Guder, before the global streamflow products use for various applica- Neshi, Dedissa, Ribb, Abbay (Bahir Dar), Gumera and Birr tions for local case studies. 42 Dynamical bias correction procedure to improve global gridded daily streamflow data We used a multiplicative daily bias reduction estimator (Bias where n is the total number of gauges within the spatial cover- Factor, BF) to correct the bias in the WaterGAP3 model prod- age of the study, and T is the full duration of the study period. uct. The correction for the WaterGAP3 bias, the pixel-based The bias correction, in this case, is applied by dividing the daily bias factors were calculated for some gauging stations and WaterGAP3 estimates by the bias factor, BF , to result in a TSC spatially interpolated with the same WaterGAP3 spatial 0.25° new set of WaterGAP3 estimates that are bias-corrected in a grid resolution. For the case of streamflow estimates, we used spatially and temporally-lumped scheme. the nearest neighbor method of spatial interpolation as well as it is applicable for the insertion of the bias factor values at each III. Spatial Variable, BFSV pixel for the entire basin with 0.25° spatial resolution. For the In this formulation, the BF is temporally lumped over the en- evaluation and analysis, three dynamical bias correction tire domain but, it is still estimated for each streamflow gauge schemes were applied and evaluated. Temporal-spatial variable, stations Equation (3): temporal-spatial constant and spatial variable bias factors were estimated to correct WaterGAP3 estimates. The new set of (3) spatiotemporal bias-corrected of WaterGAP3 streamflow grid- ded data produced by dividing the uncorrected global gridded The bias correction, in this case, is applied by dividing each data of WaterGAP3 by the spatially interpolated gridded bias WaterGAP3 field by the bias factor, BF , to result in a new set SV factor for the respective time window that is computed by the of WaterGAP3 estimates that are bias-corrected in a temporally three methods discussed below. lumped but spatially varying scheme. The bias factor at a given pixel for the three schemes was I. Temporal Spatial Variable, BFTSV computed by dividing the uncorrected WaterGAP3 to the ob- In this study, we estimated and corrected the bias in Wa- served (gauged) streamflow. This computation might lead to terGAP3 estimates as follows. The multiplicative daily bias unexpected extremely large or small (near to zero) bias factor factor (BF) of a given streamflow gauging station with the values. To avoid the propagation of large and small magnitudes WaterGAP3 grid global streamflow data formulated as follows. of the bias factor to the neighbor pixels during interpolation, we For each streamflow observation station, 365 bias factors de- restricted the bias factor value between 0.5 and 2. The prelimi- termined. The same with the daily temporal scale, the monthly nary evaluation of the global WaterGAP3 product for the most bias factor was computed using Equation (1). For monthly tem- streamflow gauging stations bias value was between 0.5 and 2. poral scale, 12 bias factors generated for each streamflow station. If the BF value of the streamflow station of a given pixel is greater than 2 or less than 0.5, the BF will be replaced automat- (1) ically by 2 or 0.5 respectively before the interpolation. The bias factor of the streamflow stations was interpolated where G and WaterGAP3 represent daily (monthly) gauge and using the nearest neighbor method as well the interpolated map global streamflow product, respectively, i refers to gauge loca- of the bias factor values shown in Figure 2. The map shows the tion, and t refers to a Julian day or month number. The interpolated map of the BF scheme for temporally and spa- TSV subscript “TSV” stands for “Temporal Spatial Variable” since tially varied schemes for randomly selected date of the first July the bias factor in this formulation estimated for a specific pixel of 2000 (peak runoff season). with 0.25° spatial resolution and a particular day or month. Each pixel has the corresponding bias factor value for a specific day (month). The bias factor is dynamic spatially and tempo- rally. This scheme helps in adjusting the bias at a pixel-based at 0.25° spatial and at a daily or monthly temporal scales (i.e., time and space varying), and based on using the BF factor TSV estimated from Equation (1). The interpolation yields a spatial and temporally varying field of BFs over the entire study area and over the whole time window. The spatially gridded Wa- terGAP3 daily or monthly streamflow fields were then divided by the spatially gridded and interpolated BFTSV bias fields for the respective time windows. This result a new set of bias- corrected WaterGAP3 estimates that as such are bias-corrected in a temporally and spatially varying scheme. II. Temporal Spatial Constant, BF TSC Temporal and spatial constant (TSC) bias correction: in this formulation, the bias factor estimated using Equation (2). The average daily WaterGAP3 and the average observed streamflow over 31 years (1980–2010) for the twelve stations computed and obtained by dividing the average WaterGAP3 product with the average observed streamflow values. Then the bias factor of the eleven stations computed and interpolated for the 0.25° grid resolution to get temporally and spatially lumped value. (2) Fig. 2. The interpolated bias factor values of the upper Blue Nile basin using the nearest neighbor method. Haileyesus Belay Lakew, Semu Ayalew Moges From the total stations, three stations of Chacha, Guder and equal to the sample size (12). The advantage is that the proce- Neshi have less drainage area from the grid cell of the global dure delivers the same results every time because all possible streamflow data (Table 1). This might have an impact on the options evaluated. rest stations that have large drainage area during the BF inter- polation. However, the BF values restricted between the mini- RESULT mum and maximum bounds of 0.5 and 2. Furthermore, the Evaluation of bias correction schemes three stations of Chacha, Guder and Neshi have independent BF values shown in Figure 2 for temporally and spatially varied BF To evaluate the three bias correction schemes that were ap- values. The three stations considered in this study due to the plied for the gridded based global WaterGAP3 streamflow shortage of streamflow stations. product. For the demonstration, we selected three streamflow gauging stations from a total of twelve stations. The three sta- Performance of bias corrected reanalysis streamflow tions represent a wide range of different spatial scales are small 2 2 (WaterGAP3) (Gilgel Abbay, 1,656 km ), medium (Kessie, 65,784 km ) and large scale (Eldiem, 199,812 km ). The recent and the available The new set of bias-corrected streamflow product computed updated observed streamflow data from 2000 to 2010 taken for by the three schemes evaluated using statistical metrics between the comparison of the three schemes. the bias-corrected and observed streamflow at a daily and The results of the three statistical metrics show that the bias- monthly temporal scales. The three statistical metrics are de- corrected data, applying the bias factor of the temporal-spatial scribed below. variable (BFTSV) outperforms the other method of bias correc- First, for statistical goodness of fit of simulated streamflow, tion for the three different spatial scales as shown in Table 2. In we utilized the Nash-Sutcliffe coefficient of efficiency (Nash the case of the Eldiem (large basin), the NSCE performance and Sutcliffe, 1970). shows significant improvement from 0.80 to 0.95 for the uncor- rected and corrected WaterGAP3 product respectively applying BFTSV bias correction scheme. The bias also reveals a high , , (4) reduction from 13.9% to 1.2% as the WaterGAP3 is corrected using BF . Similarly, in the case of the medium-scale of TSV Kessie shows a high improvement of NSCE and reduction of where is the observed streamflow of the ith day; is the , , bias applying BF scheme. The NSCE improves from 0.66 to TSV global streamflow (WaterGAP3) of ith day; and is the aver- 0.93 as well as the bias reduces from –12% to 7.2%. The NSCE age of all the daily observed streamflow values. If NSCE ≤ 0, performance result from Gilgel Abbay of small basin scale then the model provides no skill in relation to using the ob- makes sure that the scheme of bias correction using BFTSV gives served mean as a predictor and values greater than zero indicat- the best performance from among the methods used for the ing better agreement. three case studies of small, medium and large basin scales. The Second, the Pearson correlation coefficient (CC) is used to bias efficiency metric shows that the result from BFsv scores assess the agreement between simulated and observed stream- minimum bias value for medium and large basin scales. How- flow as follows: ever, the bias value of Gilgel Abbay shows highest from the other schemes. The result of bias of Gilgel Abbay and the per- , , formance of NSCE of the three case studies shows that BF is TSV (5) applicable for the streamflow bias correction in the upper Blue , , Nile basin. The BF and BF manifest poor performance TSC SV even from the uncorrected WaterGAP3 products, that is due to where is the average of all daily simulated streamflow (Wa- the bias values are highly varied temporally and spatially in the terGAP3) values. upper Blue Nile basin and should not be lumped in both as- Third, relative bias ratio assesses the systematic bias of the pects. Lumping the bias factors temporally or spatially might simulated discharge: decrease the performance even from the uncorrected WaterGAP3 product. , , (6) , Table 2. Different schemes of bias correction for the three-different sized basin scales from 2000 to 2010. The best skill occurs with NSCE = 1, CC = 1, and Bias = 0%. Uncorrected Correction Schemes for Station WaterGAP3 WaterGAP3 Cross-validation BFTSV BFTSC BFSV Eldiem NSCE 0.80 0.95 0.76 0.83 In cross-validation (CV), some parts of the datasets are held Bias 13.9 1.2 21.3 0.1 back when the model is fitted, and used for evaluating the error (100%) of the fitted model. The hold-out is called the test, the other part CC 0.91 0.98 0.91 0.91 of the training data set. Cross-validation approaches differ in Kessie NSCE 0.66 0.93 0.60 0.58 how many of the data are allocated to test/training, and how Bias –12 –7.2 –3.9 0.38 often the CV itself is repeated (Lachenbruch and Mickey, 1968; (100%) Stone, 1977). In our case, we used the leave-one-out cross- CC 0.83 0.97 0.81 0.81 validation (LOOCV) method for the analysis. In LOOCV, a Gilgel NSCE 0.66 0.76 0.67 0.61 single observation from the sample used as test data and the Abbay Bias –21.7 –5.3 –16.7 –31.3 remaining observations as training data. It leaves out only 1 (100%) data point and does that for each data point in turn (Badr et al., CC 0.83 0.87 0.83 0.83 2014). This is done repeatedly N times, with N has chosen 44 Dynamical bias correction procedure to improve global gridded daily streamflow data Evaluation of the selected bias correction scheme and due to the large upstream area errors average out more as com- streamflow pared to streamflow from smaller basins. Typical, large-scale hydrological models perform better for downstream stations The result from the evaluation of the three bias correction than for upstream stations and the rest show relatively weak schemes of the recent period from 2000 to 2010 reveals the performance below 0.6 (except Gilgel Abbay). Especially, BFTSV is applicable for the upper Blue Nile basin. Considering Abbay (Bair Dar) shows the lowest NSCE performance of –1.56 this, we implemented this bias correction scheme applying that is due to the effect of lake Tana. This result indicates that twelve gauging stations of Eldiem, Kessie, Gilgel Abbay, the uncorrected product of WaterGAP3 manifests poor perfor- Beles, Chacha, Guder, Neshi, Dedissa, Ribb, Abbay (Bahir mance, especially it has less efficiency for the stations that have Dar), Gumera and Birr in the upper Blue Nile basin The sta- a reservoir effect at the upstream of the streamflow gauging tions are with a long period of record from 1980 to 2010 at stations. daily temporal scale. There are 365 BF values for each of the The new set of bias-corrected WaterGAP3 streamflow prod- TSV eleven stations values interpolated using the nearest neighbor uct shows high NSCE efficiency improvement. From the total method and the remaining one station used for validation apply- 12 stations, 33% of the stations score above 0.8, and 58% of the ing leave-one-out cross-validation approach. This procedure stations score above 0.7 of NSCE evaluation metric in the cor- carried out for each twelve streamflow stations that were used rected WaterGAP3 new set of product. for bias correction in the upper Blue Nile basin, leaving one Based on bias performance evaluation of the corrected grid- observation out of the calibration data set, recalibrating the ded WaterGAP3 streamflow product, more than half of the model, and doing that for each data point in turn. stations score less than 10% of bias, such as Gumera and Ribb The result reveals that the bias correction applying BFTSV for score the minimum bias value of –1.9% and 3.8% respectively. the gridded streamflow data WaterGAP3 shows significant im- More than 80% of the stations score below 25% bias value for provement for long-term data for the case studies. Nash-Sutcliffe the new set of bias-corrected streamflow gridded data. coefficient (NSCE) improves by 30% and bias decreases by 19% The overall NSCE results manifest that except Guder (0.87) for the twelve streamflow gauging stations (Table 3). the stations with small watershed area show relatively weak The NSCE performance of the uncorrected WaterGAP3 performance compared to the large watersheds for the corrected product from 1980 to 2010 shows that only Eldiem scores WaterGAP3 product. The small watersheds with the weak higher than 0.70. Eldiem is the most downstream station, where performance of NSCE are Chacha (0.41), Neshi (0.52) and Table 3. Performance of daily and monthly WaterGAP3 streamflow with and without bias-correction using BFTSV (1980–2010). Daily Monthly No. Station WaterGAP3 Bias corrected WaterGAP3 Bias corrected WaterGAP3 WaterGAP3 1 Eldiem NSCE 0.73 0.85 0.79 0.85 (199,812 km ) Bias (100%) 24.2 6.1 24.2 6.1 CC 0.90 0.94 0.93 0.95 2 Kessie NSCE 0.37 0.73 0.43 0.74 (65,784 km ) Bias (100%) 7 –5.8 6.6 –0.58 CC 0.66 0.91 0.71 0.92 3 Gilgel Abbay NSCE 0.67 0.86 0.84 0.87 (1,656 km ) Bias (100%) –18 –8.6 –17 –8.4 CC 0.83 0.93 0.93 0.95 4 Main Beles NSCE 0.44 0.78 0.71 0.79 (3,431 km ) Bias (100%) –17.3 –12.5 –10.9 –6.7 CC 0.67 0.91 0.85 0.97 5 Chacha NSCE 0.12 0.41 0.45 0.82 (418 km ) Bias (100%) 63.4 25.1 63.4 25.3 CC 0.62 0.73 0.85 0.91 6 Guder NSCE 0.48 0.87 0.79 0.91 (524 km ) Bias (100%) 21.7 7.1 21.7 7.1 CC 0.77 0.92 0.91 0.94 7 Neshi NSCE 0.19 0.52 0.22 0.55 (322 km ) Bias (100%) –53.2 –38.1 –53 –38 CC 0.64 0.83 0.77 0.90 8 Dedissa (1,806 km ) NSCE 0.30 0.68 0.48 0.76 Bias (100%) –38.3 –23.2 –38 –20 CC 0.62 0.86 0.87 0.94 9 Ribb (1,592 km ) NSCE 0.54 0.84 0.75 0.88 Bias (100%) 8.6 –3.8 7.9 –29 CC 0.80 0.92 0.89 0.92 10 Abbay NSCE –1.56 0.64 –1.5 0.67 (Bahir Dar) Bias (100%) 107.7 41 106 40.5 (15,321 km ) CC 0.61 0.84 0.63 0.87 11 Gumera NSCE 0.47 0.70 0.71 0.87 (1,394 km ) Bias (100%) 15.4 1.9 15.3 1.9 CC 0.78 0.84 0.88 0.93 12 Birr NSCE 0.34 0.46 0.34 0.68 (978 km ) Bias (100%) –36.4 –25.6 –36.3 –25.2 CC 0.63 0.71 0.85 0.88 Haileyesus Belay Lakew, Semu Ayalew Moges Birr (0.46) as shown in Table 3. This indicates that large water- and Gumera shows that the corrected WaterGAP3 manifests sheds with a large magnitude of streamflow have high perfor- significant improvement from the uncorrected WaterGAP3 to mance to estimate the observed streamflow for the corrected capture the shape and the magnitude of the hydrograph for both global product of WaterGAP3. This is because of that the Wa- Gilgel Abbay and Gumera stations. This result indicates that the terGAP3 product has better performance of minimum bias for corrected WaterGAP3 product using BF bias correction TSV the larger basin than smaller basin to estimate the streamflow scheme is applicable for local applications for the data-scares magnitude. region of the upper Blue Nile. The daily WaterGAP3 and observed streamflow were aggre- As a final step of this dynamical bias correction, the analysis gated to the monthly mean flow to investigate the effectiveness extended to find out how many of the twelve observational time of the temporal-spatial variable scheme for the monthly tem- series are required to improve performance applying BFTSV bias poral scale and shown in Table 3. The bias correction of the correction scheme at a daily time scale. The analysis was car- monthly BFTSV was carried out as the same procedure as the ried out by taking a different number of stations using leave- daily temporal scale. The monthly temporal scale bias-corrected one-out cross-validation (LOOCV) technique for each trial and new product of WaterGAP3 improves by much and performs each station and shown in Table 4. The average values of the better even from the daily temporal scale for the whole stream- evaluation metrics of NSCE, Bias and CC for each analysis flow stations shown in Table 3. These all results show that the were computed and tabulated below. The result shows that for bias-adjusted new set of high resolution (0.25°) gridded stream- the upper Blue Nile basin that has a drainage area of 199,812 flow data shows high performance to estimate the observed km , the BF bias correction procedure is applicable for more TSV streamflow applying BF scheme. This new set of bias- than nine streamflow stations to improve the performance. TSV corrected data can be in use for the data-scarce region of the Taking a lesser number of stations for the analysis might make upper Blue Nile basin, Ethiopia. the performance worse even from the uncorrected WaterGAP3. For the randomly selected the year of 2000, the corrected Mainly, if the stations are less than six, it would deteriorate the and uncorrected WaterGAP3 daily products plotted using hy- original global streamflow data shown in Table 4. This is due to drograph for Gilgel Abbay and Gumera stations. From the the stations’ bias factor (BF) are spatially varied and taking an twelve case studies that used for bias correction scheme (Gilgel insignificant number of stations of less than six would make the Abbay and Gumera) stations were selected randomly to plot the performance worse even from the original global streamflow hydrograph (Figure 3). The hydrograph of both Gilgel Abbay data (uncorrected) of WaterGAP3 model. Fig. 3. Hydrograph of the corrected, uncorrected WaterGAP3 and observed streamflow for (a) Gilgel Abbay and (b) Gumera watersheds for the year of 2000. Table 4. Performance for different number of stations for the bias correction procedure in the upper Blue Nile basin. Number of stations taken for the bias correction Un corrected 12 11 10 9 6 3 2 1 NSCE 0.70 0.67 0.63 0.56 0.34 0.28 0.27 0.21 0.36 Bias (100%) 16.6 18.2 19.8 22.1 34.9 38.9 40.1 43.6 34.3 CC 0.86 0.85 0.83 0.81 0.69 0.64 0.64 0.54 0.71 46 Dynamical bias correction procedure to improve global gridded daily streamflow data DISCUSSION AND CONCLUSIONS bias correction was applied only using twelve streamflow sta- tions in the upper Blue Nile basin. We recommend further This study contributes to the improvement of the perfor- investigation based on the analysis of a reasonably large num- mance of global WaterGAP3 product by implement bias correc- ber of stations with high-quality and recently updated stream- tion schemes to develop a new set of dynamically bias- flow data. corrected gridded streamflow product for the data-scarce region Taking more streamflow stations (>9) for the procedure of the upper Blue Nile basin. The new bias-corrected Wa- makes the bias correction BF scheme applicable and im- TSV terGAP3 daily data is compared with the uncorrected and also proves the performance. On the other side, taking a lesser num- with previous study findings. Previously, (Lakew et al., 2019) ber of stations (<6), will make the performance worse even the first evaluation of the global WRR products (including the from the uncorrected WaterGAP3 model and not recommended WaterGAP3) was carried out at a range of spatial and temporal to apply for the bias correction procedure. scales in the nested watersheds (Gilgel Abbay, Kessie and Eldiem) for the upper Blue Nile basin. The results revealed that Acknowledgements. The authors would like to thank the Ethio- the WaterGAP3 product manifests better and consistent per- pian Ministry of Water, Irrigation and Energy for the updated formance from the other products. However, the findings have observed streamflow data. highlighted that the accuracy and reliability of global hydrolog- ical products vary greatly as a function of the model selected Conflicts of interest. The authors declare no conflict of interest. and spatial/temporal requirements and thus uncertainty evalua- tion and bias correction of those products should precede their REFERENCES use in practice. Therefore, this study aims to implement bias Badr, H.S., Zaitchik, B.F., Guikema, S.D., 2014. Application of correction to improve the performance of the WaterGAP3 Statistical Models to the Prediction of Seasonal Rainfall product using dynamical bias correction schemes. Anomalies over the Sahel. J. Appl. Meteorol. Climatol., 53, In the previous studies and findings, the results revealed that 614–636. https://doi.org/10.1175/JAMC-D-13-0181.1 the reanalysis streamflow cannot capture the gauged stream- Bitew, M.M., Gebremichael, M., 2011. Assessment of satellite flow for local applications. The daily reanalysis streamflow rainfall products for streamflow simulation in medium product of the uncorrected WaterGAP3 has high bias range watersheds of the Ethiopian highlands. Hydrol. Earth Syst. from 7% to 100.7%, and the NSCE performance is poor and Sci., 15, 1147–1155. https://doi.org/10.5194/hess-15-1147- between –1.56 to 0.7 for the selected streamflow gauging sta- tions in the upper Blue Nile basin. This indicates that the uncor- Eisner, S., 2016. Comprehensive evaluation of the WaterGAP3 rected global WaterGAP3 streamflow product has an error in model across climatic, physiographic, and anthropogenic streamflow generation mechanisms to estimate the observed gradients. PhD thesis. University of Kassel, Kassel, streamflow for the local basin river network and needs bias Germany. Available at http://nbn- correction or adjustment for various applications at the regional resolving.de/urn:nbn:de:hebis:34-2016031450014 scale. However, not a lot has been done to demonstrate how El-Sadek, A., Bleiweiss, M., Shukla, M., Guldan, S., Fernald, these products can be ready for various local application by A., 2011. Alternative climate data sources for distributed reducing their errors. hydrological modelling on a daily time step. Hydrol. This research focuses on improving the quality of global Process., 25, 1542–1557. https://doi.org/10.1002/hyp.7917 streamflow data by bias-correction to make the product ready Elshamy, M.E., Seierstad, I.A., Sorteberg, A., 2009. Impacts of for diverse applications in the upper Blue Nile basin, Ethiopia. climate change on Blue Nile flows using bias-corrected Three spatiotemporal dynamical bias correction schemes (tem- GCM scenarios. Hydrol. Earth Syst. Sci., 13, 551–565. poral-spatial variable, temporal-spatial constant and spatial https://doi.org/10.5194/hess-13-551-2009 variable) tested in the twelve watersheds (streamflow gauging Erkossa, T., Awlachew, S., Haileslassie, A., Denekew Yilma, stations) of the basin. For the bias correction scheme and analy- A., 2009. Impacts of improving water management of sis, from the whole observed streamflow stations in the basin, smallholder agriculture in the Upper Blue Nile Basin. twelve stations that have a better quality of records selected for https://doi.org/DOI: 10.22004/ag.econ.212433 bias correction procedure using long term set of observed data Guetter, A.K., Georgakakos, K.P., Tsonis, A.A., 1996. from 1980 to 2010 at a daily temporal scale. Hydrologic applications of satellite data: 2. Flow simulation The bias correction applying BFTSV manifests significant and soil water estimates. J. Geophys. Res. Atmos., 101, improvement compared to the other schemes for the global 26527–26538. https://doi.org/10.1029/96JD01655 streamflow dataset. The NSCE improves by 30% and bias Haberlandt, U., Kite, G.W., 1998. Estimation of daily space– decreases by 19% for streamflow gauging stations that used for time precipitation series for macroscale hydrological the bias correction in the upper Blue Nile basin. From the total modelling. Hydrol. Process., 12, 1419–1432. stations, 33% of the stations score above 0.8, and 58% of the https://doi.org/10.1002/(SICI)1099- stations score above 0.7 of NSCE. Based on bias evaluation 1085(199807)12:9<1419::AID-HYP645>3.0.CO;2-A metrics, more than half of the stations score less than 5% of Habib, E., Haile, A., Sazib, N., Zhang, Y., Rientjes, T., 2014. bias in the corrected WaterGAP3 new set of daily gridded Effect of bias correction of satellite-rainfall estimates on streamflow product. The leave-one-out cross-validation runoff simulations at the source of the Upper Blue Nile. (LOOCV) approach applied for the bias correction and valida- Remote Sens., 6, Article Number: 6688. tion procedures. https://doi.org/doi.org/10.3390/rs6076688 These results reveal that the temporal-spatial variable Hay, L.E., Clark, M.P., 2003. Use of statistically and scheme is applicable and can use as one method for the bias dynamically downscaled atmospheric model output for correction to use the global data for local applications. The hydrologic simulations in three mountainous basins in the bias-adjusted new set of high resolution (0.25°) gridded stream- western United States. J. Hydrol., 282, 56–75. flow data applying BFTSV can be in use as one source of data for https://doi.org/10.1016/s0022-1694(03)00252-x various applications in the upper Blue Nile basin, Ethiopia. The Haileyesus Belay Lakew, Semu Ayalew Moges Hossain, F., Anagnostou, E.N., 2004. Assessment of current Remote Sens., 8, Article Number: 293. passive-microwave- and infrared-based satellite rainfall https://doi.org/https://doi.org/10.3390/rs8040293 remote sensing for flood prediction. J. Geophys. Res. Müller Schmied, H., Eisner, S., Franz, D., Wattenbach, M., Atmos., 109, Article Number: D07102. Portmann, F.T., Flörke, M., Döll, P., 2014. Sensitivity of https://doi.org/10.1029/2003JD003986 simulated global-scale freshwater fluxes and storages to Jasper, K., Gurtz, J., Lang, H., 2002. Advanced flood input data, hydrological model structure, human water use forecasting in Alpine watersheds by coupling meteorological and calibration. Hydrol. Earth Syst. Sci., 18, 3511–3538. observations and forecasts with a distributed hydrological https://doi.org/10.5194/hess-18-3511-2014 model. J. Hydrol., 267, 40–52. Nash, J.E., Sutcliffe, J. V., 1970. River flow forecasting through https://doi.org/https://doi.org/10.1016/S0022-1694(02)00138-5 conceptual models part I - A discussion of principles. J. Kanamaru, H., Kanamitsu, M., 2007. Fifty-seven-year Hydrol., 10, 282–290. https://doi.org/10.1016/0022- California reanalysis downscaling at 10 km (CaRD10). Part II: 1694(70)90255-6 Comparison with North American regional reanalysis. J. Nikolopoulos, E.I., Anagnostou, E.N., Hossain, F., Clim., 20, 5572–5592. https://doi.org/10.1175/2007jcli1522.1 Gebremichael, M., Borga, M., 2010. Understanding the scale Lachenbruch, P.A., Mickey, M.R., 1968. Estimation of error relationships of uncertainty propagation of satellite rainfall rates in discriminant analysis. Technometrics, 10, 1–11. through a distributed hydrologic model. J. Hydrometeorol., https://doi.org/10.1080/00401706.1968.10490530 11, 520–532. https://doi.org/10.1175/2009jhm1169.1 Lakew, H.B., Moges, S.A., Asfaw, D.H., 2017. Hydrological Schellekens, J., Dutra, E., Martínez-de la Torre, A., Balsamo, evaluation of satellite and reanalysis precipitation products G., van Dijk, A., Sperna Weiland, F., Minvielle, M., Calvet, in the Upper Blue Nile basin: A case study of Gilgel Abbay. J.-C., Decharme, B., Eisner, S., Fink, G., Flörke, M., Hydrology, 4, Article Number: 39. Peßenteiner, S., van Beek, R., Polcher, J., Beck, H., Orth, R., https://doi.org/https://doi.org/10.3390/hydrology4030039 Calton, B., Burke, S., Dorigo, W., Weedon, G.P., 2017. A Lakew, H.B., Moges, S.A., Anagnostou, E.N., Nikolopoulos, global water resources ensemble of hydrological models: the E.I., Asfaw, D.H., 2019. Evaluation of global water eartH2Observe Tier-1 dataset. Earth Syst. Sci. Data, 9, 389– resources reanalysis runoff products for local water 413. https://doi.org/10.5194/essd-9-389-2017 resources applications: Case study - Upper Blue Nile basin Setegn, S.G., Srinivasan, R., Melesse, A.M., Dargahi, B., 2010. of Ethiopia. Water Resour. Manag., SWAT model application and prediction uncertainty https://doi.org/10.1007/s11269-019-2190-y analysis in the Lake Tana Basin, Ethiopia. Hydrol. Process., Lakew, H.B., Moges, S.A., Asfaw, D.H., 2020. Hydrological 24, 357–367. https://doi.org/10.1002/hyp.7457 performance evaluation of multiple satellite precipitation Stone, M., 1977. An asymptotic equivalence of choice of model products in the upper Blue Nile basin, Ethiopia. J. Hydrol. by cross-validation and Akaike’s criterion. J. R. Stat. Soc. Reg. Stud., 27, Article Number: 100664. Ser. B, 39, 44–47. https://doi.org/10.1111/j.2517- https://doi.org/https://doi.org/10.1016/j.ejrh.2020.100664 6161.1977.tb01603.x Mei, Y., Nikolopoulos, E., Anagnostou, E., Zoccatelli, D., Borga, M., 2016. Error analysis of satellite precipitation- Received 31 January 2020 driven modeling of flood events in complex Alpine terrain. Accepted 25 September 2020

Journal

Journal of Hydrology and Hydromechanics – de Gruyter

Published: Mar 1, 2021

Keywords: Blue Nile; WaterGAP3; Bias Correction; Water Resource Reanalysis

There are no references for this article.