Comparison and evaluation of hydrologic performance of satellite, reanalysis and climate model-based precipitation products in the Upper Awash River Basin, Ethiopia

In remote areas where there are absent or limited rain gauge station networks, hydrological simulation is significantly lacking for the planning, designing and managing of water resources. A number of open-source precipitation products offer viable options for hydrological simulation as an alternative or as a supplement to station observations. Through diverse hydrological models to simulate streamflow, precipitation products derived from various data sources such as satellite, reanalysis, and global and regional climate models are crucial. Assessing and verifying the performance of these precipitation products in relation to precipitation measured from ground stations is essential to building confidence in these datasets for further usage (Hughes 2006; Javanmard et al. 2010; Adjei et al. 2015; Musie et al. 2019; Abdulahi et al. 2022; Ougahi & Mahmood 2022). Experiences in the past demonstrated that using precipitation data from ground measuring stations can generally provide high quality and accuracy of the hydrological modelling (Sharannya et al. 2020). However, direct and accurate networks of ground rain gauge measuring stations are unavailable due to sparse distribution, nonexistence, lack of high spatial resolution, political instability, orographic effect, harsh climate, rough topography and even in some developing countries lack of commitment to fund the hydro-meteorological stations because of many other immediate socio-economic problems (Michaels et al. 1998; Javanmard et al. 2010; Ougahi & Mahmood 2022). The growing availability of high-resolution satellite driven precipitation products, reanalysis, and global and regional climate model forecasting systems can help the hydrologists to obtain alternative precipitation products for streamflow modelling where conventional (ground measuring) hydro-meteorological stations are difficult to build (Bitew et al. 2012; Worqlul et al. 2014).

Precipitation is the most significant part of the water cycle balance and the primary input for hydrological modelling. Recently, different open-source precipitation products with high quality on the basis of spatial and temporal resolutions particularly from satellite, reanalysis and climate models are increasingly available with almost global coverage; their supplies are cost effective and, hence, they are alternative means of input sources for hydrological modelling (Adane et al. 2021). Thus, finding and adopting reliable precipitation products for quality hydrological modelling for the ungauged/limited data regions is an essential and even a controversial issue. Recently, different precipitation products with diverse temporal and spatial resolutions are available as input for the hydrological modelling where there is no adequate gauged streamflow for planning and management of water resources (Bitew et al. 2012; Reda et al. 2015; Babur et al. 2016; Gebremicael et al. 2017; Hu et al. 2017; Liu et al. 2017; Sharannya et al. 2020; Abdulahi et al. 2022; Ougahi & Mahmood 2022). The common ways of obtaining reliable and high quality precipitation products for accurate streamflow modelling could be by comparing and evaluating the predictive ability of different open-source precipitation products with the gauged precipitation data from ground stations and/or through hydrological modelling. Some of the precipitation product categories are: ground/in situ observation, ground-based radar, satellite, reanalysis and climate model-based. The most commonly available spatially and temporally distributed satellite and reanalysis precipitation forcing data for hydrological modelling are: Tropical Rainfall Measuring Mission (TRMM), Climate Forecast System Reanalysis (CFSR), the NOAA/Climate Prediction Centre, Precipitation Estimation from Remotely Sensed Information using Artificial Neural Network (PERSIANN), Climate Hazards Group Infrared Precipitation with Stations (CHIRPS), Modern-Era Retrospective Analysis for Research and Application (MERRA2), European Centre for Medium-Range Weather Forecast (ECMWF) Reanalysis-5 (ERA-5) (Javanmard et al. 2010; Worqlul et al. 2014; Adjei et al. 2015; Liu et al. 2017; Colston et al. 2018; Tolera et al. 2018; Bhattacharya et al. 2020; Nkunzimana et al. 2020; Adane et al. 2021; Ougahi & Mahmood 2022), and regional and global climate models like Canadian Centre for Climate Modelling and Analysis (CGCM), Commonwealth Scientific and Industrial Research Organisation (CSIRO), Meteorological Research Institute (MRI) Max-Planck Institute for Meteorology, Germany (ECHAM5), Regional Climate Model (RACMO22T) version 2.2 (Reda et al. 2015; Babur et al. 2016; Abdulahi et al. 2022). The results obtained from these precipitation products suggested various performance levels in hydrological modelling both temporally and spatially over diverse climate/topographic conditions. Despite their few uncertainties, these products are a promising alternative means of obtaining precipitation to fill the gaps in data-scarce regions to support the assessment, planning and management of water resources.

Numerous studies have been conducted using different hydrological models from lumped to fully distributed models using precipitation data from multiple products as an input for modelling of streamflow to identify the best precipitation products for hydrological modelling. Some of the common hydrological models that have been in use are: Hydrologiska Byråns Vattenbalansavdelning model (HBV), hydrological model (HYMOD), Artificial Neural Network (ANN)-based, Génie Rural à four paramètres Journalier (GR4J), Snowmelt Runoff Model (SRM), Simplified version of the HYDROLOG (SIMHYD), Soil and Water Assessment Tool (SWAT), Variable Infiltration Capacity (VIC), Hydrologic Engineering Center-Hydrologic Modeling System (HEC-HMS), Visualizing Ecosystem Land Management Assessments (VELMA) and Variant of Système Hydrologique Européen (MIKE SHE) (Bitew et al. 2012; Worqlul et al. 2014; Reda et al. 2015; Babur et al. 2016; Hu et al. 2017; Gebremicael et al. 2017;Liu et al. 2017; Colston et al. 2018; Tolera et al. 2018; Musie et al. 2019; Bhattacharya et al. 2020; Abdulahi et al. 2022; Ougahi & Mahmood 2022). However, they indicated different performance levels as their accuracy can be affected by geographical position, topography and climate, and the algorithms they are derived from. For instance, the study conducted by Ougahi & Mahmood (2022) using CFSR, CHIRPS, ERA5, MERR2, PERSIANN-CDR precipitation products to simulate streamflow suggested that ERA5 was the most reliable from the five precipitation products, while CHIRPS showed the worst performance both at daily and monthly scales. The study described in the works of Musie et al. (2019) to evaluate the hydrologic performance of SWAT using precipitation from CFSR, CHIRPS, PERCIANN-CDR and TRMM(3B427) concluded that, while all the other precipitation products performed well to capture gauged streamflow, CFSR showed the worst performance at daily scale. Conversely, CHIRPS showed a promising result in capturing streamflow modelling at both daily and monthly time scales. Similarly, Duan et al. (2016), who evaluated the capability of precipitation products from CHIRPS, CFSR and TRMM, to simulate streamflow with SWAT hydrological model, reported that CHIRPS performed better than the other (CFSR and TRMM) precipitation products. The simulated streamflow using Global Land Data Assimilation System (GLDAS) and PERSIANN-CDR precipitation products have similar good performance with simulated streamflow from gauge-based precipitation (Liu et al. 2017). A study conducted by Bhattacharya et al. (2020) concluded that simulated streamflow with the VIC hydrological model forced by ERA-Interim precipitation has a good agreement with observed streamflow while MERRA overestimated. Precipitation products with JRA-55 (Japanese 55-year Reanalysis), WFDEI (WATCH Forcing Data ERA Interim), and CFSR datasets underestimated the streamflow. CFSR was extremely poor in modelling streamflow compared to the other products especially at daily time scale.

Precipitation products from global and regional climate models with different downscaling systems are also promising alternative precipitation sources for hydrological simulations (Gleckler et al. 2008; Räisänen et al. 2010; Yira et al. 2017; Abdulahi et al. 2022). For instance Abdulahi et al. (2022) evaluated precipitation product from climate model (RACMO22T) under RCP4.5 and RCP8.5 as an input to HBV hydrological model to investigate the future streamflow availability. The result demonstrated that simulated streamflow with RACMO22T precipitation product in the baseline period was in good agreement with gauged streamflow, particularly after bias correction was applied. Precipitation data can be obtained from various sources including ground measuring gauges, satellite remote sensing, ground radar, reanalysis, or a combination of them. Each of these sources has its strengths and weaknesses, and their performance is region- and regime-dependent; hence, they need to be evaluated against available ground measuring station networks.

A number of studies were conducted and published regarding how well different open-source precipitation products capture gauged precipitation at different regions. For instance, Adane et al. (2021), Adjei et al. (2015), Bitew et al. (2012), Gebremicael et al. (2017) and Hordofa et al. (2021) compared the applicability of different satellite precipitation products to capture gauged precipitation products over various topographies. Similarly, precipitation products from reanalysis precipitation products were also evaluated to detect their hydrologic performance by Bhattacharya et al. (2020), Hu et al. (2017), Musie et al. (2019) and Tolera et al. (2018). On the other hand, some studies have been conducted by comparing between satellite and reanalysis precipitation products (Javanmard et al. 2010; Worqlul et al. 2014; Ougahi & Mahmood 2022). A few global and regional climate models have been investigated to study the availability of streamflow in the future under the climate change scenarios (Gleckler et al. 2008; Räisänen et al. 2010; Reda et al. 2015; Babur et al. 2016; Yira et al. 2017; Abdulahi et al. 2022). However, as far as we are aware, no studies have been conducted to examine the hydrologic performance of precipitation products from the satellite, reanalysis and climate model as an input to a single hydrological model (HBV) and compared their performance to each other, particularly in the Upper Awash River basin (UARB). Therefore, this study was intended to compare and evaluate the hydrologic performance of precipitation products selected from satellite, reanalysis and regional climate model using HBV hydrological model to gauged/recorded streamflow in the UARB. The selected precipitation products were CHIRPS from satellite products, CFSR from reanalysis product and RACMO22T from regional climate model. The choice of these datasets and reason for that choice are elaborated under the dataset section. The HBV model is selected for this particular study because of its popularity, immense application proficiency, user-friendliness, the ability to simulate streamflow in data-limited regions and relatively low data input requirement for simulation of streamflow as has been tested by many researchers (Abdulahi et al. 2022).

There are two approaches to evaluate the applicability of different precipitation products: (1) comparison and evaluation of precipitation products from different data sources with precipitation from ground rain gauge measuring stations; (2) using these precipitation products as an input into the hydrological models, and the outputs/simulated flows are compared and evaluated against recorded gauge-based streamflow (Worqlul et al. 2014; Meng et al. 2015; Ougahi & Mahmood 2022). The second approach was followed for this particular study. The result of this study would help and contribute to the selection of the best and most reliable precipitation products for streamflow modelling in the limited ground-based station networks and particularly in the UARB.

Daily precipitation data of the six meteorological stations were collected from Ethiopian National Meteorological Service Agency (ENMSA) for 24 years (1991 to 2013). These meteorological stations were selected based on their long record years, less missing data (<10%), and the extent of the catchment networks. Ground measured precipitation stations were used to extract other precipitation products (CFRS, CHIRPS and RACMO22T). Measured flow at the inlet of Koka reservoir was collected from Ethiopian Ministry of Water and Energy office (MoWE) for 14 years (2000–2013). This stream flow was used to calibrate and validate HBV hydrological model and used as a reference for hydrologic performance of considered precipitation products. The details of selected ground measured precipitation stations are presented in Table 1.

The most precise data comes from individual rain gauges and gauge networks since precipitation events are highly localized, especially when they are severe. Yet, they suffer from sample measurement, under-catch problems, and geographic sparsity, particularly in remote areas. Satellite, reanalysis and climate model-based precipitation products mitigate some of the shortcomings of the gauges (Sharannya et al. 2020).

CHIRPS (Climate Hazards Group Infra-Red Precipitation with Station data) Version 2 (CHIRPS-2.0) is an open-source precipitation dataset available at daily and monthly time resolution and incorporates fine (0.05° × 0.05° to 0.25° × 0.25°) spatial resolution. CHIRPS data set was developed to provide comprehensive, dependable, and current data sets for various early warning goals, such as trend analysis and seasonal drought monitoring. It was developed for monitoring drought-sensitive areas and also tested in countries like Ethiopia. It is available from 1981 to the near present (Musie et al. 2019; Sharannya et al. 2020; Hordofa et al. 2021). CHIRPS dataset is constructed from the NOAA Climate Forecast System (CFS), TRMM-3B42V7 data and monthly precipitation climatology data (CHPclim) and rain gauge stations (Hordofa et al. 2021; Ougahi & Mahmood 2022). The spatial and temporal coverage and resolution in CHIRPS provide the ability to study hydroclimate, precipitation extremes, and droughts, especially in historically data-sparse regions such as the Caribbean and Africa (Sharannya et al. 2020; Ougahi & Mahmood 2022). Currently, CHIRPS does not have coverage outside of 50°S–50°N) and is constructed using ground observed stations. Therefore it has limitations outside this range and ground station data-sparse regions. Despite its limitations, CHIRPS rainfall product agreed well with the benchmark rainfall patterns of different topographies. CHIRPS precipitation product, which is in the satellite gauge category with the 0.05 × 0.05 degree spatial resolution, is freely available and downloaded from (ftp://ftp.chg.uscb.edu/pub/org/chg/product/CHIRPS-2.O/) and used for analysis for this study.

CFSR global weather data was chosen as an alternative to the ground-based conventional weather data as an input for the hydrological model to evaluate the applicability and performance in modelling the streamflow (Tolera et al. 2018). Data assimilation systems that integrate available observations (both in-situ and remotely sensed data) into numerical models provide useful reanalysis outputs (Sharannya et al. 2020). The ground observations and satellite rainfall product are merged to output product with 38 by 38 km spatial resolution and daily time resolution at a global scale (Ougahi & Mahmood 2022). Reanalysis products depend on the observational restrictions for their dependability, even though they offer an extended temporal estimation of precipitation. These constraints might change greatly over time and place (Moriasi et al. 2013). CFSR is derived from the Global Forecast System that was developed in the National Centre for Environmental Prediction (NCEP). One of the great advantages of reanalysis product is that it was developed by combining historical observations going back several decades, with model-based global estimates using data assimilation provide near real-time (Bhattacharya et al. 2020). Because of its great temporal and spatial resolution, the CFSR precipitation product is the most extensively utilized precipitation product used for hydrologic modelling. SWAT formatted file at high spatial and longer time series is widely and readily available and has been used in hydrological modelling. CFSR precipitation data which is freely available was downloaded from the web (https://swat.tamu.edu/data/cfsr/) from the period of 1979 to 2014. The accuracy of CFSR dataset has been improved by including the coupling of the atmosphere-ocean-land surface-sea ice system (Bhattacharya et al. 2020).

Regional Climate Models (RCMs) were widely used due to high resolution and resolved information, and their ability to model atmospheric processes and land cover changes explicitly, wider range of variables, and better representation of some weather extremes than in Global Climate Models (GCMs) (Yira et al. 2017). The regional climate model (RACMO22T) version 2.2 was downloaded from the African CORDEX domain for the period of 1991–2013. CORDEX promotes international downscaling coordination, and facilitates easier analysis by scientists and end-user communities at the local level of regional climate changes with many methods of RCMs for Africa. The daily time series precipitation data of RACMO22T was downloaded from the web (http://cordexesg.dmi.dk/esgf-web-fe/). The selection of this climate model was based on vintage, resolution, validity and the ability to capture observed precipitation in the baseline period (Gleckler et al. 2008; Rojanamon et al. 2009; Räisänen et al. 2010; Yira et al. 2017) The performance of RACMO22T climate model precipitation output in capturing the gauged precipitation product was improved after bias correction is applied (Abdulahi et al. 2022). The detailed description of RACMO22T R|M and bias correction was presented by Abdulahi et al. (2022).

HBV is a semi-distributed conceptual hydrological model which was developed in the Swedish Meteorological and Hydrological Institute (SHMI) in 1970s. It is used to forecast and simulate streamflow throughout the world including Ethiopia. The advantages of HBV model over the other hydrological models are the possibility of disaggregating the basin into a number of sub-basins, elevation and vegetation zones and relatively less data required for input for streamflow simulation (Seibert 2011; Gebre et al. 2015; Gragn et al. 2019). The required input parameters for HBV model are: daily areal rainfall (mm), daily average temperature for areas with snow cover (°C) and long term daily/monthly potential evapotranspiration (mm) (Abdulahi et al. 2022). Streamflow is also required for the calibration and validation of the model at the outlet. For this particular study, Hargreaves method was used to estimate potential evapotranspiration because of its simplicity and because it requires relatively less data input compared to the other methods like Penman (Hargreaves & Riley 1985; Nadrah et al. 2012). Detailed descriptions of the HBV hydrological model and Hargreaves method were presented in the work of Abdulahi et al. (2022) and Seibert (2011).

The predictive ability, reliability and accuracy of hydrological model simulation should be evaluated before further usage for different purposes (Rientjes 2007). HBV hydrological model parameters during simulation were calibrated and validated with gauged daily or monthly streamflow along with daily or monthly gauged precipitation as input. Five to ten years of streamflow is sufficient to calibrate and validate HBV model (Seibert 2011). A number of statistical parameters were used by the scholars and researchers to evaluate the performance of the hydrological models (Bitew et al. 2012; Adjei et al. 2015; Tolera et al. 2018; Hordofa et al. 2021). In this study, coefficient of determination (R²) and Nash–Sutcliffe efficiency (NSE) were used to evaluate the performance of HBV model (Krause & Boyle 2005). Nash–Sutcliffe efficiency indicates how well the plot of the observed value versus simulated value fits the 1:1 line and ranges from [−∞ to 1] with the higher values indicating less error variance or better agreement. Values between 0 and 1 are generally viewed as acceptable levels of performance while values <0 indicates unacceptable performance. NSE performance ratings are classified as unsatisfactory (≤0.50), satisfactory (0.50 < to ≤ 0.65), good (0.65 < to ≤ 0.75) and very good (0.75 < to ≤ 1.0) (Moriasi et al. 2013; Tolera et al. 2018). NSE was also found to be the best objective function for reflecting the overall fit of a hydrograph and less sensitive to high extreme values due to the squared differences (Moriasi et al. 2013). The strength of linear correlation between simulated and observed streamflow is determined by R² and ranges from 0 to 1. The best simulation is considered close to 1, while its values of >0.5 are generally acceptable (Krause & Boyle 2005; Moriasi et al. 2013). The model performance ratings based on R² values are classified as unsatisfactory (≤0.50), satisfactory (0.50 < to ≤ 0.60), good (0.60 < to ≤ 0.70) and very good (0.70 < to ≤ 1.0).

For this study, from a total of 14 years (2000–2013) streamflow data, two years (2000–2001) was used for warmup period, seven years (2002–2008) for calibration and five years (2009–2013) for validation.

Precipitation products derived from different data sources such as satellite, reanalysis and climate models are essential in hydrological model simulations where ground measured precipitations are unavailable. The performances of such products are detected mainly using two approaches: (1) comparing the performance of each precipitation obtained from these data sources with the long years precipitation from in-situ or station recorded daily, monthly, and quarterly and even annually; (2) assessing the performance of each of the precipitation products through hydrological modelling as an input data. The second approach is followed in this particular study. The procedure is summarized as follows:

• 1. Daily areal precipitation from six ground measured stations from 2000–2013 was fed into HBV hydrological model as an input and simulate streamflow keeping the outlet at the inlet of Koka Reservoir/lake (Figure 1).

• 2. HBV hydrological model was calibrated and validated by the measured flow at the inlet of Koka reservoir at monthly time series. The predictive ability of the HBV model was investigated by coefficient of determination (R²) and Nash–Sutcliffe efficiency (NSE).

• 3. After successful calibration and validation of HBV model, each precipitation product from CFSR, CHIRPS and RACMO22T was plugged as an input to HBV model and flow was simulated for each precipitation product separately keeping the optimum parameters during calibration constant.

• 4. Finally, the performance of simulated flow with each precipitation product was compared and quantified against measured flow through graphical plots and statistical indices (R², NSE, RMSE and PBIAS).

The two most commonly used graphical techniques to compare simulated with measured flow are hydrographs and percentage of exceedance probability curves and provide a first overview of model performance (Moriasi et al. 2013). The hydrographs indicate how simulated flow captures the measured flow especially low and high flow. On the other hand, flow duration curves (FDC) show how well the simulated flow frequency reproduces the frequency of measured flow (Rojanamon et al. 2009; Babur et al. 2016; Abdulahi et al. 2022). The FDC can be applied to study hydropower, water resources management, and low and high flow management for drought and flood management respectively.

Root mean square error (RMSE) and percentage of bias (PBIAS) are the statistical parameters used to examine the hydrologic performance of precipitation products in addition to coefficient of determination (R²) and Nash–Sutcliffe efficiency (NSE) as was suggested by Krause & Boyle (2005). R² and NSE were discussed in the Model calibration and validation section. PBIAS measures the average tendency of the model simulated flow to be smaller or larger than measured flow at a common outlet point for the same period of time. Its value ranges from (−∞ to +∞). Under optimal conditions PBIAS is 0, where positive value indicates overestimation and negative shows underestimation. PBIAS is the deviation of data being evaluated and expressed as a percentage (Adjei et al. 2015; Babur et al. 2016). Performance rating of PBIAS is classified as very good (≤± 10%), good (±10% < to ≤ ±15%), satisfactory (±15% < to ≤ ±25%) and unsatisfactory (>± 25%) (Moriasi et al. 2013; Adjei et al. 2015; Babur et al. 2016). RMSE is also another important and commonly used error index statistic. It computes the standard deviation of the model prediction error (difference between measured and simulated values). The lower the RMSE value, the better model performance, and vice versa. RMSE varies from the optimal value of 0, which indicates zero residual variation and therefore perfect model simulation, to a large positive value (Moriasi et al. 2013). Recommended RMSE performance ratings for both daily and monthly flow are: (0–49) very good, (50–59) good, (60–70) satisfactory, and (>70) unsatisfactory (Tolera et al. 2018). A summary of the statistical measures used to check the level of hydrologic performance of the precipitation products is presented in Table 2.

Another very important statistical parameter used to check the hydrologic performance of precipitation products is NSE, which indicates how well the plot of measured value fits the simulated value. NSE varies from 0.43 to 0.71 at daily time series and 0.52 to 0.98 at monthly time series. In terms of NSE, CFSR performance was unsatisfactory (NSE = 0.43), which is <0.5 at daily time series, whereas RACMO22T performance was very good (NSE = 0.7) at daily time scale (Figure 7(b)). CHIRPS showed good performance (NSE = 0.63) at daily time series but performed worse than RACMO22T. On monthly scale, the performance of all precipitation products in hydrologic simulation in terms of NSE was very good (NSE > 0.7) except CFSR (NSE = 0.52) which was satisfactory. CHIRPS was the second best precipitation product next to RACMO22T in terms of NSE. Like R² there was a significant improvement of NSE from daily to monthly time scale (Figure 7(b)).

PBIAS and RMSE are other important statistical indices used to examine the hydrologic performance of precipitation data. Simulated flow with CFSR, CHIRPS and RACMO22T resulted in positive PBIAS which means overestimated measured flow at various levels. The simulated flow with RACMO22T was good in terms of PBIAS (<± 20%) while CHIRPS and CFSR showed unsatisfactory performance (PBIAS > ±25) at both daily and monthly time scale (Figure 7(c)). Extremely large values of PBIAS (>± 180%) and (>± 75%) were seen for CFSR and CHIRPS respectively which showed worst performance. Generally the hydrologic performance of RACMO22T precipitation product showed satisfactory results whereas CFSR and CHIRPS showed unsatisfactory results in terms of PBIAS. Similarly, CFSR showed the highest RMSE both at daily and monthly scale. Greater magnitude (RMSE) error was exhibited on monthly scale. The lowest RMSE was registered for simulated streamflow from RACMO22T precipitation product. The RACMO22T performance was in the very good ranges in terms of RMSE (21.8% and 17.3%) at monthly and daily scales, respectively, compared to the other precipitation products. Generally, the hydrologic performance of RACMO22T precipitation product was considered to be the best in terms of all statistical indices (R² > 0.50, NSE > 0.70, −20 <PBIAS < +20%, and RMSE < 20).

The findings show that streamflow in the UARB may be predicted by the HBV model more accurately utilizing climate model (RACMO22T) precipitation data than satellite (CHIRPS) and reanalysis (CFSR) precipitation products. We have applied linear bias adjustment to the RACMO22T precipitation product prior to using it as an input for the HBV model for simulation, which could be one of the reasons why it was better in the performance. Applying bias correction to climate model-based precipitation products has been studied by Abdulahi et al. (2022) and Yira et al. (2017). The result shows that applying bias correction greatly improved precipitation data to capture the precipitation from ground stations. When compared to gauged flow, the hydrologic performance of the satellite-based precipitation (CHIRPS) product performed marginally better than the reanalysis precipitation (CFSR) product. Hydrologic performance of satellite-based precipitation data is more reliable in flat regions than mountainous because in the mountainous regions satellite products use passive microwave or infrared sensors which mostly fail to detect high rainfall over complex topography (Ougahi & Mahmood 2022). Similarly, a study conducted on a flat region by Sharannya et al. (2020) revealed that the satellite-based precipitation product in the flat regions has better performance than reanalysis (CFSR) in modelling flow which is in agreement with our result. Among all the precipitation products, only RACMO22T precipitation data qualified for hydrologic performance based on all four evaluation indices (R² > 0.50, NSE > 0.70, PBIAS < ±20% and RMSE < 20). The result implied that, based on the evaluation criterion applied, RACMO22T was considered to be the viable option for hydrologic modelling in the UARB. The hydrologic performance of all precipitation products was very good based on (R² > 0.70) and (NSE > 0.75) at monthly scale except CFSR (NSE = 0.52) which is in the satisfactory range. The hydrologic performance of the CHIRPS and CFSR were unsatisfactory based on PBIAS (>+ 65) both at daily and monthly scale as recommended by Adjei et al. (2015) and Babur et al. (2016). CFSR shows extremely large value of PBIAS (>+ 180) both at daily and monthly scale compared with CHIRPS. Hydrologic performance of all precipitation data based on the RMSE evaluation indices were in the very good range (<50) both at daily and monthly scales, except for the CFSR (=91.8) which is unsatisfactory as suggested by Moriasi et al. (2013) and Tolera et al. (2018).

Figure 8 depicts the simulated FDC for all precipitation products at the outlet of the UARB at a monthly time scale. It is observed that simulated flow with RACMO22T precipitation follows the same patterns as that of the gauged/measured flow (Figure 8(b)). This means that RACMO22T precipitation product is capable of simulating high, medium and low flow for UARB since the FDC exactly corresponds to observed flow FDC with a little deviation. The same finding was reported by Abdulahi et al. (2022). On the other hand, regarding the CHIRPS and CFSR precipitation products, it is difficult to judge one dataset as the best; rather it should be based on the application of the study. For example, to simulate high flows, CHIRPS may be suitable, whereas for low flow CFSR is better than CHIRPS (Figure 8(c) and 8(d)). Minimum flow corresponds to Q₉₅ and Q₁₀₀ whereas a high flows/extreme flow event corresponds to Q₁₀ and Q₂₀. Q₅₀ represents medium or average flow. CHIRPS rainfall product was capable of capturing high flows with a little deviation, and underestimated low flow by far. A Report by Sharannya et al. (2020) also reveals the same kind of situation. CFSR overestimated high flow and underestimated low flow (Figure 8(d)). CHIRPS and CFSR showed nearly the same kind of medium flow (Q₅₀) distribution. The differences observed between the simulated flow for precipitation estimates and measured flow could be attributed to a variety of agents, such as: errors in precipitation predictions because of high temporal and spatial variability, resolution, and algorithm; errors in stream gauge observation; errors in model structure/setup; and number of data years and model runs during calibration and validation. In general, considering the uncertainties, the result implied that the hydrologic performance of regional climate model-based precipitation product (RACMO22T) means that it is a promising alternative compared to the satellite (CHIRPS) and reanalysis (CFSR) precipitation products in the UARB. To determine the most promising precipitation products for the management and mitigation of water resources, additional research using a variety of regional climate models and an ensemble of them will be necessary in order to increases confidence in the results obtained.

The present work examined the hydrologic performance of open-source precipitation estimates particularly CHIRPS, reanalysis (CFSR) and regional climate model (RACMO22T) through HBV hydrological model in the UARB. First, precipitation measured from rain gauges stations was used as an input to HBV model to simulate the flow. Then the predictive ability/performance of HBV was evaluated against gauged flow at the outlet through calibration and validation. The optimum parameters obtained during calibration of HBV were kept constant during simulation of flow with each precipitation estimate (CHIRPS, CFSR and RACMO22T). Finally, the simulated flow from each precipitation estimate was compared and evaluated against gauged flow at the outlet of the basin to examine their hydrologic performance through statistical indices (R², NSE, RMSE and PBIAS) and graphical visualization at daily and monthly scales. The result indicated that the performance indicators (R², NSE, RMSE and PBIAS) were within the acceptable ranges for all precipitation estimates even though their performance levels were different, except PBIAS for CFSR and CHIRPS estimates. The performance of the RACMO22T precipitation product was ranked the first based on all performance indicators. It is observed that the performance of precipitation data widely differs based on the statistical indices. The performance indicators (R², NSE, PBIAS and RMSE) were in the ranges of 0.5–0.72, 0.43–0.72, +11.1– + 182, and 11.4–38.8 at daily scale and 0.81–0.90, 0.52–0.98, +10.5– + 207.7, and 14.1–91 at monthly scale, respectively. The hydrologic performance of these precipitation products was significantly improved from daily to monthly scales. The hydrologic performance of the CHIRPS and CFSR were unsatisfactory based on PBIAS values at both daily and monthly scales. And, CFSR was unable to detect measured flow based on the RMSE. For instance, all precipitation datasets show very good to satisfactory performance based on R² and NSE. Only RACMO22T precipitation estimates qualified for hydrologic performance based on all four evaluation indices used in the UARB.

In addition to the statistical parameters, the flow duration curve (FDC) simulated by RACMO22T precipitation product has the same patterns as that of the gauged/measured flow duration curve which confirms that the RACMO22T precipitation estimates are better in predicting flow in the UARB. This means that RACMO22T precipitation product is capable of simulating high flow (Q₁₀ and Q₂₀), medium (Q₅₀) and low flows (Q₉₅ and Q₁₀₀) for UARB. Regarding the CHIRPS and CFSR precipitation products, it is challenging to choose a single dataset as the best option; instead the purpose and the application of the study should be considered to select the best precipitation estimates for further application. For instance, CHIRPS might be appropriate to simulate high flows, whereas CFSR performed better in low flow modelling scenarios.

This result implied that, overall, the hydrologic performance of climate model (RACMO22T) based precipitation estimates using HBV model would be of great benefit considering the difficulties in accessing data across and in similar basins compared to the satellite (CHIRPS) and reanalysis (CFSR) precipitation estimates. However, to build more confidence in the results obtained, further study will be required by employing a number of climate models and even their ensembles.

We are grateful to the Ethiopian Ministry of Water and Energy (MoWE) and National Meteorology Service Agency (NMSA) for providing hydro-meteorological and GIS data. We thank the Climate Forecast System Reanalysis (CFSR) centre and Climate Hazards Group Infrared Precipitation with Station (CHIRPS) centre for providing the essential and valuable data for this study.

All relevant data are included in the paper or its Supplementary Information.

The authors declare there is no conflict.