ABSTRACT
Remote sensing contributes valuable information to streamflows. Usually, streamflow is directly measured through ground-based hydrological monitoring stations. However, in many countries like Ethiopia, ground-based hydrological monitoring networks are either sparse or nonexistent. The lack of reliable in situ observational data severely limits our ability to manage water resources in a well-informed way in these regions. In such cases, satellite remote sensing is an alternative means of acquiring such information. In this study, the application of remotely sensed rainfall data in streamflow modeling for the Gilgel Ghibe catchment in Ethiopia is reported. Ten years (2001–2010) of satellite-based-precipitation-products (SBPPs) from Tropical Rainfall Measuring Mission (TRMM) and WaterBase, were combined with a PyTOPKAPI model to generate daily streamflows. We compared the results with that of the observed streamflows at the Gilgel Ghibe Nr, Assendabo gauging station using four statistical tools (bias, R2, NS, and RMSE). The result indicates that the bias-adjusted SBPPs agree well with gauged-rainfall. Without bias-adjustment, the SBPPs tend to overestimate the simulated streamflows. We further conclude that the general streamflow patterns were well captured at daily time scale from SBPPs after bias-adjustment. However, the simulated streamflow using gauged-rainfall is superior to those obtained from SBPPs including bias-adjusted ones.
HIGHLIGHTS
Streamflow information is vital for peak flood estimation. Such information can be obtained either through ground-based monitoring station or by streamflow modeling.
Precipitation data are the key ‘input’ to streamflow modeling.
The SBPPs are viable sources of precipitation data for data limited areas. Therefore, this study explores the reliability of SBPPs for streamflow modeling in Ethiopia.
INTRODUCTION
The development of hydrological models has resulted from accumulating information about hydrological modeling and aligning it with the various development needs in both applied research and operational domains (Ciarapica & Todini 2002; Liu et al. 2005). These models can be used for hydrological research, water resources engineering, and environmental applications (Mathevet et al. 2006). Streamflow modeling from a catchment using such hydrological models is a current field of active research. The importance of streamflow modeling ranges from the assessment of the impacts of long-term climate or land use change to operational forecasting of streamflows for flood prediction and other similar water resources applications. It also provides a conceptual framework of the physical processes operating within watersheds at a variety of scales and how those processes are interlinked (Bourdin et al. 2012). Many authors have provided detailed descriptions of modeling approaches which are categorized from an empirical to a physical model structure; lumped to fully distributed (spatial-based); and event to continuous (time-based) models (Singh & Woolhiser 2002; Pechlivanidis et al. 2011; Bourdin et al. 2012). Normally, the selection of a suitable model is dependent on the aim of the project and data availability. After that, preliminary simulations are conducted, and then the model calibration is performed to achieve a close imitation of the real hydrological regime of the watershed based on in-situ observational data. Validation takes place after calibration using independent in-situ observational data in order to evaluate the model performance (Bourdin et al. 2012).
The paper is organized as follows. Section 2 describes Materials and Methods, while section 3 presents Results and Discussion. Section 4 summarizes the Conclusions and Recommendations of this study.
MATERIALS AND METHODS
Case study area
Schematic representation of the PyTOPKAPI model (Lastoria 2008; Todini 2011).
Brief description of the PyTOPKAPI model
The PyTOPKAPI model is the acronym of the ‘Python (http://www.python.org/)-based TOPographic Kinematic APproximation and Integration’ hydrological model. It is an improved version of the former physically-based fully-distributed TOPKAPI hydrological model that has been applicable at various spatial scales, ranging from the hillslope to the basin scale without missing the physically-based meaning of the model parameters (Sinclair & Pegram 2013a, 2013b). It was developed by combining kinematic wave formulations of flows in the soil, over the land and in the channel (Ciarapica & Todini 2002) that resulted in converting the rainfall–runoff processes into three ‘structurally-similar’ zero-dimension non-linear reservoir differential equations (Figure 3) describing the different hydrological processes. The parameterization of the model is relatively simple and parsimonious. Furthermore, the parameters of the model are scale-independent and can be obtained from digital elevation model (DEM), soil grid, and land use maps. The precise integration of the differential equations can yield physically-based models that are largely scale-independent and retain the physical meaning of the model parameters (Liu & Todini 2002). It is an open-source with BSD license, coded in the Python programming language, accessed directly through an interactive Python environment, and runs on most popular operating systems (Sinclair & Pegram 2013a, 2013b). Since it is a physically-based fully-distributed model, it can be likewise appropriate for streamflow modeling from ungauged catchments using the model parameter values obtained from literature and prevailing thematic maps (Liu & Todini 2002; Sinclair & Pegram 2013a, 2013b, Rabba et al. 2018).
PyTOPKAPI model application
Model set-up
The PyTOPKAPI model application starts with pre-processing of the DEM and preparing model parameters, which are then used in the model to simulate reasonable streamflow data along the drainage system. In this study, the model was applied to Gilgel Ghibe catchment at its flow outlet defined by Ghibe Nr Assendabo gauging station in Ethiopia. For simulating steam flows from the catchment, the model was set up utilizing the input data of the study catchment. The model set-up steps are as mentioned in the following.
1. The DEM of the study area was loaded into the GIS, and then it was re-projected to its respective UTM zone.
2. The DEM was pre-processed to eliminate the false outlets and the sinks so that the flow direction and the basin Closure Cell is uniquely identified.
3. As per Todini's recommendation, a threshold area of 25 km2 was used to define the stream network (Sinclair & Pegram 2013a).
4. The location of the outlet (pour point) was chosen with care. The total watershed of the catchment was then delineated.
5. By the same token, the land use and soil grid maps of the study catchment were loaded to GIS and then they were extracted by the defined watershed as a mask.
6. The values of the literature parameters were also added to the attribute tables for each map.
7. The necessary GIS files for generating model parameters were then created.
8. Next, the cell parameters were generated using Python scripts and consequently modified to eliminate zero slopes.
9. After that, the rainfall and evapotranspiration (ETo) data were converted to HDF5 forcing files using the relevant Python scripts.
10. Finally, after the model input files were successfully prepared and introduced to the model, the model simulated streamflows for simulation period.
Model input data
The PyTOPKAPI model requires input data such as DEM, soil grid, land use, and hydro meteorological data (observed streamflow, rainfall records, and ETo or air temperature). Such data of the study catchment would be used to build the model parameters and forcing files. Brief descriptions of these data are presented hereunder.
Digital elevation model
The Shuttle Radar Topographic Mission (SRTM) DEM (Jarvis et al. 2008) was used for this study. The SRTM DEM data, produced by NASA originally, is a major breakthrough in digital mapping of the world, and provides a major advance in the accessibility of high quality elevation data for large portions of the tropics and other areas of the developing world. The NASA SRTM has provided digital elevation data (DEMs) for over 80% of the globe. The resolution of the DEM was set to 1 km and this resolution has been respectively adopted for all the terrain maps.
Evapotranspiration
The ETo data used were obtained by taking the average value of ETo computed by the methods of Blaney–Criddle (Blaney & Criddle 1962) and Thornthwaite (Thornthwaite & Mather 1955).
Land use map
The land use map was obtained from the United States Geological Survey (USGS) Global Land Cover Characterization (GLCC) database and is accessible through the SWAT web page (https://swat.tamu.edu/data/). These maps are available in the form of zip files containing one or more tiles for each continent/region. They are in two resolutions: the original at approximately 400 m (at the equator) and the resampled at 800 m. For this study, the ‘400 m resolution’ was used as it is finer. The PyTOPKAPI model requires the values of the Manning roughness coefficient for all grid cells for each land use class. These parameters were obtained from the land use grids data of USGS Land Use/Land Cover System Legend-Modified Level 2 (GLCC 2008) and the table of the Manning's roughness values used for various land cover classes in GeoSFM (Asante et al. 2008).
Soil data
The soil data were obtained from the Harmonized World Soil Database (HWSD) (FAO/IIASA/ISRIC/ISS-CAS/JRC 2012). The HWSD is composed of a raster image file linked with attribute database. The resulting raster database consists of 21,600 rows and 43,200 columns, which are linked to harmonized soil property data.
Precipitation data
Precipitation data is a key ‘input’ to any hydrological model (Kite & Pietroniro 1996; Liu et al. 2014). In areas of sparse or nonexistent rain gauge distribution, satellite RS is the alternative method of providing precipitation data to perform streamflow modeling. It is a tool to help alleviate some of the hydrometric data collection and management problems facing many of the developing nations (Kite & Pietroniro 1996). Generally, RS offers the components of such information in digital form for streamflow modeling with the advantage of providing continuous, huge coverage and free or low cost data to users. Given that rainfall is the most vital input parameter that directly influences the result of simulation (Kite & Pietroniro 1996; Tapiador et al. 2012), this section concerns aspects of rainfall in the streamflow modeling. More recently, several efforts have been directed on the use of widely available SBPPs to complement in situ hydrologic observations over vast ungauged regions (Khan et al. 2012). There are numerous freely available global satellite-based precipitation products (SBPPs) that can be used for streamflow modeling (Tapiador et al. 2012). Various studies proposed the optimal use of SBPPs. The advantage of these products is the global availability over regions where ground networks are nonexistent (Khan et al. 2012). In this study, we used two SBPPs (Tropical Rainfall Measuring Mission (TRMM) and Waterbase) as ‘input’ for the model.
Tropical Rainfall Measuring Mission
The TRMM is a joint project between NASA and the Japanese space agency (JAXA) which was launched in November, 1997 (Huffman et al. 2007) with the aim of providing global rainfall observations that can be used for scientific studies. The TRMM Multi-satellite Precipitation Analysis (TMPA) is a new SBPP dataset designed to combine precipitation estimates from various satellite systems, as well as land surface precipitation gauging stations when possible and provides calibrated precipitation estimates at spatial scales of 0.25° and at 3-h time intervals (Huffman et al. 2007; Almazroui 2011). TMPA products are available in two versions: post-real-time version (3B42) and near-real-time version (3B42RT), based on calibration by the TRMM Combined Instrument and TRMM Microwave Imager precipitation products, respectively. Only the post-real-time version (3B42) incorporates gauge data at the present (Tan et al. 2014). The dataset covers the latitude of 50°N–50°S for the period from 1998 to the present. Finer-scale TMPA is successful at approximately reproducing the surface observation–based histogram of precipitation, as well as reasonably detecting large daily events (Huffman et al. 2007).
Various authors have described that the TRMM rainfall dataset is a promising way of obtaining precipitation data for streamflow modeling. Duan & Bastiaanssen (2013) found that TRMM3B43 and TRMM3B42 data were reliable in the Caspian Sea Region in Iran for most months and years during the period 1999–2003. Dinku et al. (2007) also evaluated performance of 10 different satellite rainfall data sets and categorized them into two main group: (1) low spatial (2.5°) – low temporal (monthly), and (2) high spatial (0.1°–1°) – high temporal (3 hourly to 10 daily) resolutions over Ethiopia in Africa and found that TRMM3B43 from the first group, and TRMM3B42 from the second group, performed reasonably well. Similarly, Behrangi et al. (2011) evaluated the performance of several SBPP [TRMM Multi-satellite Precipitation Analysis real-time (TMPA-RT); TMPA bias-adjusted (TMPA-V6); Precipitation Estimation from Remotely Sensed Information Using Artificial Neural Network (PERSIANN); PERSIANN bias-adjusted (PERSIANN-adj); and Climate Prediction Center morphing algorithm (CMORPH)] for streamflow modeling over Illinpis River Basin in the United States. The results indicated that all products generate good simulations at sub-daily and monthly time scales.
Khan et al. (2011) conducted a study in the Nzoia basin of Lake Victoria in Africa to integrate the best available satellite products with a distributed CREST hydrologic model for characterizing the extent of flooding over sparsely gauged or ungauged basins. The study presented a methodology based entirely on satellite RS data to calibrate a hydrologic model, simulate the extent of flooding, and evaluate the probability of detecting inundated areas. They used satellite-based data sets (SRTM DEM, TRMM rainfall, FAO soil, MODIS land use/cover, and FEWS NET ETo). They found that RS data can be used with sufficient accuracy in predicting flood extents and, hence, potentially improve flood management strategies in ungauged basins.
Khan et al. (2012) also developed a novel framework by integrating microwave satellite RS along with rainfall estimates from TRMM as a key forcing data set into a distributed hydrologic model for flood early warning in data-poor regions with case studies in the Okavango Basin in Southern Africa. This proof-of-concept study demonstrated the efficacy of satellite RS data for flood prediction in poorly gauged basins.
Pombo et al. (2015) carried out a study to identify the RS product that offers the best monthly precipitation estimates for Angola. Four (TRMM, the Global Precipitation Climatology Project (GPCP), Precipitation Estimation from Remotely Sensed Information using Artificial Neural Networks (PERSIANN), and the Climate Prediction Centre (CPC) Morphing Technique (CMORPH)) products were examined by comparing monthly precipitation estimates with ground observation measurements. They found that TRMM offers the most consistent and accurate estimates of monthly precipitation for Angola, probably because it is the first product to include measurements from precipitation radar and because it uses in situ ground measurements from monitoring stations.
According to Huffman et al. (2007), the ‘near-real-time’ product makes the estimates useful to several new classes of users. Consequently, we used the daily 0.25° × 0.25° near-real-time version of TRMM precipitation products (TRMM3B42RT) (Islam & Uyeda 2007; Scheel et al. 2011) that can freely be assessed and obtainable through the GES-DISC Interactive Online Visualization And analysis Infrastructure (Giovanni) as part of the NASA's Goddard Earth Sciences (GES) Data and Information Services Center (DISC) (Immerzeel et al. 2009). The data could be obtained from the TRMM website at http://trmm.gsfc.nasa.gov or https://gpm.nasa.gov/missions/trmm.
Waterbase-Satellite-Precipitation Product
This is the second SBPP that could be obtained with other weather data such as temperature (°C), wind speed (m/s), relative humidity (fraction) and solar radiation (MJ/m2) from the Waterbase project, and is currently accessible through the SWAT web page (https://swat.tamu.edu/data/). These weather data are available from 1/1/1979 to 7/31/2014. For this study, the remotely sensed rainfall data for a 10-year period (2001–2010) were used.
The reference daily-observed precipitation data were obtained from Ethiopian National Meteorological agency from 1986 to 2010 for the three (3) rain gauging stations (Jimma, Asendabo and Yebu) located within the Gilgel Ghibe basin. The daily streamflow observation data at the basin's outlet (Ghibe NrAssendabo) was also obtained from Ministry of Water and Energy in Ethiopia.
Model performance evaluation
In order to quantitatively analyze the performance of TRMM3B42RT and WaterBase precipitation products against rain gauge observation and their effect on streamflow simulation, four widely used performance evaluation statistical tools were selected. The Relative Bias (Percent Bias) was used to measure the agreement between the averaged values of simulated rainfall data. Percent Bias (PBIAS) quantifies the average tendency of the simulated data to have different values from observations. The optimum PBIAS is zero, where low values indicate better simulations. Positive values indicate model underestimation, and negative values indicate model overestimation (Yousefi & Moridi 2022).
A smaller value of RMSE indicates a better model performance.


The closer R2 is to one, the better the model explains the data. In the case of a precise fit, R2 = 1 and naturally values greater than 0.5 are considered suitable (Legates & McCabe 1999; Moriasi et al. 2007).
The simulation outcomes are considered to be good if NS ≥ 0.75, and satisfactory if 0.36 ≤ NS ≤ 0.75 (Van Liew & Garbrecht 2003,;Heathman & Larose 2009 ;Demissie et al. 2013). Otherwise, they are considered as unsatisfactory.
Model calibration and validation
As indicated above, the modeling process by the PyTOPKAPI model starts with pre-processing of the DEM and preparation of model input parameters with GIS and then incorporates them into the model to simulate the streamflow data along the drainage system. In this study, streamflow modeling was performed using 10-year SBPP data (2001–2010) over the Gilgel Ghibe basin located upstream of Gilgel Ghibe Nr. Assendabo gauging station (Figure 2). The watershed is one of the two test catchments for the application of PyTOPKAPI model in Ethiopia (Rabba et al. 2018). In order to generate a more reliable streamflow information, the model parameters need to be calibrated. The calibration was performed using 15-year gauged rainfall data (1986–2000) (Rabba et al. 2018). The remaining dataset (2001–2010) of the gauged rainfall was used for verification of the results. Subsequently, the PyTOPKAPI model generated daily streamflow comparable to the daily streamflow observation at the basin's outlet.
RESULTS AND DISCUSSION
In order to analyze the performance of the SBPPs for streamflow modeling, it is important to evaluate the effect of individual SBPP against the gauged rainfall data. Therefore, the evaluations were performed for both the precipitation inputs and the corresponding simulated streamflow outputs. In this case, the precipitation/streamflow evaluations were conducted at daily time scales with statistical measures. Details are given in the following.
Comparison of precipitation inputs
Basin-averaged precipitation: (a) gauged, (b) TRMM3B42, (c) WaterBase, (d) TRMM bias-adjusted, and (e) WaterBase bias-adjusted.
Basin-averaged precipitation: (a) gauged, (b) TRMM3B42, (c) WaterBase, (d) TRMM bias-adjusted, and (e) WaterBase bias-adjusted.
Evaluation of simulated streamflow
Simulated and observed streamflows, with gauged rainfall, for the calibration period (1986–2000).
Simulated and observed streamflows, with gauged rainfall, for the calibration period (1986–2000).
Simulated and observed streamflows for the validation period (2001–2010): (a) gauged rainfall, (b) TRMM, (c) WaterBase, (d) TRMM bias-adjusted, and (e) WaterBase bias-adjusted.
Simulated and observed streamflows for the validation period (2001–2010): (a) gauged rainfall, (b) TRMM, (c) WaterBase, (d) TRMM bias-adjusted, and (e) WaterBase bias-adjusted.
While the WaterBase product largely missed the higher peak flows, the TRMM product more or less adequately captured most of the peak flows. The statistical comparisons of the three precipitation products (Gauged /TRMM/WaterBase) showed that the streamflow simulations forced by gauged rain data had better value of the statistical metrics than those based on TRMM and WaterBase precipitation products in the validation period (Table 1). Interestingly, the TRMM rainfall-based simulations had similar, but slightly worse performance than the gauged rainfall-forced simulations in the validation period.
The statistical comparisons of the three precipitation products (bias-unadjusted)
Precipitation products . | Precipitation input . | Simulated streamflow . | ||||
---|---|---|---|---|---|---|
R2 . | RMSE . | Bias (%) . | R2 . | RMSE . | NS . | |
Gauged | – | – | – | 0.899 | 28.58 | 0.669 |
TRMM | 0.487 | 5.853 | 10.41 | 0.753 | 40.37 | 0.339 |
WaterBase | 0.283 | 7.943 | 26.39 | 0.752 | 60.51 | − 0.486 |
Precipitation products . | Precipitation input . | Simulated streamflow . | ||||
---|---|---|---|---|---|---|
R2 . | RMSE . | Bias (%) . | R2 . | RMSE . | NS . | |
Gauged | – | – | – | 0.899 | 28.58 | 0.669 |
TRMM | 0.487 | 5.853 | 10.41 | 0.753 | 40.37 | 0.339 |
WaterBase | 0.283 | 7.943 | 26.39 | 0.752 | 60.51 | − 0.486 |
Simulated and observed streamflows for 2001–2010 with unadjusted rainfall.
The statistical comparisons of the three precipitation products (bias-adjusted)
Precipitation Products . | Precipitation input . | Simulated streamflow . | ||||
---|---|---|---|---|---|---|
R2 . | RMSE . | Bias (%) . | R2 . | RMSE . | NS . | |
Gauged | – | – | – | 0.899 | 28.6 | 0.669 |
TRMM | 0.487 | 5.65 | 0.00 | 0.766 | 34.7 | 0.513 |
WaterBase | 0.283 | 6.99 | 0.04 | 0.774 | 39.0 | 0.382 |
Precipitation Products . | Precipitation input . | Simulated streamflow . | ||||
---|---|---|---|---|---|---|
R2 . | RMSE . | Bias (%) . | R2 . | RMSE . | NS . | |
Gauged | – | – | – | 0.899 | 28.6 | 0.669 |
TRMM | 0.487 | 5.65 | 0.00 | 0.766 | 34.7 | 0.513 |
WaterBase | 0.283 | 6.99 | 0.04 | 0.774 | 39.0 | 0.382 |
Simulated and observed streamflows for 2001–2010 with bias-adjusted rainfall.
Flow-duration curves for observed and simulated flows (bias-adjusted).
CONCLUSION AND RECOMMENDATIONS
Climate change has led to an intensified impact on the stability of water resources (Nazari Mejdar et al. 2023), including rivers, lakes, aquifers, and sources of fog water (Morichi et al. 2018). Moreover, the dynamic properties of water enable it to flow between different locations, which can lead to flooding in some areas and drought in the other regions (Salehi et al. 2020). An extensive knowledge of watershed hydrological information is therefore fundamental for the efficient water resources management and flood control (prevention, protection and mitigation) for which continuous hydrological records are indispensable (Akbari et al. 2022; Budhathoki et al. 2024). That is to say, an accurate streamflow information has great importance for planning/designing of water resource projects, peak flood estimation, early flood warning systems, dry season water management, climate foresight, etc (Belina et al. 2024; Budhathoki et al. 2024). Such streamflow information can be obtained either through ground-based monitoring station or by streamflow modeling. Precipitation data is the key ‘input’ to the streamflow modeling process. The SBPPs are viable sources of precipitation data, particularly for developing regions/countries like Ethiopia with poor or nonexistent ground-based streamflow measurements. In this study, 10 years (2001–2010) of daily precipitation estimates from two SBPPs (TRMM and WaterBase) were evaluated by comparing them with measurements of precipitation from ground stations over the Gilgel Ghibe catchment in Ethiopia. These products were then introduced to the PyTOPKAPI model to generate streamflows at daily time scale and the results were compared with streamflow observations at the same catchment outlet (Ghibe Nr. Asendabo gauging station) with the help of four statistical tools (Bias, R2, NS, and RMSE). The results indicate that the bias-adjusted SBPPs agree well with gauged rainfall compared to bias-unadjusted ones. The SBPPs with no bias-adjustment tend to overestimate (high bias and high RMSE) extreme precipitation events and the corresponding simulated streamflow outputs, particularly during wet months (June–September) and underestimate the streamflow prediction over few dry months (January and February). Overall, we conclude that the general streamflow patterns were well captured at daily time scale from SBPPs after bias adjustment. However, simulated streamflow using the gauged rainfall is superior to those obtained from the remotely sensed rainfall products tested, including bias-adjusted ones.
Despite their global coverage, SBPPs are not universally integrated into operational hydrologic modeling in Ethiopia mainly due to lack of information on the reliability of such products at basin scale. Therefore, studies on the reliability of SBPPs for streamflow modeling in data sparse areas like Ethiopia should be the major concerns of the future.
ACKNOWLEDGEMENTS
The author acknowledges the Ethiopian National Meteorological Agency (ENMA) of Ethiopia, the Ministry of Water and Energy in Ethiopia for their necessary data in this study. The author is also very thankful to School of Engineering, University of Kwa-Zulu-Natal for providing all sorts of support to conduct this research work.
DATA AVAILABILITY STATEMENT
All relevant data are included in the paper or its Supplementary Information.
CONFLICT OF INTEREST
The authors declare there is no conflict.