Flood Modeling using SWAT and GR4J Models in the Oti River Basin in Togo (West Africa)

 Corresponding author: Koungbanane Dambré Copyright © 2025 Author(s) retain the copyright of this article. This article is published under the terms of the Creative Commons Attribution License 4.0. Flood Modeling using SWAT and GR4J Models in the Oti River Basin in Togo (West Africa) Koungbanane Dambré 1, 2, *, Lemou Faya 1, Kodja Japhet Domiho 2, Totin Vodounon Henri Sourou 2, 3 and Amoussou Ernest 2, 3 1 Laboratory of Biogeographic Research and Environmental Studies, Department of Geography, University of Lome, Department of Geography, Faculty of Human and Social Sciences, Lome-Togo. 2 Laboratory Pierre PAGNEY, Climate, Water, Ecosystems and Development (LACEEDE), Université d’Abomey-Calavi, Department of Geography, Faculty of Arts, Humanities and Social Sciences, Cotonou-Benin. 3 Laboratory of Tropical Climatology and Ethnoclimatology, Department of Geography and Regional Planning, Faculty of Arts, Humanities and Social Sciences, Université de Parakou, Benin. World Journal of Advanced Research and Reviews, 2025, 28(01), 2046-2061 Publication history: Received on 15 September 2025; revised on 25 October 2025; accepted on 27 October 2025 Article DOI: https://doi.org/10.30574/wjarr.2025.28.1.3607 Abstract The Earth's climate evolution over the past few decades has been marked by global warming, leading to hydroclimatic variability and impacting water resource availability. Hydrological modeling has become a solution for reproducing river flows and is an essential tool for understanding watershed dynamics and mitigating natural disasters related to floods. The objective of this study is to reproduce flood flows in the management of floods in the Oti River Basin in Togo. To achieve this objective, the semi-distributed physically-based SWAT (Soil and Water Assessment Tool) and conceptual GR4J (Génie Rurale Quatre paramètres Journaliers) hydrological models were chosen as the base model to calculate flood hydrographs. The data used for this study consisted of a DEM, land use map, soil map, rainfall, temperature, flow, and PET observation data. Sensitivity analysis revealed the most significant parameters of the SWAT and GR4J models. The values of these parameters were determined during the calibration and validation phases. The Nash criterion value for the SWAT model is 0.83 and 0.78 for calibration and 0.72 and 0.81 for validation. Additionally, the KGE values of GR4J range from 0.73 to 0.85 during calibration and 0.62 to 0.81 during validation. These results show excellent performance of the SWAT and GR4J models in simulating flood flows in the watershed. However, the analysis of the results shows that the SWAT model is more efficient than GR4J in simulating high flows and could be used for flood management in the Oti River Basin. Keywords: Modeling; Rainfall-Runoff; Soil and Water Assessment Tool; Floods; Oti 1. Introduction In West Africa and Togo, floods have become recurrent in recent years, with significant socio-economic and ecological impacts, including loss of human life and material damage [1]. These floods have highlighted the need for hydrological forecasting to reduce the vulnerability of populations. In Togo, very few studies have addressed flood-related issues due to a lack of data caused by the limited number of stations and the frequent destruction of limnigraphs by high waters. Hydrological modeling is a solution for reproducing river flows, particularly in the context of flood risk. It is an essential tool for understanding watershed dynamics, rational water resource management, and mitigating natural disasters related to floods [2, 3]. World Journal of Advanced Research and Reviews, 2025, 28(01), 2046-2061 2047 To protect the riparian populations of the Oti River Basin from recurrent floods caused by high waters, it is essential to have a reliable and operational forecasting tool, as flood forecasting is a pressing issue with undeniable importance today. To this end, mastering and rationally managing surface water requires a better understanding of episodic flows and the influence of extreme climatic events, all through conceptual rainfall-runoff modeling. With climate change and variability, as well as population growth, it has become increasingly necessary to use water resources as rationally as possible [4]. This constraint requires an in-depth understanding of watershed physics and a thorough comprehension of how hydrological processes related to water flow in the watershed occur. The hydrological model, which allows calculating fluxes (flows, sediments, nutrients, chemicals, etc.) within the watershed, has become indispensable today, both for fundamental hydrological research and for water resource planning and management. The Oti River Basin in Togo, having experienced significant land use changes, is prone to recurrent floods. Therefore, hydrological modeling using SWAT and GR4J allows evaluating the performance of these models in simulating flows in the watershed under changing surface conditions. Considering this aspect is currently a major challenge in modeling tropical river flows, given the significant anthropization recently observed in their watersheds, with well-known hydrological impacts. The SWAT and GR4J models have already proven their effectiveness in several West African watersheds. In the Oti River Basin at the Mango outlet in Togo, research on hydrological modeling is scarce, with [5] being one of the few studies that used a semi-distributed model. A rainfall-runoff model is particularly interesting in developing countries, as it can estimate or represent available resources for development purposes and predict the evolution of these resources in the coming years in the context of climate change [6, 7]. It also allows appreciating the rainfall-runoff relationship that can generate floods. From a spatial perspective, there are global and distributed models. Global models are one-dimensional and do not account for spatial heterogeneity or variations in hydrological variables and processes. Distributed models, on the other hand, have a spatial structure, dividing the watershed into smaller grid cells that can capture the spatial variability of hydrological processes and watershed characteristics. From a process description perspective, hydrological models can be conceptual or physically-based. Conceptual models represent different hydrological cycle processes using storage elements, with transfers between these reservoirs represented by mathematical functions. Physically-based models typically represent different hydrological cycle processes using non-linear partial differential equations derived from fluid mechanics conservation laws. Spatially distributed and physically-based models provide detailed information on watershed functioning and are used to deepen the understanding of hydrological process dynamics and simulate the hydrological response of a watershed at a spatial scale. This justifies the use of the semi-distributed physically-based SWAT model and the conceptual GR4J model in the Oti River Basin. In terms of performance, comparing global and semidistributed approaches in rainfall-runoff modeling is a long-standing issue [8]. Scientific research on this topic presents a complex picture, with some studies clearly affirming the benefits of using semi-distributed approaches [9], while others present opposing conclusions [10]. This study aims to reproduce flood flows for sustainable flood management in the Oti River Basin, Togo. 2. Material and methods 2.1. Study area The Oti River basin in Togo is located in the north of the country, in the Volta River basin, and drains part of the watershed areas of the countries bordering Togo, namely Burkina Faso, Benin and Ghana. At the Mango outlet, the Oti basin covers an area of 3.652 square kilometres and is located between 11˚05' and 10˚96' N and 0˚12' and 0˚95' E (Figure 1). World Journal of Advanced Research and Reviews, 2025, 28(01), 2046-2061 2048 Figure 1 Location of the Oti watershed in Togo The topography of the watershed is based on a morphology that includes the Gourma peneplain, the DapaongBombouaka plateaus and the Oti plain. The total elevation ranges from 106 to 506 meters. Its relatively dense hydrographic network is controlled by the Oti River. The Oti river, the main water collector (167 kilometers long in Togo), rises in northern Benin on the eastern slopes of the Atacora chain, under the name of Pendjari. He follows a south-westnorth-east orientation, crosses the Atacora region, and then re-verses and returns south-west, arriving in Togo, where it takes the name of the Oti [11]. The Oti river forms the natural border between Togo and Benin from its confluence with the Koumongou River, south of the Ga-langachi classified forest, to the Ghanaian commune of Sabari, south of which the Oti river flows into Lake Volta. Its tributaries are the Wabga, Namiélé and Sansargou on its right bank. Its left bank tributaries are the Koumongou, Kéran and Kara. The riverbed of the Oti is cut into the clay formations and here and there bordered by a forest gallery. When the Oti overflows its banks during the rainy season, it causes major flooding with serious consequences for the people living along its banks. According to the 5th General Census of Population and Housing (5thRGPH) in 2022, the Oti watershed had a population of 1.143.520. Several activities are practiced by these populations. These include agriculture, livestock, fish-ing, hunting, handicrafts, commerce, tourism, sand and gravel extraction, and transportation. However, agriculture seems to be the main activity of the peo-ple living in the watershed. 2.2. Data used Several types of data were used for this study. These include DTM, land use data, soil data and hydroclimatic data. 2.2.1. DTM (Digital Terrain Model) HydroSHEDS data were used to delineate watersheds. HydroSHEDS is derived from SRTM (Shuttle Radar Topography Mission) elevation data at a resolution of 3 arc seconds (WWF, 2018). The original SRTM data were hydrologically conditioned using a sequence of automated procedures. Existing data enhancement methods and newly developed algorithms were applied, including gap filling, filtering, flow engraving and upscaling techniques. Manual corrections were also made. Preliminary quality assessments indicate that the accuracy of HydroSHEDS far exceeds that of existing global hydrographic and river maps. 2.2.2. Land cover data This data comes from the ESA's Globcover database. It was produced between 2004 and 2006 and was derived by automatic and regional classification of a full-resolution MERIS surface reflectance time series. It has a resolution of 300 m. World Journal of Advanced Research and Reviews, 2025, 28(01), 2046-2061 2049 2.2.3. Soil data The Harmonised World Soil Database (HWSD) version 1.2 produced by the Food and Agriculture Organisation of the United Nations (FAO), the International Institute for Applied Systems Analysis, the Institute of Soil Science of the Chinese Academy of Sciences and the Joint Research Centre (JRC) of the European Commission, was used to generate the soil data required in SWAT. 2.2.4. Hydroclimatic data For this study, the data used are hydrometric and climatic data. The hydrometric data come from the database of the Water Resources Directorate in Lomé. These are the average daily flows observed at the Mango station over the period 1961–2022. The rainfall data observed covers the period 1961–2022 and was collected at seven (07) meteorological stations of the National Meteorological Agency of Togo (ANAMET) in Lomé. These are the stations of Dapaong, Toaga, Mango, Mandouri, Borgou, Barkoissi and Tankpamba. The data covers the period from 1 January 1961 to 31 December 2022. Similarly, maximum and minimum temperature data and ETP data were used. 2.3. Methods 2.3.1. Presentation of the SWAT hydrological model SWAT is a continuous, spatially semi-distributed, semi-physical, daily-timetable hydrological model that can be downloaded free of charge from https://swat.tamu.edu/software. The SWAT model can simulate a large number of physical processes in a catchment area with high spatial resolution. To do this, the watershed is divided into sub-basins, which are themselves subdivided into Hydrologic Response Units (HRUs). HRUs are homogeneous spatial units characterised by the same soil, vegetation cover, and slope. The water balances calculated for the HRUs are used to determine the flows at the outlets of the sub-watersheds. These different outlets are connected to each other by the hydrographic network. The water balance simulated by SWAT uses the equation below [12, 13] 𝑆𝑊𝑡=𝑆𝑊0+∑(𝑅𝑑𝑎𝑦 −𝑄𝑠𝑢𝑟𝑓 −𝐸𝑎− 𝑊𝑠𝑒𝑒𝑝 −𝑄𝑔𝑤 ) 𝑡 𝑖=1 Where: t: time in days, SWt (mm) and SW0 (mm) represent the moisture content in the soil at time t and at the initial time, respectively, Rday (mm) is the rainfall at time i, Wseep (mm) is the rainfall infiltrated into the soil from the surface at time i, Qsurf (mm) is the rainfall available for runoff at time i, Qgw (mm) is the base flow from the water table at time i, and Ea (mm) is the evapotranspiration at time i. In this study, the Penmann-Monteith method is used because it is the most widely used. The water depth available for runoff is determined using the Curve Number method developed by the Soil Conservation Service (SCS). The flow from the watershed surface to the hydrographic network is simulated using the kinematic wave equation. River flow is calculated using the Muskingum method [12, 13]. 2.3.2. Structure and operation of the GR4J model There are several versions of the GR4J model. The one used in the present study is the version adapted by [14] and [15] and illustrated in Figure 2. World Journal of Advanced Research and Reviews, 2025, 28(01), 2046-2061 2050 Source: [14, 15] Figure 2 How the GR4J model works The GR4J model is a daily model with four optimizable parameters: X1 is the capacity of the production reservoir (mm) ; X2 is the underground exchange coefficient (mm) ; X3 is the one-day capacity of the routing reservoir (mm) and X4 is the unit hydrograph (HU1) base time in days. The first reservoir, called the production reservoir, X1, is used to assess the observed precipitation at watershed scale. From the net precipitation, it is possible to define the watershed discharge, taking into account evapotranspired water (precipitation and PTE). The second reservoir, the X3 routing reservoir, enables the quantity of water linked to the flow rate to be broken down over time using previously defined parameters. The model uses as input the average rainfall over the watershed area P (mm) and the po-tential evapotranspiration PTE (mm), and provides as output the flow rate Q (m3/s). The soil water balance is determined by gross rainfall plus potential evapotranspiration (PTE). This action is mathematically represented by: -If P ≥ E then Pn = P-E and En = 0 -If P ˂ E then En = E - P and Pn = 0 In the production reservoir, the rainfall remaining after withdrawal by the action of the PTE is divided between the quantity of water going into the soil reservoir Ps and that of Pn - Ps which transits towards the outlet as follows: World Journal of Advanced Research and Reviews, 2025, 28(01), 2046-2061 2051 The reservoir content is thus updated by: S = S - Es + Ps and the percolation of the production reservoir is written as Perc, which is added to Pn - Ps such that The reservoir content is then again identified by: S = S - Perc. Rainfall is separated into two flow components in the model. A pseudo-direct flow is routed by a unit hydrograph (SH1(X4)) representing 10% of the effective rainfall, i.e. each day j For J = 0, SH1 (j) = 0 For 0 ≤ j ≤ X4, SH1(j) = For j ˃ X4, SH1 (j) = 1 The remaining 90% of effective rainfall is routed via a unit hydrograph (SH2(X4)) and a routing reservoir (transfer). For j = 0, SH2(j) = 0 For 0 ≤ j ≤ X4, SH2(j) For x4 ˂ j ≤ 2X4, SH2(j) For j ˃ 2X4, SH2(j) = 1 At the level of the underground reservoir, the exchange function is provided by F(X2), which can be either an underground inflow or an underground outflow for the basin. Il se It is presented as follows Where R is the level in the routing reservoir (groundwater). The level in the routing reservoir R is updated by adding the output Q9 of the hydrograph (SH1(X4)) and F: R = max (0 ; R + Q9 + F). It then empties to an output Qr given by the following formula: World Journal of Advanced Research and Reviews, 2025, 28(01), 2046-2061 2052 The level in the reservoir then becomes Δ: R= R – Qr The output Q1 of the hydrograph (SH2(X4)) is subjected to the same exchanges to give the direct flow component (Qd) of the stream and the delayed flow (Qr), representing groundwater recharge and materialized by the following formula: Q = Qr + Qd. The utility of this model is to reproduce flood flows [16, 17, 18], as well as flood hazards from rainfall and PET to alert communities [17]. 2.3.3. Calculation of potential evapotranspiration (PTE) The method used to calculate PTE is that of [19] Oudin et al. (2005), previously tested by [20, 21]. It consists of estimating daily PTE for hydrological modelling and is derived from the Jensen-Haise and McGuinness models. These models take into account only the average daily air temperature and solar radiation, which depends on latitude and the 365 days of the year. They are presented as follows 𝑆𝑖 𝑅𝑒𝑇𝛼 + 𝐾2 > 0, 𝑃𝐸 = 0 If 𝑅𝑒𝑇𝛼 + 𝐾2 ˃ 0,𝑃𝐸 = 0 Where PE: daily potential evapotranspiration (mm.d-1); Re: solar radiation (MJ.m-2.d-1) ; Ta: daily mean temperature (°C); γ: latent heat of evaporation of water at 20° (taken equal to 2.45 MJ. Kg-1); r : density of water (kg.m-3). K1 (°C) and K2 (°C) are fixed model parameters because they are adjusted during the sampling of the entire basin and are not specifically calibrated for each watershed [19, 22]. 2.3.4. Optimization quality criteria Hydrological modeling using SWAT and GR4J models allows evaluating water availability by highlighting their performance in reproducing flood flows, low flows, and average flows, based on Nash, R2, and KGE criteria. The most well-known and widely used model optimization criteria are Nash-Sutcliffe Efficiency (NSE) and Kling-Gupta Efficiency (KGE). The Nash criterion is used for the SWAT model, which is one of the most commonly used statistical indicators by hydrologists. Additionally, the KGE criterion was used for GR4J because the study aims to find the criterion that best corresponds to GR4J in reproducing high-water flows in the Oti watershed. Therefore, this study is based on both Nash and KGE criteria to analyze the performance and robustness of SWAT and GR4J models in simulating flood flows. The Nash and KGE values range from -∞ to 1. The closer they are to 1, the closer the simulated flows are to the observed flows. Table 1 Presents the performance of SWAT and GR4J models using Nash and KGE criteria in the Oti watershed, Togo Criteria Nash/KGE Model performance 0.75 < Nash/KGE ≤ 1 Very Goog 0.65 < Nash/KGE ≤ 0.75 Good 0.50 < Nash/KGE ≤ 0.65 Sufficient Nash/KGE ≤ 0.50 No Sufficient Source: [23] World Journal of Advanced Research and Reviews, 2025, 28(01), 2046-2061 2053 2.3.5. Model calibration and validation The calibration and validation procedure involves first dividing the time series data into two independent periods of equal size (P1 and P2). Then, the model is calibrated on the first period (P1) and validated on the second (P2). Alternatively, it can be recalibrated on the second period (P2) and validated on the first (P1) depending on the objectives. All chosen periods should be preceded by a one-year warm-up period to allow for the initialization of reservoir contents. To ensure that the model accurately reproduces flood flows during calibration and validation, four periods were considered in the time series. Initially, the models were calibrated from 2016-2017 and validated from 2018-2019. Subsequently, they were calibrated from 2019-2020 and validated from 2021-2022. The choice of periods for calibration and validation in this study is justified by the fact that these periods have fewer missing data. 2.3.6. Model calibration and sensitivity analysis of SWAT Calibration is the process of estimating the values of the most significant model parameters by comparing the calculated quantities with observed quantities for a fixed set of parameters. It involves testing the model with known inputs and outputs to adjust the model parameters. The SWAT model has a large number of parameters with a wide spatial and temporal variation spectrum. Its calibration is complex and computationally intensive. Therefore, it is necessary to eliminate insensitive parameters to simplify the calibration process. This exercise is called sensitivity analysis. Sensitivity analysis provides a final list of parameters ranked in decreasing order of sensitivity. These parameters are used for calibration. Both calibration and sensitivity analysis can be performed manually or automatically using the SUFI-2 algorithm in SWAT-CUP [24, 25]. 2.3.7. Validation of SWAT and GR4J models The evaluation of the calibrated model allows verifying the ability of the estimated model parameters to realistically reproduce the watershed response using performance indicators that are either graphical or statistical. Thus, the model results after calibration are compared with direct observation and measurement results. Graphical indicators mainly involve comparing the hydrographs calculated by the models with observed hydrographs. Among the existing statistical indicators, the Nash and KGE criteria, which are widely used by hydrologists, were chosen for this study. 2.3.8. Determination of flood thresholds Flood thresholds in the Oti watershed concern discharge rates, and for this research, the Standardized Flow Index (SFI) is used for floods. Thus, from daily discharge rates at multiple time scales, the SFI indices were calculated. The standardized index is based on the equiprobability transformation of discharge values, aggregated over k-months into standard normal values, with k typically fixed according to the analysis objectives, such as k = 1, 3, 6, 9, 12, 24, 36 months [26]. The classical standardized index (denoted Z) was used but with a change in time step to a daily scale. Discharge data are used to calculate these indices using the following formula: where y is the unit value of the considered parameter, μ and σ denote the mean and standard deviation of the variable, respectively. The SFI categories (Table 2) allowed classifying the maximum daily flow values in terms of flood hazards for different risk levels (limited, moderate, significant, and critical) in the Oti watershed. This categorization was made possible by transposing the classification of [27] to the daily data used. Table 2 Classification of SFI values, flood categories, and risk levels SFI threshold values Flood category Risk levels 2.00 and more Catastrophic Critical 1.5 to 1.99 Severe Significant 1 to 1.49 Negligible Moderate 0 to 0.99 No impact Limited Source: Adapted of [28, 27] World Journal of Advanced Research and Reviews, 2025, 28(01), 2046-2061 2054 3. Results 3.1. Analysis of the results of the calibration and validation of the SWAT and GR4J models Figure 3 illustrates the variability of observed and simulated flows during calibration (2016–2017 and 2019–2020) and validation (2018–2019 and 2021–2022) with SWAT and GR4J. Figure 3 Variability of observed and simulated flows during calibration and validation with SWAT and GR4J From a hydrological point of view, it is worth noting the flow variability in the Oti watershed in Togo, marked by a succession of low and high water. Surface runoff evolution shows that in some years, daily flow rates are lower than in others, resulting in daily flow variability. With the model GR4J, the variability of observed and simulated flows in calibration and validation gives very good KGE values (ranging from 0.73 to 0.85 in calibration and 0.62 to 0.81 in validation), demonstrating the GR4J's strong performance in simulating flood flows. The observed flow rates are consistently lower than the simulated flow rates during the calibration and validation phases, which can be explained by their position below the simulated flow rate curve. Simulated flows overestimate observed flows during low water periods and underestimate them during high water periods (August to September). Similarly, SWAT modeling results showed very good performance in simulating flood flows. Sensitivity analysis revealed the most significant model parameters. The values of these parameters were determined during calibration and validation. The value of the Nash criterion is 0.83 and 0.78 for the calibration phase and 0.72 and 0.81 for the validation World Journal of Advanced Research and Reviews, 2025, 28(01), 2046-2061 2061 [31] B.A. Tolson and C.A. Shoemaker, ‘’Cannonsville reservoir watershed SWAT2000 model development, calibration and validation’’ Journal of Hydrology, 337(1-2): 68-86, 2007. [32] K.A. Anoh, T.J.J. Koua, K.J. Kouamé, J.P. Jourda, F. Laurent, ‘’Modelling water flow in a complex watershed in humid a tropical area using swat: a case study of Taabo watershed in Ivory Coast’’. International Journal of River Basin Management, 16 (2): 157-167, 2018b. [33] K.A. Anoh, T.J.J. Koua, S.G. Eblin, K.J. Kouamé, J.P. Jourda, ‘’Modelling Freshwater Availability Using SWAT Model at a Catchment-Scale in Ivory Coast’’. Journal of Geoscience and Environment Protection, 05 (13): 70-83, 2017. [34] J. Schuol, K. C. Abbaspour, H. Yang, R. Srinivasan, J.B.A. Zehnder, ‘’Modeling blue and green water availability in Africa’’. Water resources research, 44: 1-18, 2008. [35] Y. Tramblay, D. Ruelland, S. Somot, R. Bouaicha, and E. Servat, ‘High-resolution regional climate simulations from the Med-CORDEX model for hydrological impact studies: a first assessment of the ALADIN climate model in Morocco.’ Hydrol. Earth Syst. Sci. 17, 3721–3739. doi: 10.5194/hess-17-3721-2013. [36] D. T. Rwasoka, C. E. Madamombe, W. Gumindoga, and A. T. Kabobah, ‘Calibration, validation, parameter identification and uncertainty analysis of a parsimonious two-parameter monthly rainfall model in two catchments in Zimbabwe.’ Phys. Chem. Geol. A/B/C/ 67-69, 36–46, 2014. doi: 10.1016/j.pce.2013.09.015. [37] V. S. H. Totin, E. Amoussou, L. Odoulami, M. Boko, B. A. Blivi, ‘Rainfall thresholds for flood risk levels in the Ouémé basin in Benin (West Africa)’. In ‘XXIXth Colloquium of the International Association of Climatology’, Lausanne – Besançon, 2016, pp. 369-374.