SSRG International Journal of Geo-informatics and Geological Science Volume 10 Issue 1, 29-35, Jan-Apr 2023 ISSN: 2393–9206 / https://doi.org/10.14445/23939206/IJGGS-V10I1P103 © 2023 Seventh Sense Research Group® This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/) Original Article Modeling River Discharge using Deep Learning in the Ouémé catchment at Savè outlet (Benin, West Africa) Zohou Pierre Jérôme 1,2,3,4 , Biao Iboukoun Eliézer ,4,5 , Aoga John 1,3 , Houessou Oscar 5 , Alamou Adéchina Eric 4,5 , Eugène C.Ezin 1,3 1 Université d’Abomey-Calavi (UAC), Bénin. 2 Chaire Internationale en Physique Mathématique et Applications (CIPMA-Chaire UNESCO), UAC, Bénin. 3 Laboratoire de Recherche en Sciences Informatiques et Applications (LRSIA), UAC, Bénin 4 Laboratoire d’Hydrologie Appliquée (LHA), UAC, Bénin 5 Université Nationale des Sciences, Technologies, Ingénierie et Mathématiques (UNSTIM), Abomey, Bénin Received: 22 February 2023 Revised: 25 March 2023 Accepted: 08 April 2023 Published: 20 April 2023 Abstract - This paper presents a modeling approach based on Artificial Neural Networks (ANNs) in the Ouémé river catchment at Savè. To do this, we used precipitation data as input over the period 1965 -2010 to simulate river discharge in the study area by using two ANNs models such as the Long Short Term Memory (LSTM) and Recurrent Gate Networks (GRU) models. Indeed, the description of the stochastic nature of the data is better presented today by ANNs models than the statistical models. We compared the performance of these two models based on different evaluation criteria. The predictions made using these models show a strong similarity between the observed and simulated flows. The deep learning models gave good results. Indeed, in calibration and validation, the Nash Sutcliffe Efficiency (NSE) and the coefficient of determination (R²) are very close to one (calibration: R²= 0.995, NSE= 0.991, and RMSE= 0.18; validation: R² = 0.975, NSE= 0.971, and RMSE= 0.41). This good performance of LSTM and GRU confirms the importance of models based on Artificial Intelligence in modeling hydrological phenomena for better decision-making. Keywords - Artificial Neural Networks, Modeling, Ouémé catchment at Savè, Long Short Term Memory, Gated Recurrent Unit. 1. Introduction Precipitation is a natural phenomenon and is generally the largest contributor to the water balance in a watershed. They comprise drizzle, ice, frost, snow, hail, sleet, and rain. However, in West Africa, specifically in Benin, rain feeds the water tables and the various rivers [8]. However, excessive rainfall leads to natural disasters such as flooding. It is, therefore, necessary to better control the phenomenon of precipitation, and this requires its modelling. Over the past few decades, fully data-driven (empirical) models have begun to emerge with breakthroughs in new deep-learning methods in runoff prediction [11]. These breakthroughs were mainly made possible by the availability of large volumes of water-related data. We propose using recurrent neural networks models such as LSTM and GRU to model the rainfall-runoff relationship. To achieve this, we will optimize the hyperparameters of the models, simulate the river discharge at the outlet of the catchment area and finally evaluate the performance of the recurrent neural network models. 2. Materials and Methods The Ouémé is a river that covers at Bonou, the most advanced station before the Delta, an area of 46,990 2 . It rises at the foot of Atacora, in the Djougou region, crosses Benin towards the coast, and flows into Lake Nokoué, just north of Cotonou (Fig. 1). It is thus the longest river in Benin, draining more than a third of the territory alone. The Ouémé basin at Savè (09°12’N; 02°16’E) is the area whose data are used in this project. Its natural outlet, located a few kilometers downstream from the confluence of the Ouémé with the Yérou-Maro, is the Bétérou station, created in 1952; the area covered by the Ouémé in Bétérou is 10,475 2 . Precipitation data used comes from Météo-Bénin (National Meteorological Agency of Benin), while the National Directorate of Water (DG-Eau) provides the river discharge data. The study area contains seven rainfall stations (Savè, Ouesse, Kokoro, Tchaourou, Bassila, Penessoulou, Toui) covering the period from 1965 to 2010.