Abstract

The Oueme basin in Benin, like several other basins in the region, is increasingly affected by floods that have numerous environmental, social, and economic impacts. In this context, it is essential to analyze rainfall occurrence to propose mitigation or adaptation measures for water management. This study analyzes rainfall occurrence over the period 1921 to 2020 in the Oueme River basin at the Savè outlet, using Markov chain models and Artificial Neural Networks (ANN). The methodological approach consisted of occurrence of successive rainy years by using Markov chains and RNAs of order one to three. From the rainfall data, the test on sequential trends with the rainfall index confirmed the existence of three major periods during the study period. We observe a wet period from 1921 to 1940 followed by a normal period from 1941 to 1979 and a dry period from 1980 to 2020. We note a decrease in rainfall from the 1970s onwards, which became more pronounced in 1981. A comparison of the probabilities of the Markov matrices of order 1 to 3, as well as those of the ANR over the sub-periods and the total period of 100 years of study in the Oueme River basin at the Savè outlet, shows a general decrease in the distribution of rainfall in all stations. The study also showed that the probability of having four dry decades is certain over the entire basin. The return of a rainy period as long as the 1920s, 1930s, and 1940s is expected as early as 2025.

Key words: Oueme basin at Savè, Markovian approach, artificial neural networks, rainfall occurrence.