Imputing Missing Precipitation Data at Benin Synoptic Stations (West Africa) by Using Machine Learning Methods

Precipitation is a key hydrometeorological variable in environmental, hydrological, and agroclimatic studies. Unfortunately, in developing countries as Benin Republic, daily precipitation records from synoptic stations are often characterized by substantial gaps. Thus, to obtain complete datasets, gap- filling is necessary. However, applying inappropriate gap-filling may lead to partial or biased results. This study aims to evaluate the capabilities of five Machine Learning Models (MLM) in estimating missing daily rainfall data over the period 1953-2010 across six Benin synoptic stations. The considered MLM are: (a) Decision Tree (DT), (b) Random Forest (RF), (c) Support Vector Machine (SVM), (d) Multi-Layer Perceptron (MLP), and (e) Multiple Linear Regression (MLR). Mean Absolute Error (MAE) and Root Mean Square Error (RMSE) are applied to evaluate their performances. The results show that the SVM and RF models generally outperformed the other models. But these performances varied unpredictably across stations and months. In contrast, the MLP and DT models showed poor performance. Moreover, all models generally performed better during dry months than the wet months. Finally, their performances was higher at synoptic stations located at higher latitudes.