Amélioration de la simulation des précipitations en Afrique Centrale à partir du modèle climatique régional RegCM4

Abstract

This work proposes the possibility, more or less limited, for the regional climate model RegCM4.4 to represent the reality of precipitation in Central Africa (CA) when there could be uncertainties in these aspects after the simulations. In this sense, harmonic analysis and bias correction methods are essential to improve their representativeness at the output of RCMs. The first step of this study examined the reliability of the downscaling tool RegCM4.4 to simulate the characteristics of the diurnal cycle of precipitation from June to September 2002-2006. Besides their diurnal cycles, the spatial and temporal patterns in precipitation intensity, amount and frequency over CA are investigated. Diurnal variance, phase and amplitude based on 3-hourly model simulations are obtained by diurnal harmonics from each 24-h period. Two statistical measures are used to evaluate model performance : the root mean square error (RMSE) and the index of agreement (IA). The results shows that the RegCM outputs are well simulated compared with reference data in revealing the temporal and spatial patterns of precipitation amount and frequency over the continental area with some systematic wet biases over Cameroon highlands area. Diurnal variability of precipitation frequency and amount are properly well reproduced by the model with an afternoon peak around 1800 LST over entire domain except Atlantic Ocean sub-region. The model does not properly describe the observed diurnal variation of precipitation intensity over the study area. One of the prominent results is that the pattern of precipitation frequency is quite similar to that of precipitation amount. This strong relationship between these two precipitation characteristics over the entire region of interest suggests that the diurnal precipitation variability is generally determined by how often it rains. The second step consists of bias correction of the simulated rainfall during the four seasons and the annual cycle for the control period of 1991-2001 and the evaluation period of 2002-2012. It is found that the variance method corrects better for precipitation in CA.