Improving the quality of Gridded Precipitation Datasets over Burkina Faso Using merging Methods

Abstract

The management of climate risks such as droughts, floods and heat waves requires high quality historical climate data that offers good spatial and temporal distribution. To achieve this, rain gauges are installed to provide the most reliable measurement of rainfall. However, the rain gauge network is generally very poor in developing countries, leading to many uncertainties, especially in areas where no rain gauge is installed. To fill this gaps, rainfall estimation based on satellites seems to be a good and cost-effective alternative because they supply information for these areas at a relative low cost. However, these datasets are subject to systematic and random errors inherent to the observation method; therefore, there is a need to adjust them before their use for operational applications and decision making. This study proposes a rigorous method in three-step to improve satellite estimation data for Burkina Faso. The first step is devoted to assessing the accuracy of seven satellite precipitation datasets. Then the best dataset is bias corrected using Empirical Quantile Mapping (EQM) and Time and space-variant (TSV) biasadjustment approaches. The final step is to generate blended datasets between the best corrected datasets and in-situ gauges data to produce more robust estimates of precipitation datasets. This blending was performed with Regression kriging (RK) and Mean Field Bias (MFB) with two interpolation techniques namely Shepard and Spheremap. The main results of the study are the following: The evaluation revealed that TAMSAT and CHIRPS were the best for daily and monthly time scales respectively. EQM method outperformed TSV at daily scale, while at the monthly scale the TSV was more suitable for bias correction. Morever, RK-Spheremap was the best of the four methods for merging satellite and in situ data at both time scales. Thus, the approach proposed in this study has improved the correlation coefficients improved the correlation of the daily data by 85.2% (from 0.147 to 0.999), the Bias by 12.4% (from 0.875 to 0.999) and the RMSE by 95.6% (from 26.494 to 1.175). Concerning the monthly dataset, the correlation coefficients are enhanced by 8.4% (from 0.916 to 1), the Bias by 2.3% (from 0.977 to 1) and the RMSE by 99.9% (from 35.654 to 0.042). This study may help in improving floods and droughts monotoring, as well as climate model validation over Burkina Faso.