An evaluation of satellite precipitation downscaling models using machine learning algorithms in Hashtgerd Plain, Iran

Satellite precipitation products are one of the sources of precipitation estimates. However, their spatial resolution is often too coarse for use in local areas or parameterization of meteorological and hydrological models at basin and plain scales. Therefore, the main objective of this study is to evaluate the accuracy and uncertainty analysis of the downscaled monthly precipitation modeling of the CHIRPS satellite product based on rain gage station data. Five high-resolution ancillary data were used for this purpose including land surface temperature (LST), leaf area index (LAI), Normalized Difference Vegetation Index (NDVI), and Enhanced Vegetation Index (EVI) extracted from the MODIS/Terra satellite database, and elevation data obtained from the Shuttle Radar Topography Mission (SRTM) of NASA JPL satellite database. In addition, precipitation data were extracted from CHIRPS satellite data. Four machine learning models were used to model the downscaling, including support vector machine (SVM), gradient tree boost (GTB), random forest (RF), and classification and regression tree (CART). For accuracy evaluation, root mean square error (RMSE), Bias, multiplicative bias (mBias), correlation coefficient (CC), and optimum index factor (OIF) error estimators were applied. The results of the accuracy evaluation results showed that the CART and GTB models with the best performance according to the error estimators were the best models for stations at lower altitudes and latitudes (Karim Abad, Najm Abad, and Somea) and stations at higher altitudes and latitudes (Valian, Sorheh, and Fashand), respectively. The uncertainty analysis was performed using the bootstrap method. The results showed that the CART and GBT models had a more reliable estimate and lower uncertainty than the other models. This study highlights the power of the Google Earth Engine and machine learning algorithms in downscaling modeling to improve the resolution of satellite precipitation data.