Precipitation Estimation by Integrating CHIRPS with Weather Parameters Using Machine Learning Approaches in Peninsular Malaysia

In Malaysia, precipitation estimation is always the main focus as it is tied to flood prediction. This study attempted to evaluate the performance of the raw satellite estimations extracted from climate hazards group infrared precipitation satellite (CHIRPS) by comparing them with the rain gauge observations. In addition, artificial neural network (ANN) and random forest (RF) were proposed to merge CHIRPS and weather parameters (i.e., skin temperature, radiation (long-wave), and evapotranspiration). The capabilities of the developed models in estimating precipitation for 515 rainfall stations across Peninsular Malaysia from 2011–2020 on overall scale, regional-based and seasonal scale were evaluated, through correlation coefficient (CC), percent bias (PBIAS), mean absolute error (MAE) and root mean square error (RMSE). ANN performs better in daily precipitation estimation across Peninsular Malaysia if compared with RF as it has a higher value of CC (ranging from 0.66 to 0.99), as well as a lower value of PBIAS (ranging from −12.69% to 4.39%), MAE (ranging from 0.19 mm/day to 3.18 mm/day) and RMSE (ranging from 0.29 mm/day to 5.46 mm/day). It is worth highlighting that in terms of regional-based and seasonal scale assessments, the ANN model shows a better performance for daily precipitation estimation in the east region of Peninsular Malaysia and during the northeast monsoon, respectively. The output of the study may provide an effective tool for the relevant stakeholders while developing strategies for water resources management.