Prediction of Index Rainfall Using a Cubist Model: A Case Study of Cheliff Watershed (Algeria)

Abstract

This research paper investigates a cubist method as a rule-based regression predictive model for index rainfall (IR) estimation. The IR is required both in the regional frequency analysis procedure and in the evaluation of probable maximum precipitation. This IR is still considered a basic means in the rainfall-runoff transfer process. Data used include annual maximum rainfall from 75 rain gauge stations in the Cheliff watershed (Algeria). The data have geographic information and annual precipitation values. The adopted model was trained on 70% of the available data with optimized hyper-parameters using the leave one out cross-validation (LOOCV) technique. The remaining (30%) of the data were used as a testing set for evaluation. Three metrics: Correlation Coefficient (R), Root Mean Squared Error (RMSE) and Mean Absolute Error (MAE), were used to measure the prediction performance of the regression model. Finally, the results compare models with and without introducing climatic input.