<![CDATA[Unpredictable rainfall is considered a major challenge for agricultural systems in Indonesia, especially in the preparation of planting calendars. Therefore, accurate rainfall predictions are essential for agricultural systems to be more adaptive to climate change. The aim of this study is to analyse and compare the accuracy of two monthly rainfall prediction schemes, namely recursive and semi-recursive approaches, by using the Random Forest algorithm. Climatological data from four stations in the Bogowonto Watershed were used, and the modelling process included data pre-processing, feature engineering (lag, rolling window, and seasonal transformation), and grid search with cross-validation to obtain the optimal parameter combinations. Model performance was evaluated on out-of-sample test and validation data by using RMSE, MAE, NSE, and R². The semi-recursive approach improved NSE from 0.20–0.33 to 0.53–0.61 and reduced RMSE from 210.68–255.75 mm to 150.79–211.31 mm, while R² values increased from 0.23–0.50 to 0.72–0.82 across the four stations. These results indicate that the semi-recursive approach is more stable in predicting monthly rainfall and thus is recommended for planning a planting calendar in Bogowonto Watershed.]]>
Copyrights © 2026
