Pengembangan Model Penentuan Produksi Padi Berbasis Data Indeks Vegetasi dan Data Iklim Menggunakan Algoritma Random Forest

Abstract

Jember Regency, one of the major rice production centers in East Java, continues to face production fluctuations driven by vegetation and climatic factors. Conventional data collection methods, such as the Agricultural Statistics survey form and eye estimate approach, are limited by lengthy data collection processes and suboptimal accuracy. This study aims to develop an accurate and efficient rice production estimation model by integrating vegetation index data and climate data. The research was conducted in two main stages: dataset construction and Machine Learning modeling. Vegetation indices (NDVI, GNDVI, and EVI) were extracted from Landsat 8 imagery using Google Earth Engine (GEE) platform, smoothed using the Savitzky–Golay (SG) filter, and aggregated using the trapezoidal Area Under Curve (AUC) method. Climatic data, comprising rainfall and temperature, were obtained from the CHIRPS and ERA5-Land datasets, respectively. Modeling was performed using the Random Forest algorithm, incorporating outlier analysis, multicollinearity testing via Variance Inflation Factor (VIF), and a comparison of 70:30 and 80:20 train-test split scenarios. Multicollinearity analysis indicated that AUC_GNDVI was considered to be eliminated due to a high VIF value; therefore, the final model utilized four variables AUC_NDVI, AUC_EVI, rainfall, and temperature derived from 305 observations across 31 sub-districts. The best-performing model was obtained using the 80:20 train-test split, achieving a coefficient of determination (R²) of 0.9509, an RMSE of 2741 tonnes, and an MAE of 1,930.7 tonnes. Feature importance analysis revealed that AUC_NDVI contributed the most (27.2%), followed by mean temperature (25.8%), AUC_EVI (25.0%), and total rainfall (22.0%). The integration of satellite-derived data with the Random Forest algorithm proved effective and highly accurate in estimating rice production in Jember Regency.