Modernizing harvested paddy areas estimation through satellite imagery time series and machine learning

Abstract

Paddy data is essential for policymakers to formulate Indonesia's national food security strategies, especially regarding harvested area. Currently, estimates are generated monthly using ground truth data from sampled locations through the Area Sampling Frames (ASF) process. However, the high cost and various field obstacles highlight the need for alternative methods. Satellite Imagery Time Series (SITS) data, mainly Sentinel-1 historical imagery, offers a promising alternative for detecting phenological stages. However, SITS data require machine learning modeling—for example, XGBoost—which has proven successful in several classification tasks. This study presents an alternative approach in West Java province, a major region for paddy production. The workflow includes preliminary analysis, data preprocessing, region-specific modelling, prediction, and estimation. Most regional clusters demonstrate high accuracy in classifying phenological stages, and harvested area patterns closely align with official statistics, demonstrating the effectiveness and potential of this approach.