Analisis Produktivitas Tanaman Padi Berbasis Indeks Vegetasi dari Citra Landsat 8 OLI Menggunakan Geographically Weighted Regression (GWR) di Kabupaten Jember

Abstract

Rice productivity is an important indicator for supporting food security and is influenced by biophysical factors whose relationships may vary across locations. Research on productivity aligns with SDG Goal 2, which emphasizes the importance of achieving food security. This study aimed to analyze the spatial distribution of rice productivity and evaluate the performance of the Ordinary Least Squares (OLS) and Geographically Weighted Regression (GWR) models in explaining rice productivity variation in Jember Regency. The study used secondary data from 31 sub-districts in 2025. The explanatory variables consisted of harvested area, rainfall, elevation, and the Green Normalized Difference Vegetation Index (GNDVI), derived from Landsat 8 OLI imagery. The analysis included satellite image processing, exploratory spatial data analysis, multicollinearity testing, OLS modeling, GWR modeling, and spatial mapping. The multicollinearity test indicated that NDVI and SAVI were excluded from the final model; therefore, harvested area, rainfall, elevation, and GNDVI were retained as explanatory variables. The OLS model produced an R² value of 0.3170, an adjusted R² of 0.2119, and an AICc of 138.1168. Meanwhile, the GWR model produced a higher R² value of 0.6953 and an adjusted R² of 0.3539, indicating its greater ability to explain local variation in rice productivity. However, the GWR model had a higher AICc value of 151.1917, indicating greater model complexity. The predicted rice productivity ranged from 57.58 to 63.82 kw/ha, with relatively higher productivity concentrated in the western and southern parts of Jember Regency, while lower productivity was generally found in the northern and central areas. The results show that the relationships between harvested area, rainfall, elevation, GNDVI, and rice productivity vary across sub-districts. Therefore, GWR is useful for identifying local spatial variation, although its results should be interpreted alongside model- complexity considerations.