Integrated Remote Sensing Approaches for Predicting Sugarcane Yield in Fragmented Agricultural Lands

Abstract

Sugarcane is a key raw material for sugar production and plays a significant role in Indonesia’s economy. Efficient monitoring and yield estimation are therefore crucial, yet remain challenging due to geographic variation and climatic conditions. Previous studies on sugarcane yield estimation using satellite data have been largely conducted in extensive, homogeneous plantations. However, little attention has been given to fragmented smallholder sugarcane fields, where historical yield data are scarce and land management is heterogeneous. This study addresses this gap by developing and validating an estimation model using Sentinel-2 derived vegetation indices in smallholder sugarcane fields in East Java, Indonesia. Utilising Sentinel-2 satellite imagery, various vegetation indices—NDVI, GNDVI, NDII, NDRE, and SAVI—were employed as predictors. Data were collected from 38 sugarcane fields across four sub-districts and analysed in relation to actual yields. The results show that the maximum values of the vegetation indices exhibit a strong correlation with sugarcane yield, with NDVI having the highest correlation coefficient (r = 0.72). A multiple linear regression model revealed that index combinations provided the best accuracy: NDVI + GNDVI (R² = 0.65, RMSE = 11.32 t/ha), NDVI + NDII + NDRE (R² = 0.67, RMSE = 11.36 t/ha), and NDVI + NDII + SAVI (R² = 0.68, RMSE = 11.36 t/ha). These findings highlight the applicability of remote sensing-based approaches for yield prediction in fragmented smallholder systems, particularly sugarcane, offering a practical pathway to improve forecasting and support farm-level decision-making.