Horticulture acreage estimation: remote sensing vis-a-vis sample surveys

Timely and spatially explicit estimates of horticultural crop acreage are required to support agricultural statistics, resource allocation, and production planning. This study assessed the agreement between satellite-derived acreage estimates and house-to-house survey data for selected villages in Erode, Chittoor, and Nashik districts of India. Multispectral imagery from Resourcesat-2 LISS-IV and Sentinel-2 was subjected to radiometric and geometric correction, band stacking, mosaicking, and normalized difference vegetation index-based vegetation masking. Maximum Likelihood Classification (MLC), Spectral Angle Mapper (SAM), and Support Vector Machine (SVM) algorithms were evaluated for mapping banana, tapioca, mango, and onion. Classification performance varied among crops. MLC achieved overall accuracies of 84.83% for banana and 76.81% for mango, with kappa coefficients of 0.767 and 0.917, respectively. SAM produced an overall accuracy of 78.70% and a kappa coefficient of 0.797 for tapioca, whereas SVM performed best for onion, attaining an overall accuracy of 76.41% and a kappa coefficient of 0.713. Satellite-derived acreage estimates were strongly associated with survey estimates for banana (R 2 = 0.94), tapioca (R 2 = 0.93), and mango (R 2 = 0.94). Compared with the survey estimates, remote sensing underestimated the total areas of banana, tapioca, mango, and onion by 9.04%, 5.35%, 5.88%, and 26.08%, respectively. The results indicate that classification performance depends on crop structure, field size, spatial fragmentation, and classifier selection. Satellite imagery provided more reliable estimates for relatively large and contiguous orchards than for small and fragmented crop fields. Integrating satellite-based mapping with targeted field verification can therefore provide a scalable and spatially explicit framework for operational horticultural acreage estimation.