Integrating PRISMA hyperspectral data with Sentinel-1, Sentinel-2 and Landsat data for mapping crop types and land cover in northeast Thailand

Abstract

Accurate crop types and land cover maps are pivotal for effective land management and agricultural policy, particularly in regions with complex agricultural landscapes and small field sizes. Northeast Thailand, a significant agricultural hub, faces challenges in crop classification due to its diverse crop patterns, cloud cover, and smallholder plots. This study integrates satellite data from PRISMA, Sentinel-1 (S1), Sentinel-2 (S2), and Landsat-8/9 (L8/9) imagery to address these challenges. A total of 1305 reference point were randomly collected between November and December 2022 to train and validate the proposed crop classification. Specifically, 15 different combinations using a random forest (RF) classifier were tested. The combination of all datasets achieved the highest overall accuracy (OA) of 91.5 %, followed by S1 + S2 + L8/9 (89.8 %), while PRISMA alone yielded a lower accuracy (63.8 %). The study identified nine dominant land cover classes, with cassava, rice, and sugarcane as primary crops. A strong correlation (r = 0.91) with the official Land Development Department (LDD) statistics demonstrates the robustness of the method. This research highlights the technical advantage of multi-sensor integration in overcoming the limitations of single-sensor data, providing a reliable tool for accurate crop mapping, and supporting sustainable agricultural practices in challenging environments.