Detection of Banana Fusarium TR4 Disease in Lebanon Using Satellite Data and Geospatial Machine Learning

Fusarium wilt Tropical Race 4 (TR4), caused by Fusarium oxysporum f. sp. cubense, threatens Cavendish banana production worldwide, with no effective soil recovery once infection is established. In Lebanon, where coastal banana cultivation supports smallholder livelihoods and regional export, TR4 was confirmed in 2013, yet surveillance has remained reactive and spatially limited. This study develops an integrated satellite-based framework for banana plantation mapping and TR4 detection along Lebanon's coastal plain using Sentinel-2 and Landsat-8 imagery, spectral-indices (NDVI, EVI, NDMI, NDWI), topographic variables, and Random Forest classification in Google Earth Engine. Ground truth data of 1,282 healthy and 307 infected plants were collected along the Lebanese coast in 2023. Random Forest, proving its superiority in remote sensing-based plant disease classification, was selected as the classifier for TR4 detection in this study. For plantation mapping, Landsat-8 achieved 84.74% accuracy (98.75% banana-class precision) against 76.4% for Sentinel-2, with topographic variables contributing roughly 70% of feature importance across both sensors. For TR4 detection, Sentinel-2 reached 84% post-processing map accuracy versus 73.15% for Landsat-8, driven by balanced contributions from both four spectral-indices reflecting TR4's effects on vegetation vigor and canopy moisture. Infected areas were stratified into mild, moderate, and severe classes via k-means clustering on standardized indices. Multi-sensor analysis from 2014 to 2024 revealed a 50% contraction in total banana cultivation area following TR4 confirmation, with infection persisting at 19.8% of plantation extent in 2024. These findings establish satellite-based Random Forest classification as a scalable framework for TR4 surveillance and proactive agricultural biosecurity in Lebanon.