Detecting the invisible enemy in maize: Machine learning classification of fall armyworm damage in maize

The fall armyworm (FAW), Spodoptera frugiperda, is a highly destructive pest of maize, exploiting the crop’s nutritional and structural characteristics and causing significant yield loss. This study investigates the potential of remote sensing to detect and classify FAW infestations in Bangladesh maize fields using freely available Sentinel-1 (radar backscatter) and Sentinel-2 (optical reflectance) satellite datasets. Field observations were conducted during the 2019–2020 maize growing season in maize fields across six administrative divisions, covering both infested and non-infested sites across various maize growth stages. The corresponding Sentinel images were downloaded and processed for analysis. Radar backscatter values, spectral reflectance profiles, and vegetation indices were extracted from the Sentinel data, and statistical tests were performed to identify significant differences between infested and non-infested maize fields. Machine learning models were used to classify infestation severity based on five different combinations of Sentinel data and vegetation indices as inputs. Additionally, the results were validated through cross-validation techniques to ensure model accuracy. Statistical tests revealed significant differences between infested and non-infested maize across growth stages. Infested fields showed reduced near-infrared reflectance (Sentinel-2) and distinct radar backscatter patterns (Sentinel-1 VH polarisation), with notable variations at silking and maturity stages. The Random Forest classifier demonstrated superior accuracy and robustness, especially when integrating multi-source data. The red edge (740 nm) and near-infrared (865 nm) bands were particularly effective in distinguishing infestation levels, while the combination of spectral and radar data significantly improved classification accuracy by leveraging their complementary strengths. These results underscore the potential of integrating multi-source remote sensing data for scalable and accurate pest monitoring. Freely available Sentinel satellite imagery is a valuable source of information for early pest detection and management, aiding policymakers in identifying high-risk areas, implementing timely interventions, and promoting sustainable pest management strategies to protect maize production and reduce economic losses.