Symbiotic Artificial Intelligence and satellite imagery for rapid assessment of war-induced agricultural damage in Ukraine

This study aims to enhance damage detection methods for the agricultural sector in Ukraine, which has been severely affected by ongoing conflict. While existing approaches, such as the method by Kussul et al. (2023), are among the best for monitoring damage, they are limited by the use of static threshold coefficients that can lead to inaccurate results, particularly false positives. To address these issues, we introduce a new approach using Symbiotic Artificial Intelligence (SAI), which integrates human oversight with machine learning to enable real-time adjustments to detection sensitivity based on field-specific characteristics. The proposed SAI-based method was tested using high-resolution satellite imagery from MAXAR for fields in Donetsk and Kherson. Results demonstrated a significant reduction in false positive rates, from 8.5% to approximately 1%, while maintaining a high rate of correctly identified undamaged areas. However, a slight decrease in true positive detections was observed, indicating a necessary balance between false positive reduction and sensitivity to actual damage. The SAI method effectively minimized false detections at field boundaries and other non-damage-related anomalies. This approach showcases the potential of combining human expertise with AI to improve accuracy and adaptability in damage detection. While the results are promising, further research should focus on automating the adjustment of detection thresholds for broader application, such as developing regression models to optimize field-specific coefficients.