Evaluating Multi-Temporal Optical and Radar Images for Crop Area Mapping in the US

Accurate mapping of agricultural lands is critical for resource management, economic forecasting, and food security planning. Optical and radar remote sensing are effective in distinguishing crop from non-crop land due to their complementary abilities to capture unique agricultural characteristics. Although efforts have been made to integrate both data sources, these typically involve computationally intensive workflows. In this research, I evaluated the effectiveness of a coefficient of variation (CV)-based approach for agricultural land mapping using moderate-resolution (10 m) optical and Synthetic Aperture Radar (SAR) imagery collected throughout the growing season (May to October). Specifically, I compared Sentinel-2 NDVI and Sentinel-1 VH polarization texture metrics (Angular Second Moment [ASM] and Entropy) for their ability to classify crop versus non-crop areas. Among evaluated features, NDVI-based CV consistently achieved the highest accuracy (95.8%), followed by ASM (91.6%) and Entropy (86%). Additionally, generalized thresholds (Average Youden’s theshold) derived from calibration consistently performed equal to or better than site-specific thresholds, especially for NDVI, highlighting their practical applicability in broad-scale operational mapping efforts. Furthermore, a high correlation (Pearson’s r > 0.72) between NDVI and ASM-derived CV metrics indicates that both metrics capture complementary aspects of agricultural landscape variability. This relationship is modulated by landscape homogeneity, with higher correlations associated with increased homogeneity. The integration of optical and radar-derived metrics, particularly NDVI and ASM, is recommended to enhance classification reliability. This integrated radar-optical approach, combined with strategically generalized thresholds, provides a practical, scalable, and highly accurate framework for operational agricultural land classification.