Work

Improving the transferability of SAR-derived vegetation indices for crop monitoring

Year: 2026

Type: article

Abstract: • SAR enables frequent crop monitoring when optical data are limited by clouds • VH backscatter is the strongest SAR predictor for NDVI/EVI in corn–soybean • DAP and AGDD improve accuracy and cross-region transfer of SAR-derived VIs • Transferability drops in dry/subtropical zones due to moisture and biotic stress Optical vegetation indices (VIs) are widely used for crop monitoring, yet their operational value is often limited by cloud contamination and data gaps. Synthetic aperture radar (SAR) provides cloud-independent observations and is sensitive to crop structure, offering a pathway to estimate VIs when optical imagery is unavailable. A key challenge, however, is the transferability of SAR-based VI models across regions and seasons due to environmental and climatic differences. Here, we evaluated whether incorporating time-related proxies of crop growth such as accumulated growing degree days (AGDD) and days after planting (DAP), together with environmental variables can improve the transferability of SAR-estimated VIs across major U.S. corn and soybean production regions. Using dense Sentinel-1 SAR with Harmonized Landsat–Sentinel (HLS) reference data, we trained LightGBM regression models under multiple predictor configurations. Across climate zones, adding time-related features consistently improved generalization relative to SAR-only baselines, with DAP providing the most robust performance. For NDVI, the DAP-based model achieved MAE values ranging from 0.083 to 0.115 and R² scores between 0.422 and 0.781. For EVI, the model achieved MAE values ranging from 0.081 to 0.111 and R² scores between 0.388 and 0.721. Performance peaked in humid continental zones (Dfa/Dfb) but declined in semi-arid (BSk) and subtropical zones (Cfa), where unaccounted moisture and biotic stressors limited spatial transferability. This study offers insights into developing a scalable framework that ensures high-frequency crop monitoring across diverse geographies.

Source: Environmental Challenges

DOI: 10.1016/j.envc.2026.101411

Authors: Henry Poku Osei, Anita Simic Milas, Mir Md Tasnim Alam

Institution: Bowling Green State University

Language: English

Topic: Remote Sensing in Agriculture

Subfield: Ecology

Field: Environmental Science

Domain: Physical Sciences

SDGs: Zero hunger, Life on land

Open Access status: gold

Estimated APC paid: $2,140