Estimating crop biophysical parameters using self-supervised learning with foudnation models and SAR-optical observations

Authors: HASHEMI, M - Michigan State University; ALEMOHAMMAD, H - Clark University; JALILVAND, E - Nasa Goddard Institute For Space Studies; TAN, P - Michigan State University; JUDGE, J - University Of Florida; Cosh, Michael; DAS, N - Michigan State University

Submitted to: Remote Sensing of Environment
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 7/1/2025
Publication Date: 9/1/2025
Citation: Hashemi, M., Alemohammad, H., Jalilvand, E., Tan, P.N., Judge, J., Cosh, M.H., Das, N. 2025. Estimating crop biophysical parameters using self-supervised learning with foudnation models and SAR-optical observations. Remote Sensing of Environment. 327. Article e114825. https://doi.org/10.1016/j.rse.2025.114825.
DOI: https://doi.org/10.1016/j.rse.2025.114825

Interpretive Summary: Vegetation biophysical parameters are useful to estimate for many applications and decision making tools. These estimates can be made from satellite observations, typically in the optical or near optical spectrum. However, these estimates need a variety of other land surface parameters, along with the satellite observations, to produce reasonable and accurate estimates. Therefore, a study was conducted using several field experiments to quantify the impact of various land surface parameters on the production of biophysical characteristics. This work will be useful for satellite missions and landscape modelers who need this information as a part of their operational work.

Technical Abstract: Accurate knowledge of vegetation water content (VWC) and crop height is crucial for many applications providing societal benefits and for satellite-based retrieval algorithms for geophysical variables. Traditional methods to estimate VWC primarily rely on optical indices, which has limitations of biomass saturation, and sensitivity to atmospheric conditions. This study introduces a novel application of geospatial foundation models (FMs), leveraging extensive, unlabeled datasets through self-supervised learning to enhance the skill of VWC and crop height estimation. We developed a comprehensive model integrating Sentinel-1A C-band SAR and Sentinel-2 indices with weather parameters to estimate soybean and corn VWC and crop height. Our research study area spans a variety of climatic zones and management practices, from the humid continental climate of Iowa and Michigan to the subtropical environment of Florida, encompassing both irrigated and non-irrigated fields as well as diverse tillage practices. We compared the performance of Single-Task Learning FM (STL-FM), Multi-Task Learning FM, Random Forest (RF), and XGBoost (XGB) to evaluate their effectiveness in estimating VWC and crop height. Results demonstrated that STL-FM outperforms other methods in accuracy and generalizability. For VWC estimation, STL-FM achieved R² values of 0.90 and 0.89 for soybean and corn, respectively. For crop height, R² values reached 0.95 for soybean and 0.98 for corn. The integration of SAR, optical, and climate data provided more reliable estimations than using individual data sources. Feature importance analysis identified NDVI, NDWI, VH backscatter, and precipitation as key drivers for accurate VWC and height estimations. The red-edge band emerged as significant for VWC estimation but showed limited importance for height prediction. Notably, surface roughness demonstrated a substantial impact on corn VWC and height estimations, while soil moisture exhibited less influence than initially anticipated. Notably, without directly incorporating soil moisture and surface roughness data, but by including diverse field conditions in training and validation, the STL-FM model demonstrated strong generalization capabilities. This study highlights the potential of geospatial FMs in advancing crop monitoring techniques, offering more reliable data for precision agriculture, and supporting sustainable farming practices across diverse agricultural landscapes.