Effects of Spatial Extent and Resolution on Soybean LAI, Biomass Estimation and Model Transferability

Timely monitoring of the development of crops throughout the growing season is essential for informed decision-making regarding the mitigation of climate-related risks and promotion of food sustainability. Two crop growth indicators—leaf area index (LAI) and aboveground biomass (AGB)—are important for monitoring overall crop productivity. In the past few decades, remote sensing has been proven a promising approach for estimating soybean LAI and AGB at the field scale. Moreover, previous studies have estimated soybean LAI and AGB at different geographic extents, especially at smaller scales. However, the impact of spatial extent and satellite sensor spatial resolution on estimating soybean LAI and AGB and transferring prediction models across different spatial extents and resolutions remains underexplored. This study evaluated the influence of spatial extent (1 m2, 25 m2, and 100m2) and spatial resolution (UAV-RGB, PlanetScope, and Sentinel-2) on soybean LAI and AGB estimation using Stepwise Multiple Linear Regression (SMLR) and Random Forest (RF) models. Model transferability across spatial extents and resolutions was also assessed. LAI estimation accuracy improved with increasing spatial extent, with the RF model achieving the best performance at 100 m2 (R2 = 0.32, rRMSE = 11.42%). For AGB, the highest accuracy was obtained by the SMLR model at 100 m2 (R2 = 0.22, rRMSE = 35.02%). Across spatial resolutions, Sentinel-2 combined with RF yielded the most accurate estimates for both LAI (R2 = 0.45, rRMSE = 10.34%) and AGB (R2 = 0.23, rRMSE = 33.80%). Cross-scale analysis demonstrated robust transferability of RF models for LAI across spatial extents, whereas SMLR models showed poor transferability for AGB. In contrast, both models exhibited limited transferability across spatial resolutions. These findings highlight the importance of spatial scale in soybean LAI and AGB estimation and model transferability.