The Effect of Atmospheric Correction Approaches on Leaf Area Index Retrieval Accuracy in Heterogeneous Croplands

Abstract

This study evaluates four atmospheric correction approaches, including Sentinel-2 Correction (Sen2Cor), Rayleigh correction, image correction iCOR, and Dark Object Subtraction (DOS) for Leaf Area Index (LAI) retrieval in heterogeneous croplands in southern Mozambique, using the Sentinel Application Platform (SNAP) biophysical processor with PROSAIL neural network inversion. Field LAI measurements from 270 sample points across multiple vegetation types were used as reference data. The results showed that iCOR delivered the best overall performance, with a Root Mean Square Error (RMSE) of 1.315, a Mean Absolute Error (MAE) of 1.063, and a Bias of -0.684. This was followed by DOS, with an RMSE of 1.423, Sen2Cor with an RMSE of 1.514, and Rayleigh correction with an RMSE of 1.567. All methods exhibited negative coefficient of determination (R²) values from-1.223 to -2.155, indicating systematic LAI underestimation and limited predictive capability. Performance varied substantially across stratification levels. LAI-class analysis revealed that DOS performed the best for low LAI with an RMSE of 0.478, while iCOR excelled at medium LAI with an RMSE of 0.956, and high LAI with an RMSE of 1.929. Strong positive correlations between absolute error and field LAI (Pearson's correlation coefficient (r) = 0.733 to 0.895) indicated increasing retrieval challenges at higher canopy densities. Significant spatial variation was observed, with ESUDER Campus exhibiting lower errors (RMSE = 0.518-0.745) compared to Machengue (RMSE = 1.435-1.741). Vegetation-type specific analysis showed that Miombo Forest had the lowest errors (RMSE = 0.542 for DOS), while Mango orchards presented the greatest challenges (RMSE = 1.754 for iCOR). These findings demonstrate the importance of selecting appropriate atmospheric correction methods based on specific vegetation characteristics, canopy density, and local environmental conditions for operational LAI monitoring in heterogeneous agricultural landscapes.