A data-driven approach to estimate leaf area index for Landsat images over China

Leaf area index (LAI) is a fundamental parameter for assessing the structure and dynamics of terrestrial vegetation ecosystems. Current long-term LAI datasets are typically derived from medium-resolution satellite imagery, which limits their utility in fine-scale applications. We present a high-resolution LAI mapping algorithm using 30-meter Landsat surface reflectance data across China. The algorithm is based on ~390,000 LAI samples uniformly distributed across the country and integrates MODIS-derived LAI and Landsat reflectance as the target variable and primary predictor, respectively. For each of the eight major vegetation community types in China, a Random Forest model was trained using rigorously filtered and optimized samples. Cross-validation results indicated that the model achieved good accuracy (coefficient of determination (R²) = 0.899, bias = −0.007 m²/m², mean absolute error (MAE) = 0.180 m²/m², root mean square error (RMSE) = 0.382 m²/m², mean absolute percentage error (MAPE) = 23.8%, and normalized root mean square error (NRMSE) = 0.057), although the performance varied by vegetation type. An independent validation using 156 ground-based LAI measurements from the DIRECT V2.1 dataset for two vegetation community types (grass and cropland) yielded R2, bias, MAE, RMSE, MAPE, and NRMSE values of 0.595, −0.675 m²/m², 0.968 m²/m², 1.203 m²/m², 36.39 %, and 0.237, respectively.