Estimation of grassland leaf area index by remote sensing under low-coverage conditions

Accurately estimating the leaf area index (LAI) of low-coverage grasslands holds significant importance for monitoring their growth conditions and optimizing grassland management practices. Previous studies of LAI have primarily focused on grasslands with medium-to-high coverage, and relatively limited attention has been given to low-coverage grasslands. Leveraging Google Earth Engine (GEE), essential feature variables were extracted from Landsat-8 satellite data. Feature selection was performed based on the correlation of these variables with LAI values and their importance within the model, determining the optimal number of variables to construct a machine-learning model for estimating the LAI value in low-coverage grassland areas. The results indicated that the gradient boosting regression tree model selected based on feature correlation performed well in estimating the LAI value in low-coverage grasslands, with a test set coefficient of determination (R²) of 0.686 and a root mean square error of 0.101. These findings suggest that machine-learning models constructed using feature selection have significant practical utility in estimating LAI values in low-coverage grassland conditions.