Estimating leaf area index in semi-Mediterranean oak forests using Sentinel-2 and machine learning algorithms: cross-comparison with global satellite products

Leaf area index (LAI) provides crucial insights into forest canopy structure, vegetation density, and overall ecosystem functioning. In the Zagros Mountains of northwestern Iran, traditional silvopastoral practices rely on tree foliage as animal fodder for local communities. Understanding and estimating LAI in these forests can therefore support the adoption of appropriate strategies for sustainable management. This study aims to map LAI in the Zagros semi-Mediterranean Forests (ZMFs) in Iran using Machine Learning Regression Algorithms (MLRAs) combined with Sentinel-2 (S-2) data and a Digital Elevation Model (DEM). We employed three MLRAs, i.e., Support Vector Regression (SVR), K-Nearest Neighbors (KNN), and Random Forest (RF). To train and evaluate the algorithms’ performance, in situ LAI data were collected in 60 field plots measuring 20×20 m. The results revealed that the RF algorithm performed best, with R2 = 0.82, RMSE = 0.16 m2 m-2, and rRMSE = 16.84. We further cross-compared the generated LAI map with three satellite-derived LAI products: Sentinel-2 SNAP, MODIS, and PROBA-V. The results indicated that the uncertainty of global LAI products varies from 30% to 55% higher than that of our applied approach. The study concludes that integrating S-2 data with DEM and MLRAs provides an accurate and reliable method for mapping LAI in open-canopy ZMFs. However, global LAI products show higher uncertainty because of lower spatial resolution and the lack of suitable empirical transfer functions for these forest types.