Estimation of forest leaf area index using satellite multispectral and synthetic aperture radar data in Iran

Abstract

Different satellite datasets, including multispectral Sentinel 2 and synthetic aperture radar Sentinel 1 and ALOS2, were tested to estimate the Leaf Area Index (LAI) in the Zagros forests, Ilam province, in Iran. Field data were collected in 61 sample plots by hemispherical photographs, to train and validate the LAI estimation models. Different satellite data combinations were used as input in regression models built with the following algorithms: Multiple Linear Regression, Random Forests, and Partial Least Square Regression. The results indicate that Leaf Area Index can be best estimated using integrated ALOS2 and Sentinel 2 data; these inputs generated the model with higher accuracy (R2 = 0.84). The combination of a single band and a vegetation index from Sentinel 2 also led to successful results (R2 = 0.81). Lower accuracy was obtained when using only ALOS 2 (R2 = 0.72), but this dataset is helpful where cloud coverage affects optical data. Sentinel 1 data was not useful for LAI prediction. The optimal model was based on the traditional Multiple Linear Regression algorithm, using a preliminary input selection step to exclude multicollinearity effects. To avoid this step, the use of Partial Least Square Regression may be an alternative, as this algorithm was able to produce estimates similar to those obtained with the best model.