Retrieving Leaf Area Index from Remotely Sensed Data Using Advanced Statistical Approaches

Abstract

Mapping and monitoring leaf area index (LAI) is critical to model surface energy balance, evapotranspiration, and vegetation productivity. Remote sensing helps in rapid collection of LAI on individual fields over large areas, in a time and cost-effective manner using empirical regression between LAI and spectral vegetation indices (SVI). However, these relationships may be ineffective when sun-surface sensor geometry, background reflectance and atmosphereinduced variations on canopy reflectance are larger than variations in the canopy itself. This requires development of superior and region-specific LAI-SVI models. In recent years, statistical learning methods such as support vector machines (SVM) and relevant vector machines (RVM) have been successful over the ordinary least square (OLS) regression models for complex processes. The objective of this study is to develop and compare OLS, SVM, and RVMbased reflectance models to estimate LAI for major summer crops in the Texas High Plains. The LAI was measured in 47 randomly selected commercial fields in Moore and Ochiltree counties. Data collection was made to coincide with Landsat 5 satellite overpasses on the study area. Numerous derivations of SVIs were examined for estimating LAI using OLS, SVM, and RVM models. Analysis of the results indicated that the SVI-LAI models based on the ratio of TM bands 4 and 3, and normalized difference vegetation index (NDVI) are most sensitive to LAI. The R2 values for selected models varied from 0.79 to 0.96 with the SVM model producing the best results. However, accuracy of reported LAI models needs further evaluation that accounts for in-field spatial variability in the LAI for wider applicability.