Estimating leaf area index of salt marsh vegetation using airborne hyperspectral data

Disclaimer This document describes work undertaken as part of a programme of study at the International Institute for Geo-information Science and Earth Observation. All views and opinions expressed therein remain the sole responsibility of the author, and do not necessarily represent those of the institute. Abstract Information on Leaf area index (LAI) can be important for an improved understanding of the ecological processes of the salt marsh ecosystem. It is very difficult to estimate LAI given the heterogeneity and large spatial gradients within salt marshes. Hyperspectral remote sensing has demonstrated wide applicability in the area of estimating and mapping LAI of forest and agricultural ecosystems. This study attempts to evaluate different empirical models in estimating the LAI of salt marsh area from a hyperspectral remote sensing data. The empirical models were derived from three different regression techniques: simple linear regression, partial least square regression and neural network. Narrow band optimal normalized difference vegetation index (optimal NDVI) and optimal modified soil adjusted vegetation index (optimal MSAVI) derived from all possible two-band combinations were used in simple linear regression. The full spectrum data (455nm-1622nm) and the useful channels selected by a genetic algorithm were used in partial least square regressions. Feature extractions for neural networks were carried out using principle component analysis and minimum noise fraction. Cross validation procedure was used to assess the prediction power of the regression models. Analyses were performed on the entire data set (subset A, n=78) or on subsets stratified as high marsh (subset H, n=45), low marsh (subset L, n=33) and elymus vegetation type (subset E, n=15). Partial least square regression coupled with genetic algorithm provided the highest accuracy in predicting the LAI of the study area. Neural networks performed inconsistently with different subsets. Neural network provided higher accuracy than simple linear regression when entire dataset (n=78) was used in analysis. For subsets, H, L and E, simple linear regressions provided higher accuracies than neural networks. The study showed that the partial least square regression can be a useful tool for the estimation of LAI of the study area using hyperspectral data. Partial least square regression can utilize a greater number of channels of hyperspectral data than simple linear regression; whereas it was found to work well with minimum data size compared to neural network. The highest accuracy obtained for subset E emphasized the need of building vegetation type specific model to estimate LAI of the …