Retrieval of Fuel Moisture Content from hyperspectral data via Partial Least Square

As an important indicator of vegetation moisture status, Fuel Moisture Content (FMC) is commonly used for predicting vulnerability to wild fire. Currently, the FMC estimation using spectral data is mainly based on spectral indices derived from several bands and these methods do not make full use of the entire spectrum. Partial Least Square (PLS) is a new multivariate statistical method which can effectively reduce collinearity. In this paper, using LOPEX dataset, we mainly explored the performance of PLS coupled with different feature selection methods for FMC retrieval. According to the results, PLS shows great potential to extract FMC from spectral data; when coupled with different band selection approaches, the models also generate high estimation precision; with band selection, the PLS coupled models involved fewer bands, lowering the model complexity. Thus, the high estimation precision and much simpler modeling make band selection-PLS coupled methods superior to original PLS for FMC retrieval.