Remote sensing techniques for vegetation moisture and fire risk estimation

This dissertation is aimed at evaluating and improving remote sensing techniques for vegetation moisture and fire risk estimation. Empirical retrievals of vegetation moisture using liquid water absorption based spectral indices such as the NDWI (Normalized Difference Water Index) and NDII (Normalized Difference Infrared Index) may have uncertainties, since these indices cannot fully normalize the reflectance variability due to other biophysical, biochemical, soil and illumination viewing geometry factors. Coupled leaf-canopy reflectance models, National Fire Danger Rating System data and the FARSITE fire behavior model were used to estimate the effect of Live Fuel Moisture Content (LFMC) retrieval uncertainties on fire spread rate predictions. The uncertainty estimation was focused on the Okefenokee National Wildlife Refuge where errors in LFMC retrievals using NDWI and NDII were shown to result in considerable fire spread rate prediction errors at lower LFMC levels. Soil reflectance contamination driven by soil moisture variability was identified as a problem causing errors in Vegetation Water Content (VWC) retrievals over low vegetation conditions. Analysis of canopy reflectance simulations from coupled soil-leaf-canopy reflectance models revealed that VWC isolines were curved and did not converge at the origin of the 1.64mum--0.86mum space. These were identified as causes for the soil moisture contamination of the spectral index NDII. As an improvement strategy an origin transformed NDII, called the SANDII (Soil Adjusted NDII) was designed to minimize soil contamination. Further separate regression models between VWC and the SANDII for different soil moisture classes were proposed to account for the curved nature of VWC isolines. The new technique which requires categorical soil moisture information was shown to reduce VWC estimation errors by about 20% over grassland conditions. The approach was supported using data collected over pastures during the Soil Moisture Experiment 2003 field campaign. Finally as an alternative to current subjective fire risk indices a new Fire Susceptibility Index (FSI) based on physical concept of pre-ignition energy was proposed. FSI uses remotely sensed estimations of fuel temperature and LFMC. Its physical basis is expected to allow computations of ignition probabilities and fire spread rates. FSI can be used compare fire risk across ecoregions and yet has the flexibility to be localized for an ecoregion for improved performance. A good agreement with the well tested FPI (Fire Potential Index) over Georgia, suggests the validity of FSI as a fire risk estimator. These new approaches would be helpful in fire risk monitoring, agriculture and climate studies.