Evaluating the effects of spatial resolution on hyperspectral fire detection and temperature retrieval

Hyperspectral data covering a wavelength range of 1.2-2.5 mu m can be used to detect fires and model fire temperature and background land cover. Previous work has used hyperspectral data acquired from airborne platforms, limiting spatial resolution to finer than 20 m. The Hyperspectral InfraRed Imager (HyspIRI), a proposed hyperspectral/thermal infrared mission, will provide hyperspectral data over a spectral range of 0.35-2.5 mu m at a spatial resolution of 60 m. This study uses airborne hyperspectral data to investigate changes in modeled fire temperature and area as spatial resolution is varied from 5 m to coarser than 60 m. Four images containing active fires were acquired by the Airborne Visible Infrared Imaging Spectrometer (AVIRIS), with spatial resolutions ranging from similar to 5 to similar to 20 m. Gaussian and aggregation resampling methods were compared for one scene containing fire, and both resampling methods were found to produce similar radiance values. As spatial resolution coarsened, the area flagged as having fire by the hyperspectral fire detection index (HFDI) increased. Fire temperature modeled using a multiple endmember spectral mixing model decreased at coarser spatial resolutions, while the modeled fire fractional area increased. Coarser spatial resolution hyperspectral data, including data collected by HyspIRI, are likely to provide increased fire area and lower temperatures when compared against simultaneously acquired higher spatial resolution data. Saturation in shortwave infrared (SWIR) bands was found in all four images, and increasing SWIR saturation thresholds could lead to improvements in fire characterization. (C) 2012 Elsevier Inc. All rights reserved.