Browsing College of Natural Science and Mathematics (CNSM) by Subject "Detection"
Now showing items 1-1 of 1
Satellite remote sensing of active wildfires in Alaska's boreal forestThis research addresses improvements to the detection and characterization of active wildfires in Alaska with satellite-based sensors. The VIIRS I-band Fire Detection Algorithm for High Latitudes (VIFDAHL) was developed and evaluated against existing active fire products from the Visible Infrared Imaging Radiometer Suite (VIIRS) and the Moderate Resolution Imaging Spectroradiometer (MODIS). This new algorithm is based on VIIRS 375 m spatial resolution imagery and was tuned using fires in Alaska's boreal forest. It provides improved fire detection of low-intensity fires, especially during daytime and at sensor zenith angles smaller than approximately 50° off nadir. Low-intensity active fires, which represent residual combustion present after the passage of a high-intensity fire front, are not very well detected by existing active fire products. A second topic was fire remote sensing with ~30 m resolution imaging spectrometer (or hyperspectral instrument), the Hyperion sensor on NASA's EO-1 spacecraft, which was in use from 2000 to 2016. Hyperion had a much higher spectral resolution than VIIRS or MODIS, but no repeat imagery of the same active fire was available in Alaska. The investigation relied on absorption and emission features in the radiance spectra acquired at every pixel location. Three fire detection methods were evaluated using archived Hyperion data from three fires in interior Alaska from 2004 and 2009: A version of the Hyperspectral Fire Detection Algorithm (HFDI) produced excellent active fire maps; an approach that relies on a shortwave infrared carbon dioxide absorption feature and associated Continuum Interpolated Band Ratio (CO₂ CIBR) proved to be useful, but was affected by sensor noise and clouds; finally, a potassium emission feature from biomass burning was not detectable in the Hyperion data. Fire temperatures were determined using the Hyperion shortwave infrared spectra between 1400 nm and 2400 nm. The temperatures of active fire, the corresponding partial pixel areas, and the pixel areas occupied by unburned and already-burned vegetation, respectively, were modeled within each fire pixel. A model with two reflected background components and two temperature endmembers, applied to the same three study scenes, yielded an excellent fit to Hyperion spectral radiance data. Fire temperatures ranged from approximately 500-600 K to approximately 800-900 K. The retrieved lower fire temperatures are within the range of smoldering combustion; high-temperature values are limited by Hyperion's saturation behavior. High-temperature fire occupying 0.2% of a pixel (2 m²) was detectable. Sub-pixel fire area and temperature were also retrieved using VIIRS 750 m (M-band) data, with comparable results. Uncertainties were evaluated using a Monte Carlo simulation. This work offers insight into the sensitivity of fire detection products to time of day (largely due to overpass timing), spatial distribution over the study area (largely due to orbital properties) and sensor zenith angle. The results are relevant for sensor and algorithm design regarding the use of new multi- and hyperspectral sensors for fire science in the northern high latitudes. Data products resulting from this research were designed to be suitable for supporting fire management with an emphasis on real-time applications and also address less time-sensitive questions such as retrievals of fire temperature and time series of fire evolution.