Quantitative evaluation of active fire detection capabilities from VIIRS

The VIIRS (Visible/Infrared/Imager/Radiometer Suite) instrument on board the JPSS (Joint Polar Satellite System) and the NPOESS Preparatory Project (NPP) satellites will provide radiometric measurements that offer useful information for the detection of active fires. The baseline algorithm for the VIIRS Active Fire Mask Application Related Product (ARP) uses primarily the medium resolution M13 (4.05 μm) and M15 (10.76 μm) measurements, aggregated from the native resolution observations into medium resolution pixels according to a scheme aimed at reducing the decrease of resolution towards the edges of the scanline. In this work we analyzed the fire detection potential of the VIIRS sensors, including the use of native spatial resolution radiometric measurements. Such a product would potentially enable earlier detection of fires to support fire early warning, monitoring and management. For the analysis we developed a modeling framework using a set of coincident fire observations by the ASTER (Advanced Spaceborne Thermal Emission and Reflection Radiometer) and MODIS (Moderate Resolution Imaging Spectroradiometer) sensors on the NASA Earth Observing System Terra satellite. The cases analyzed included a wide range of fire characteristics, and environmental and observing conditions. MODIS can be considered as the precursor instrument to VIIRS, providing similar radiometric measurements for fire observations at 1km nadir resolution for the mid-IR and IR bands. The principal issue of using MODIS as a proxy for VIIRS for fire pixels is the need to account for the variation of the sub-pixel area of active burning and the resulting differences between the integrated radiometric signals over the MODIS and VIIRS pixel footprints. This is particularly important when attempting to simulate VIIRS radiances in the presence of fires at the native resolution. To perform the necessary geometric transformation (in addition to a radiometric adjustment) we used the fire signal from the 30m shortwave infrared (SWIR) bands of coincident ASTER observations, along with 90m thermal infrared ASTER data to characterize background thermal conditions. The pairs of MODIS-ASTER scenes were used for both generating more realistic VIIRS proxy data for a number of fires and for fire detection algorithm refinement and testing. The cases analyzed are a subset of a dataset of ~3000 scenes used for the global validation of the standard MOD14 product using ASTER data mapped into MODIS pixel footprints as reference to determine sub-pixel fire characteristics. The simulation of the MIR M13 radiance values was done by calculating the relative contribution of each thermal component within a 30m grid, as defined by the ASTER SWIR pixels, over the medium resolution pixel footprint. For the fire-free gridcells we used the ASTER Surface Kinetic Temperature product (AST08), remapped to 30m. We used Planck's equation and sensor spectral response functions for the spectral transformation, and Point Spread Functions defined for MODIS for the spatial weighting of each thermal component. To define the spatial extent and location of active burning within the pixel footprint, we used ASTER fire masks derived from L1B ASTER radiances. The areas defined by the 30m fire gridcells were used as a constraint for actual fire area. Fire temperatures were then retrieved using an iterative process to match the simulated radiances with the actual MODIS observations. In these series of forward calculations the mid-IR the contribution of the reflected solar radiation and atmospheric attenuation was also considered. Simulated VIIRS radiances were then calculated as a weighted sum of thermal components over the VIIRS pixel footprint, using the previously retrieved fire temperatures. Radiance values for the top-of-atmosphere TIR M15 radiances were generated from 90m ASTER band 13 (10.6 µm) and 14 (11.3 µm) Level-1B top-of-atmosphere radiances. The pixel-integrated VIIRS radiances were calculated by integrating the contribution of the individual thermal components estimated from ASTER, including those containing active burning. The baseline for the proposed VIIRS fire detection algorithm is the latest version of the algorithm used to generate the standard MODIS fire product (MOD14) within NASA's MODAPS processing system. This is an improved version of a hybrid thresholding and contextual algorithm to separate the thermal signal of active fires from fire-free background both radiometrically and spatially. Necessary adjustments for the VIIRS sensor characteristics were determined and evaluated using the proxy dataset described above. One of the fundamental adjustments was the tuning of the thresholds used in the contextual tests to account for the increased spatial resolution of VIIRS compared to MODIS and the consequent increase of surface heterogeneity detected by MODIS.