Information requirements for accurate forward modeling of fire emissions on heterogeneous landscapes

Edward J. Hyer, Naval Research Laboratory,, edward.hyer@nrlmry.navy.mil (Presenting)
Jeffrey S. Reid, Naval Research Laboratory,, jeffrey.reid@nrlmry.navy.mil

Forward models of fire emissions have advanced considerably during the MODIS era as a result of broad-scale databases providing information on fuel amount, structure and condition, as well as improved fire detection systems. The traditional &ldquomap-based&rdquo approach to emissions modeling uses the location information from fire detection systems to extract information on fuels from spatial databases. The spatial resolution of fire detection systems and the accuracy of land cover maps are known sources of uncertainty in estimated emissions. We constructed a simple model based around a single piece of spatial information, namely, forest vs. nonforest. In many fire-affected ecosystems, this is the most important single determinant of intensive emissions, i.e. emissions per area burned. We demonstrate that the native resolution of fire detection systems from moderate-resolution sensors such as GOES and MODIS results in large uncertainties in determination of forest vs. non-forest fires. In addition to large uncertainties, this spatial error in many cases leads to significant biases in estimated emissions. Using a case study in the Amazon basin in South America, we demonstrate that using GOES active fire detections to assign land cover type of fire-affected areas results in a positive bias of 6-22% in estimated burning emissions. The bias resulting from use of MODIS active fire detections is smaller, but for both sensors, the spatial resolution strongly limits the ability of emission models to capture the spatial and temporal variability of emissions from heterogeneous landscapes.