| dc.description.abstract | Fire is a critical tool for managing rangeland ecosystems, particularly in the savannah. However, the increasing wildfire occurrence poses a considerable danger to rangeland ecosystem continuity. Burned area extent and fire severity have been mapped over the years using different methods. There is a need to avail tools and techniques to reliably and accurately map burned areas and fire severity early enough for improved management of fires and restoration of rangeland burned areas. Thus, predicting fire occurrence and mapping wildfire danger is critical in managing savannah rangelands. This study developed a random forest (RF) prediction model using observed fire occurrence points and selected environmental variables in Kgalagadi District, Botswana. A wildfire probability map was also developed using a Logistic regression model (LR) applied to best-performing variables. The study used 107,883 active fire points from Visible Infrared Imaging Radiometer Suite (VIIRS) sensors from 2015 to 2021 and randomly created non-fire points. A dataset of remotely sensed predictor variables was developed using ArcMap 10.7. These are Dry matter productivity (DMP), Soil moisture Content (SM), Land surface temperature (LST), Live Fuel Moisture content (LFMC), and Dead Fuel Moisture content (DMFC). The selected RF model with an Out of Bag (OOB) error of 9.91% had an overall accuracy of 90.15% for classifying fires and non-fires for the test dataset. Results show a Kappa coefficient of 0.803, with 88.25% producer accuracy and 91.76% user accuracy for classifying fires. The DMP was the most important variable (MDA= 1,055.20 and MDG= 9.328.62), followed by SM (MDA= 828.39 and MDG= 15,745). The LR model indicated a relatively weak but significant ability to discriminate fires from non-fire points with an overall accuracy of 56.05% and an Area Under the Curve (AUC) of 56.05%. A probability map produced using the LR model indicates that more than 39.11% of the study area had a high and very high chance of fire ignition before the 2021 Kgalagadi Mega fire. In this study, the burned area was estimated, and also fire severity assessed for the 2021 fires in the Kgalagadi District using the Monitoring Trends in Burn Severity (MTBS) Fire Mapping Tool (FMT) from Landsat 8 Operational Land Imager data. The burn area perimeter was delineated using the FMT tool and compared to aggregated Visible Infrared Imaging Radiometer Suite (VIIRS) active fires. Severity indices, including Normalized Burn Ratio (NBR), differenced NBR (dNBR), and Revitalised dNBR (RdNBR), were developed and assessed using sentinelderived indices. The results from the FMT severity indices also showed significant….. | en_US |
