Modelling the spatial pattern of forest fires using NDVI and land surface temperature in Southern African landscapes

Abstract

In this study we used maxent to test whether; and for which month of the growing season SPOT derived normalized difference vegetation index (SPOT NDVI) data can be used to predict fire occurrence in August, September and October (AUC>0.5). We also tested whether and for which time of the day MSG SEVIRI derived land surface temperature (MSG SEVIRI LST) data can be used to predict forest fire occurrence. Results of receiver operating characteristic (ROC) curve show that both NDVI and land surface temperature data have high predictive capacity for forest fire occurrence (AUC>0.5). Furthermore, our results of jacknife of regularized training gain for fire show that March NDVI data significantly predict fires that occur in August and September while April NDVI data significantly predict fires that occur in October and mid-day temperatures (1200-1300pm) are important in predicting fire occurrence.