Spatiotemporal analysis of wildfire in the Tigris and Euphrates basin: A remote sensing based wildfire potential mapping

Abstract

In recent years, there has been a significant increase in the occurrence of wildfires in rangelands and forests worldwide. Such events are even more critical in arid and semi-arid regions with sparse rangeland and forest covers. In this paper, we studied how wildfires behave and what drives them, and used a random forest algorithm to map wildfire potential in the Tigris and Euphrates basin. A stratified random sampling method was used to collect 53,053 wildfire points and an equal number of non-wildfire points in rangelands and forests of the basin between 2000 and 2019. The spatial-temporal behavior of environmental parameters affecting wildfire potential was analyzed. Among them, precipitation, temperature, wind speed, vegetation condition index, drought index, actual evapotranspiration, and reference evapotranspiration were identified as the main drivers of wildfires. The performance of our wildfire potential mapping model was compared with the Fire Weather Index (FWI). Our method showed superior results with an overall accuracy of 96% and a kappa coefficient of 93% compared to the FWI which had an overall accuracy of 58% and a kappa coefficient of 17.5%.