Accuracy assessment of MODIS fire products in African Savanna woodlands

Abstract

Fire poses a continuous threat to forest ecosystems and can dramatically reduce valuable timber species in forest woodland areas. The occurrence of fire in Baikiaea plurijuga woodlands warrants the need of an active and timely fire detection system. Rudimentary methods of detecting fires are still in use in most of Zimbabwe’s forest reserves, yet remote sensing has played a pivotal role around the globe in detecting and monitoring both active incidences and post fire burnt areas. Several satellite systems have been validated in different biomes of the world for both MODIS MOD14A1 and MCD14ML.However, there still remains a gap of knowledge in the accurate detection of fires in Baikiaea plurijuga woodlands. In this study we evaluate the accuracy of MODIS fire products using, confusion matrices, kappa statistic, true skill statistic (TSS), remote sensing and Geographical Information techniques where employed to assess the accuracy of MODIS MOD14A1 burnt area product and MCD14ML active fire product in fire detection. This is the first time results of accuracy assessment of fire products are reported in Baikiaea plurijuga woodlands. In both study sites, course resolution 1 km MODIS MOD14A1 burnt area fire product has a continually poor index of agreement with ground data kappa 0 and TSS value is 0. However, for MODIS MCD14ML we found high kappa and true skill statistic values, showing a high. However, we recorded high kappa and TSS values for the MODIS MCDML active fire product. These results, are consistent with the premise that, increase in spatial resolution reduces the sensors ability to detect fires in African Savanna woodlands