Assessing the Potential of Earth Observation Data to Differentiate Between Burned Area and Harvested Agricultural Land

Abstract

Fire is a common disturbance for natural ecosystems, which could bring costly effects not only to humans but also global environments. Forest fires, especially, are closely connected with the global carbon cycle and climate modelling. Therefore, fire management system is essential to estimate the extent of the damage and to model the impact on the climate as well as for forestry planning. Precise knowledge regarding the location the extent and the frequency of the fire incidents are important when monitoring fires. Remote sensing techniques provide a promising potential for this purpose, since the earth's surface can be observed in a large-scale and with high frequency. In this study, the potential of various Earth observation data were researched to distinguish between a real burn and harvested agricultural lands to map the fire in a more reliable and fast way. Sentinel-1 and -2 data as well as MODIS thermal data were adopted for the discriminant analysis, and active hotspot data and land cover were used for probabilistic approach. Discriminant analysis did not show a distinct feature space to classify between the burned area and harvested agricultural lands. Consequently, a probabilistic approach was considered to exclude the crop fields. The results showed an average producer's accuracy of 80 % and user's accuracy of 89 %, which slightly improved the classification after the probabilistic approach was implemented.