Forest recovery after wildfire disturbance in Great Xing' an Mountains by Multiple Endmember Spectral Mixture Analysis

Fire is one of the main natural disturbances to regulate the forest structure, function, and dynamics in boreal region. Post-fire ecosystem recovery usually consists of a mixture of vegetation and substrate, which is represented as mixed pixels in low- and medium-resolution (> 10 m) remote sensing images. Therefore, analyzing the ecosystem recovery at the sub-pixel scale is the key to monitor the post-fire ecological process by remote sensing image monitoring. In this analysis, a burn patch of 8700 hm² in Huzhong nature reserve of Great Xing'an Mountains in 2000 was chosen as a case study. Two medium resolution Landsat ETM+ images (30 m) on June 1, 2014 and June 22, 2010 were chosen to monitor post-fire recovery. The green vegetation cover was calculated from Multiple Endmember Spectral Mixture Analysis (MESMA) and a traditional vegetation index (e.g. Normalized Vegetation Index, NDVI), and then validated by a high spatial resolution (2 m) WorldView-2 image (July 1, 2014). The results showed that the accuracy of the green vegetation cover was not significantly different between NDVI (R² = 0.700) and MESMA method (R² = 0.691). The green vegetation cover was higher under medium severity burn than that under low and high severity burn. To assess the transferability of best endmembers, we applied best endmembers selected for image acquired in 2014 to image acquired in 2010. The mean RMSE for image acquired in 2014 and 2010 is 0.0015 and 0.0065, respectively, indicating that the best endmembers can be used as transfer among different Landsat images. This study shows that the MESMA method can effectively monitor the post-fire vegetation in northern boreal forests, and can be used to monitor the dynamics of post-fire ecosystem recovery with time series remote sensing images.