Study On Forest Fire Damage Monitoring Method Based On HJ-1Satellite Data

Over the last decades, forest fire research has been receiving increasing attentionthroughout the world, due to the wide range of economic, social and ecological influence. Atglobal scale, biomass burning has been recognized as an important source of greenhouse gasemissions and aerosols, which has been amply demonstrated to be a driver of global climatechange. According to current estimates, the global biomass burning accounts for approximately40%of total carbon emission on average. At local scale, forest fire destroys forest structure andenvironment, leading to the unbalance of ecosystem and productivity decline. In addition, theatmospheric pollution can cause degradation in the air quality that may affect the health of thelocals affected in the region. Remote sensing due to its synoptic and repetitive coverage canprovide global to local monitoring of forest fire over extended periods of time. Satellitesystems have proved useful in detecting and monitoring of fires for three primary purposes:identification of active fires, mapping of burned areas, post-fire and characterization of fires.This paper utilized the spectral and temperature change after forest fire based on HJ-CCDand HJ-IRS data to evaluate forest fire. We developed a burned area mapping method and aburn severity rating system that is suitable for this data, which is instructive for the forest fireevaluation application of similar middle and high resolution data. The main contents andresults are as follows:(1) Multi-temporal burned area mapping method based on HJ-CCD data. A modifiedregion grow algorithm for burned scar mapping was utilized. There are two key points whichcould significantly influence the accuracy of region grow algorithm: selecting a set of one ormore starting points (“seed” pixels) and the formulation of a stopping rule. This study presentsa methodology to obtain “seed” pixels based on the magnitude between the modulus of changevectors before and after the fire which consists of multi-spectral indices (Near Infrared,Normalized Difference Vegetation Index, Enhanced Vegetation Index, Global Environment Monitoring Index). The formulation of a stopping rule was determined by a binary logisticmodel which was a very effective method to discriminate the burned pixels. The algorithm hasbeen validated using a set of HJ-CCD data in China. The validation results showed that theoverall accuracy of the burned area maps is higher than90%, better than the region growingaltorithm results based on “seed” pixels by using the threshold method of spectral indices.(2) Uni-temporal burned area mapping method based on HJ-1B-CCD and HJ-1B-IRS.The problems of optimal parameters selection and thresholds setting for forest fire burned areamapping have been very difficult to address. In this paper, HJ-1B-CCD data and HJ-1B-IRSwere combined to make a contribution to the spectral indices. An ordered weighting averagingoperator(OWA) based fuzzy set theory was applied to aggregate the positive evidence andnegative evidence which was used to revise the positive information to reduce the commissionerror. Thermal infrared band was added to aggregate the negative evidence to test its validity.Then, the revised positive evidence was input for a region grow algorithm to produce the result.The performance of the method was tested for two HJ-CCD/IRS images of Skovorodino inRussia and Xunke County in Hei Longjiang Province in China. The results show that theoverall accuracy is higher than85%in the research areas, which indicates that the methodproposed in this paper could meet the application need for burned area mapping.(3) The burn severity has been evaluated based on the combination of HJ-1B-CCD andHJ-1B-IRS data. The fusion of resampled HJ-1B-IRS data and HJ-1B-CCD data was applied toproduce a different Normalized Burned Ratio(NBR) to classify the burn severity.