Sensitivity Analysis Of Forest Fire Burn Area Based On Spectral Index And Sar Polarization Characteristic Parameters

The frequent occurrence of forest fires not only damages the balance of local ecosystem,but also has many adverse impacts on people’s living conditions.The extraction of forest fire information can not only reveal the changes of forest ecosystem after forest fire,but also continuously monitor the vegetation and temperature in the burned blanks,and provide essential guidance and reference for post-disaster restoration and,soil remediation.Optical and radar remote sensing technology is currently the optimal way to obtain large-scale forest fire information.One method based on optical remote sensing is mainly observing the sudden change of both spectral reflectance and remote sensing index caused by forest fire,the other method based on synthetic aperture radar（SAR）is commonly using vegetation structure,biomass and soil moisture inside polarimetric scattering.However,optical remote sensing is easily limited by rain and cloud,and it is hard to interpret some radar information.Therefore,it can offer an approach for multi-source remote sensing data based research to integrate the advantages and applicability of spectral index and SAR polarization characteristic parameters when extracting forest fire information.Mostly,complement of optical and radar remote sensing can be much more timely and reliable when monitoring and observing forest fire,and then strengthen the technical support of forest fire emergency response and forest fire trace research.In this paper,based on“the 3·30 forest fire in Xichang City in 2020”,based on sentinel-1A and sentinel-2 data,18 parameters including GEMI、SAVI、NBR、MIRBI、Span、r_vv-vh and Alpha were extracted from spectral index and polarization characteristic parameters,then the mean difference of these parameters before and after forest fire and the sensitivity of extracting forest fire information were analyzed,so the main onclusions of this paper include the following aspects:（1）Compared with before the forest fire the biggest mean value change of vegetation index and burn index appears in RENDVI and NDSWIR in the burned area,which respectively decreasing by 60%and 214%after forest fire;The biggest parameter change of backscattering intensity,depolarization and scattering mechanism respectively appears in Span,σ_vv and Anisotropy,which respectively decreasing by23%and 30%and the Anisotropy increasing by 15%;In other area,the target parameters above have no obvious change;among the five target parameters,GEMI,SAVI,MIRBI,NBR,Span,r_vv-vh and Alpha are the most sensitive to the separation of the forest fire slash from the burned area before the fire and the unburned forest after the fire;（2）Among the forest fire sensitivity burn index,vegetation index and polarization characteristic parameters,NBR,GEMI and Span are the most availible to identify burned blanks and unburned area,with kappa coefficients of 0.94,0.88 and 0.82respectively;（3）The accuracy of fire sensitive vegetation index and burn index when estimating the burned area are all above 94%,among them,GEMI and MIRBI have higher estimation accuracy,which are 95.57%and 97.99%respectively;The accuracy of Span,r_vv-vhand Alpha’s estimation of the fire area is less than 80%,however,the estimated accuracy of the fired area from the combined multi-band data of the three can reach91.69%,compared with the results extracted separately,the burned area extracted by the combined multi-band data of the three is more referential.（4）Based on d NBR data as a reference,this paper concludes that GEMI,SAVI,MIRBI,Span,r_vv-vh and Alpha are separated in different degrees for different forest fire intensity levels,among them,SAVI has a higher degree of discrimination for all levels of forest fire intensity.Therefore,apart from NBR,SAVI is also suitable for grading forest fire intensity in this study area.