Standardization Study Of Land Surface Temperature Inversion In Multi-source Remote Sensing Monitoring Of Forest Fires

The status and role of satellite disaster reduction application projects are highlighted in the national comprehensive disaster prevention and mitigation plan.The advantages of monitoring forest fire by satellite remote sensing are increasingly prominent as an important part of China’s comprehensive warning and monitoring system of forest fire.However,due to the different kinds of sensors carried by satellites,or the aging and decay of sensor after years,the acquired remote sensing images may have different geometric,radiometric and temporal characteristics.Therefore,it is difficult to interoperate the data among sensors.So the research on the standardization of multi-source sensors can help improve data quality and detection accuracy of forest fire.The paper outlines situations of world forest fires,reviews the current status and trends of remote sensing monitoring of forest fires,as well as current researches on land surface temperature inversion and land surface temperature normalization.Guided by forestry,remote sensing and geographic information science,the paper proposes a standardized method for land surface temperature inversion of forest fire monitoring based on multi-source remote sensing data using GIS and mathematical modelling methods.The aim is to reduce the differences between remote sensing images with different geometrical,radiometric and temporal characteristics,to achieve "data networking" across sensors,and to enhance the application capability of remote sensing data.The main research contents and results are as follows.(1)Analysing and selecting land surface temperature inversion algorithms of multi-source remote sensing imagery data.The current algorithms for land surface temperature inversion are relatively mature.But due to the specificity of the study area,the variability of sensor performance,and the uncertainty of external factors such as the atmosphere,the suitability of land surface temperature inversion algorithms needs to be analysed.Six land surface temperature inversion algorithms are compared through experiments to extract land surface temperature values with high accuracy.The results show that the most suitable land surface temperature inversion algorithms for each sensor in the study area are the NOAA-based split-window algorithm,the MODISbased diurnal method,the FY-4A-based split-window algorithm and the Himawari-8based split-window algorithm.(2)The study proposes a mixed-effects model normalisation(MEMN)method for land surface temperature inversion.Based on Himawari-8 data,the longitude,latitude,solar altitude angle,solar azimuth,surface emissivity,slope,slope direction,elevation and land surface temperature values of the sample pixel in the study area are collected as model parameters,combined with multiple stepwise regression and mixed-effects model.The mixed-effects model normalization(MEMN)method of land surface temperature inversion is proposed in multi-source remote sensing monitoring of forest fires.The method can satisfy both radiation normalisation and observation angle normalisation,and reduce the differences of fire points monitoring among multi-source sensors.The results show that the model has the decision coefficient(R2)of 0.8418 when tested using Himawari-8 data.Meanwhile,the MEMN method is more accurate than the random forest normalisation(RF)method(R2 of 0.7318),the pseudo-invariant feature(PIF)method(R2 of 0.7264)and the automatic scatter control regression(ASCR)algorithm(R2 of 0.6841)after comparison.(3)The paper constructs a standardized model for land surface temperature inversion.After normalizing the multi-source remote sensing images,the paper compares and calculates the mean and standard deviation between the land surface temperature values and the measured values of each satellite,and quantitatively analyses the mean and standard deviation between the land surface temperature inversion and the measured value.And Himawari-8 is identified as the reference satellite for the standardized model.A standardized model of forest fire multi-source remote sensing land surface temperature inversion with Himawari-8 as the reference is established.At the same time,the standardized model is validated by using NOAA,MODIS,FY-4A,Himawari-8 data and fire point statistics from the National Forest and Grassland Fire Prevention and Suppression Information Sharing Platform.The results show that the remote sensing data processed by the MEMN method and the standardized model can significantly improve the validity of detection rate of fire points.The detection rate of NOAA increases by 27.28%,that of MODIS increases by 27.28%and that of FY-4A increases by 18.18%.In summary,the MEMN method can effectively eliminate the "pseudo-variation"of remote sensing data.The method combined with the land surface temperature inversion standardisation model can meet the standardisation requirements among different sensors,realise the "data networking" between multi-source sensors,and significantly improve the detection rate of forest fires.