Dendrolimus Tabulaeformis Disaster Detection And Prediction With Remote Sensing Image At Multi-scale

As a main forest pest in north China the Dendrolimus tabulaeformis casues very serious damages to Pinus tabulaeformis and other host forest every year.One of the main reasons of the disaster high frequency of outbreak and wide prevalence is lack of the high precision and availability of the risk forecasting and occurrance real-time monitoring.Remote sensing technology makes it possible to quickly and accurately obtain types,degree,area and other information of the damage,so as to realize the dynamic monitoring and prediction of forest insect pests and diseases,which can provide powerful evidence for effective disaster prevention.This paper took Jianping county,Liaoning province,China as the study area,chose the D.tabulaeformis disaster as the main research object.And we used low-altitude unmanned aerial vehicle(UAV)hyperspectral data,high spatial-resolution CCD images and time series of satellite remote sensing images to research on the disaster degree identification,damaged area monitoring and prediction methods at individual tree,plot and regional scale respectively.Based on the dimensionality reduction of hyperspectral data,spectral-spatial classification and statistical analysis,a bands selection algorithm for hyperspectral images,"Instability Index between Classes-Successive Projection Algorithm"(ISIC-SPA)and optimized support vector machine spectral-spatial classification algorithm were proposed.Based on time series analysis of satellite images and meteorological data,a technological framework for regional scale disaster prediction under climate changes was proposed.On this basis,the D.tabulaeformis disaster degree identified model,the D.tabulaeformis disaster area monitoring and short-time prediction models were constructed at individual tree,plot and regional scale respectively.Thus,the D.tabulaeformis disaster monitoring and predicting technical system of "satellite-airborne-ground" integration was proposed.The major contents and results are as follows:(1)At the individual tree scale,the D.tabulaeformis disaster degree identification method based on the band selection of UAV hyperspectral image was proposed,and the identification model of D.tabulaeformis disaster degree was constructed.First,the mean spectrum of all pixels within the scope of the crown were extracted as spectral information of a single wood.The scope of a crown was got from the smoothed hyperspectral image based on the position that obtained from the field investigation.Then,according to comparisons of the results of principal component analysis(PCA),successive projection algorithm(SPA),and instability index between classes algorithm(ISIC),the ISIC-SPA algorithm was proposed for extraction of sensitive bands to leaf loss rate and the estimation model of leaf loss rate was constructed by partial least squares regression(PLSR).The sensitive bands selected by the ISIC-SPA include 466nm、522nm、618nm、702nm、714nm and 950nm,and the model is:P=231.5840-0.0513*Band466-0.0406*Band522-0.0717*Band618-0.0118*Band702-0.0082*Band714-0.0052*Band950.The model fitting precision achieved 78.9%,which shows that the ISIC-SPA-PLSR method is feasible and effective for D.tabulaeformis damage degree identification based on the UAV hyperspectral images.(2)At the plot scale,an optimized support vector machine(SVM)spectral-spatial classification method was proposed which can be used to extract the damaged pinus tabulaeformis on plot scale.This method uses the edge-preserving filtering to optimize the initial probability graph of SVM classification,and ensure the pixel class by the probability maximization criterion.By comparing the mean structure similarity index of joint bilateral filter and guided filter based on two different guided images,the CCD image got from field investigation and the untrue-color image got from the PCA decomposition of hyperspectral images,it was found that the overall accuracy and damaged pinus tabulaeformis identifying precision of the method by using the CCD image as the guided image and the guided filter to optimize the SVM classification are 95.13%and 90.21%respectively.It shows that the optimized SVM spectral-spatial classification method could identified the damage pinus tabulaeformis effectively at the plot scale(3)A D.tabulaeformis disaster area monitoring model at the regional scale based on the vegetation index was constructed.At first,the Landsat TM/ETM+and Landsat 8 OLI images from 1990 to 2016(besides 2005 and 2013)were used as the original data to extract the pinus tabulaeformis area of each year by random forest classification method,which the accuracy reached 89.44%,and the D.tabulaeformis damage area ratio of each year were calculated.Then six indice including AI,MIR,MSI,NDVI,NDII5 and RA were selected as the independent variables by correlation analysis to build the monitoring model.In order to eliminate the multi-collinearity between factors,the stepwise partial least squares regression(SPLSR)method was used based on secondary variables selection,and the final damage area ratio monitoring model is Y=-16.0413-1.5442AI+6.7884MIR+10.7365MSI-5.8108NDVI+10.3746NDII5+9.8881RA with the fitting precision of 80.80%,which can meet the application requirement.(4)A D.tabulaeformis damage area ratio prediction model at the regional scale was constracted with the future meteorological factors data prediction by using the time series analysis,with which the short-time D.tabulaeformis damage area prediction can be implemented under the future climate changes.Using the three years average meteorological factor as the independent variables and the D.tabulaeformis damage area ratio as the dependent variable.Supported by the correlation analysis,the temperature(T),extremely high temperature(HT),extremely low temperature(LT),mean relative humidity(H),percipetation(P),sunshine duration(S),and average wind velocity(W)were selected as the major factors to build the predicting model by SPLSR.The model is:Y=-13.0221+0.5481T+0.0381HT-0.0188H-0.0039P+0.0032S+0.2680W,which has the fitting precision of 81.17%.On this basis,D.tabulaeformis damage area from 2017 to 2021 were predicted by time series analysis of the corrected meteorological factors.The result shows that,under the condition of without human intervention,the damage area in the study area of D.tabulaeformis are mainly concentrated in the slight and medium hazardous in the future four years,and in 2020 the damage area shows uptrend,which provides a basis for disaster prevention and control.