Methodology And Application For Wildfire Risk Assessement Based On Remote Sensing Techniques

With the global warming,the increasingly severe forest wildfire has caused serious damage to the earth's ecological environment.Therefore,the assessment and early warning of wildfire risk is very urgent.The occurrence of wildfires is affected by a combination of factors.At present,the existing methods for assessing wildfire risks in China only consider meteorological factors.In fact,according to the wildfire triangle model,the occurrence of wildfire is closely related to ground combustible information and topographical elements,in addition to meteorological factors.Therefore,the research on wildfire risk assessment methods based on multi-source data sets such as meteorology,combustibles and topography will help to form an effective wildfire risk assessment method system and provide scientific fire prevention decision-making basis for forestry and other relevant departments.The thesis takes Yunnan Province as the research area,and analyzes the temporal and spatial characteristics of wildfires in the study area based on MODIS remote sensing data,ERA-Interim meteorological reanalysis data and GMTED2010 topography data,and constructs the assessment model of wildfire risk through data mining technology.On this basis,based on meteorological forecast data(GFS)and near real-time remote sensing data,the wildfire risk in the next six days will be predicted.The main work and results can be summarized as follows.(1)According to the MCD64A1 product provided by MODIS,the wildfire samples of the study area from 2007 to 2016 are extracted.At the same time,under the constraints of the time and space,the equal number of non-wildfire samples are randomly extracted based on the semivariogram.In order to eliminate the significant structure differences between different vegetation types,according to the IGBP classification rules,the extracted samples are divided into two parts: forest and grassland,and then the training data set is constructed for the assessment model of wildfire risk.(2)The thesis analyzes the temporal and spatial characteristics of wildfire in the study area in detail.The temporal characteristics includes the variation of the number of wildfires on the annual and monthly scales,and the mechanism is analyzed.The spatial characteristics analyzes the spatial distribution density of wildfires based on the land cover type and the administrative division map of the study area.In addition,the thesis takes the wildfire sample of grassland as the example to analyze the relationship between the number of wildfires and the factors.(3)The logistic regression model is used to calculate the wildfire risk index(WRI)that is used to characterize the wildfire risk.The model training process is divided into modeling,internal verification and external verification.In the modeling,70% of the samples from 2007-2014 are used to training model and the remaining 30% of the samples are used for internal validated.External validation refers to that the model which has been build is validated again by the samples from 2015-2016.The verification results are characterized by the AUC value of the ROC curve.The AUC values of grassland and forest for the internal verification are 0.91 and 0.86 respectively,and the externally verified are 0.89 and 0.83 respectively.The internal verification results show that the model has excellent assessment performance,and the external verification results show that the assessment performance is still excellent when the model is used in the future.(4)Based on the scalability of the assessment model and the feasibility analysis of wildfire risk prediction,combined with meteorological forecast data(GFS)and near real-time remote sensing data,the wildfire risk in the next six days will be predicted.At the same time,in order to avoid the lack of wildfire risk predicted result due to the lack of remote sensing data,the thesis uses the gap filling algorithm to fill the prediction results.