| Forest grassland fires are regarded as a worldwide forestry disaster,affecting the vegetation coverage and the diversity of the ecosystem,and causing great damage to forest grassland resources and forestry and animal husbandry.However,wildfires are caused by a combination of climatic factors(such as temperature,rainfall,relative humidity,wind,etc.),combustible materials,topography,human activities,etc.Due to the regional differences of the influence variables,the occurrence of wildfire shows a certain volatility in space and time,affecting the temporal and spatial characteristics of wildfire season,burning frequency and burned area.With the rapid development of satellite remote sensing technology,space-borne sensors can obtain multi-band information on the surface and monitor forest and grassland fires due to their unique advantages of high spatial and temporal resolution and large-scale monitoring,making remote-sensing satellite-derived data an effective data source for wildfire monitoring.In recent years,forest grassland wildfires have occurred continuously in many areas of Sichuan province in winter,causing serious casualties and loss of forest resources,and seriously destroying the balance of ecological environment.At the same time,the complex environmental problems in Sichuan,such as the drastic relief of the terrain,the changeable local climate,and the diverse forest types and structures,make the forest wildfire control and rescue work extremely difficult.Therefore,quantitative assessment of the distribution of temporal and spatial characteristics of wildfires in Sichuan Province can optimize the spatial allocation of fire fighting resources and provide scientific basis and data support for forest fire management in Sichuan Province.However,it is difficult to obtain spatio-temporal continuous land surface remote sensing observation in the cloudy meteorological environment in Sichuan Province,and cloud and fog interfere with the analysis of reflectance after fire.The shadow produced by mountainous geographical environment and heterogeneous landscape will reduce the intensity of spectral signal and affect the quality of remote sensing data.The collaborative use of multi-source spatiotemporal data is expected to overcome this problem.Based on this,the thesis uses remote sensing technology to collect a variety of fire point information,a variety of wildfire induced environmental variables and other multidimensional characteristic data,evaluates the detection accuracy of a variety of remote sensing fire data to Sichuan wildfire,and studies the temporal and spatial distribution characteristics of forest and grassland fires in this region,providing theoretical and decision-making basis for fire prevention activities and reasonable allocation of fire resources for fire departments.The specific research contents are as follows:(1)Accuracy evaluation of four satellite-derived fire products based on cloudy and mountainous regions in China.Based on four products: MCD64A1,Fire＿CCI51,MCD14 ML,VNP14IMGTDL＿NRT to evaluate fire accuracy.The ST-DBSCAN temporal and spatial clustering analysis and random sampling and estimation based on fire density were used to compare the fire detection capabilities of four remote sensing fire products in the cloudy and mountainous region of Sichuan Province,and the detection accuracy of four fire products for small fires was verified based on fire events.The results show that the four fire products have poor fire detection ability in this region(F1-Score<0.5),VNP14IMGTDL＿NRT has the best performance,followed by MCD14 ML,Fire＿CCI51 and MCD64A1.MCD14 ML has the best small fire detection ability.(2)Temporal and spatial characteristics of forest grassland wildfire and correlation analysis of environmental variables in Sichuan Province.Based on multi-source remote sensing fire point data(MCD64A1,Fire＿CCI51,MCD14ML),meteorological,terrain,and fuel thematic data,this study aims to screen and extract effective fire area data or vector fire point information from 2001 to 2021 according to three kinds of fire data and their quality information.Combined with MCD12Q1 land cover product,forest and grassland fire database was constructed to study the spatial and temporal distribution and temporal trend of forest and grassland wildfire in Sichuan Province from 2001 to 2021.The results showed that the forest and grassland fires fluctuated and increased during2001-2014,and reached a peak during 2012-2014,while the fire frequency and burning area plummeted and remained at a low level since 2015.Forest and grassland fires have strong spatial heterogeneity.Forest fires are mainly concentrated in the southern cities of Panzhihua and Liangshan Prefecture.The distribution characteristics of grassland fires are less in the middle and more in the periphery,but there is an obvious increase of fires in the northeast in recent years.On the other hand,based on the mathematical statistics method,combined with the long-term meteorological(temperature,rainfall,relative humidity,wind speed),combustible(fuel moisture content FMC and leaf area index LAI)and static terrain variables(slope,slope direction,elevation)and other influencing factors,An adaptive fuzzy neural network model was constructed to generate a comprehensive fire risk index(CFDI)to evaluate fire risk,and its correlation with the temporal and spatial characteristics of forest and grassland wildfires as well as the main driving factors were analyzed.The results show that there is a strong negative correlation between FMC and forest fires,and a strong correlation between meteorological factors and grassland fires.Rainfall and relative humidity are negatively correlated,while temperature and wind are positively correlated.Environmental variables affect the occurrence of forest and grassland fires.The interference of human factors is not excluded.The trend of forest fires in recent years has a strong correlation with the trend of CFDI.Environmental variables may be the main driving factors of forest fires,and the interference of human factors and relevant fire prevention policies of the government affect the time characteristics of grassland fires.(3)Long time series analysis and prediction of wildfire potential risk in Sichuan Province.Three wildfire potential variables(VPD,FWI,WDI)and three fuel load variables(FFL,AGB,SIF)representing wildfire risk were analyzed based on the MannKendall trend test and the long and short term memory network LSTM in the past 21years(2001-2021)and the next 10 years(2022-2032)by remote sensing.The results showed that there was no significant increase or decrease in wildfire potential variable trend during 2001-2021,but there was a significant fuel load accumulation trend(p< 0.05).The effectiveness of wildfire management model in Sichuan Province was verified when the terrain factor was stable.At the same time,in the case of increasing fire risk and fuel load in the future,reasonable allocation of fire fighting resources should be considered to avoid the "fire trap". |
PDF Full Text Request