Research On Forest Fire Monitoring Method Based On MODIS Data

Forest fire will bring huge safety threat and economic loss to the country and people.China is one of the countries with high incidence of forest fire.Research and analysis show that satellite remote sensing data can be used to effectively control forest fires from three aspects: pre-disaster prediction can reduce the occurrence of fire;real-time monitoring during the disaster can reduce the loss to the maximum extent;efficient and rapid extraction of fire marks can provide a reliable basis for forest post-disaster assessment and reconstruction.At present,MODIS data is one of the most widely used remote sensing data,which has the advantages of high real-time performance,high efficiency,high accuracy and low cost for forest fire monitoring.This paper focuses on using MODIS data to study the three stages of forest fire: before fire stage,during fire stage,and after fire disaster stage,and to improve the problems existing in the process of forest fire monitoring based on MODIS data.The followings are the main analysis and conclusions of this paper:(1)Building a prediction model based on multi-forest fire risk factors.This paper combines the static and dynamic fire risk factors by their weights to carry out forest fire risk prediction according to analytic hierarchy process,and solves problems caused by choosing single risk factor or multiple risk factors that are not correlated to each other.Imputing MODIS data of April 25 and 26,2009 in Heihe city,the model reveals that the prediction accuracy is 83.33% and 87.50% respectively,while the accuracy of traditional prediction model is only 56.25% and 68.75% respectively.The results tell that the model can predict the fire risk of Heihe city effectively.In addition,using fire point date to classify static fire risk factors can improve the applicability of the model to different regions to some extent.(2)Building a smoke recognition model based on BP neural network.Through analysis of ground features,this research chooses Ref7,Ref8,BT32-BT20,MNDFI,NBR and NDVI as the input layer of the model and takes single seasonal and regional ground feature sample to build smoke identification model.The research result verifies the reliability of model: its overall accuracy and Kappa coefficient are 96.96% and 95.78% respectively.In the last step,tested by four regional data and compared with the multi-channel threshold method,this model demonstrates that it can effectively monitor the smoke over the land in different seasons and regions,and improve the efficiency of identifying forest smoke.In addition,MNDFI-NDVINBR false color composite image is better than true color composite image for smoke display over land.(3)Using the method of maximum variance among classes to extract the fire field.Firstly,this paper analyzes the sensitivity and separability of NDVI,NBR,NSTv2 and NSEv1 to the fire field.Secondly,this research carries out extraction of fire field by using the method of maximum interclass variance,and then evaluates the accuracy of the extraction results based on fire_cci5.1 data.The result shows that overall accuracy of NSEv2 index and Kappa coefficient are 93.47% and 83.08% respectively,which are much more accurate than other indexes.Thirdly,the applicability of the method to multiple regions has been evaluated and the Kappa coefficients are 76.42% and 75.06%,respectively,which are highly consistent.Therefore,by using the maximum inter-class variance method with NSEv2 together,researchers can improve the accuracy and efficiency to obtain fire filed,which is a better way to solve delay problem caused by using other product data.