Forewarning Method And Application Of Forest Wildfire Fanger In The Great Xingan’s Mountain Region

The occurrence and development of forest fires are largely dependent on the weather.With global warming,the occurrence and burning area of forest fires will continue to increase,especially in the global boreal forest.The Great Xingan’s Mountain region of China is a major national forest region,known as the "green treasure house".However,this region is threatened by frequent and severe forest fires that endanger both people’s safety and homes as well as the environment.In the Great Xingan’s Mountain region,researchers usually established statistical models between forest fires and fire risk factors based on the analysis of fire risk factors,such as weather,fuel,terrain,vegetation cover,lightning strike,humanity,et al.However,there are three difficulties and challenges in current forest fire risk assessment researches.Firstly,how to choose the appropriate fire risk factors to express the dynamic changes in fire risk caused by fine-scale changes of fire risk factors.Secondly,how to choose the appropriate fire risk assessment model to more stably and accurately express the complex nonlinear relationship between fire risk and fire risk factors.Thirdly,how to assess the long-term applicability and stability of the fire risk model and fire risk factors.In order to support the fire prevention management decision and scientifically adjust fire prevention resources in the Great Xingan’s Mountain region,thesis studied the fire risk factors and fire risk models in view of the above difficulties,summarized a set of fire risk assessment ideas and methods suitable for the Great Xingan’s Mountain region,and applied to the fire prevention work.The main work and achievements can be summarized as follows:(1)Attempting to solve the issue of selecting the appropriate fire risk factors.Thesis takes all fire events recorded in the Great Xingan’s Mountain region from 2000 to 2020 as the research content.Firstly,the time series features of weather factors(daily maximum temperature,daily average relative humidity,daily precipitation,daily average wind speed)and fuel factors(FFMC,DMC,DC,ISI,FWI)within 16 days before the fire were analyzed,and the time windows and feature types of each factor were determined.Based on this,we divided four feature groups: feature group without time series values,feature group with time series values,feature group with Tsfresh transformation of time series values,feature group with gradient and cumulative transformation of time series values,and trained the random forest model respectively.The results showed that the evaluation values of all feature groups with time series features(f1-score>0.9)were significantly higher than those of the feature group without time series values(F1-score<0.8).The feature group with gradient and cumulative transformation of time series values had the best evaluation values across all models,and its fire risk maps highlighted the real extremely high fire risk in the western Great Xingan’s Mountain region at that time.Therefore,selecting appropriate time series features and incorporating them into the model helps to improve the accuracy of model and the effect of fire risk assessment in the Great Xingan’s Mountain region.(2)Attempting to solve the issue of selecting the appropriate fire risk models.Based on the Contents 1,thesis compared the evaluation values and fire risk maps of the models for imbalanced datasets(the balanced random forest model and the under-sampled Ada Boost model)and typical machine learning models(the random forest model and the XGBoost model).The results showed that the models for imbalanced datasets significantly improve the fire class identification accuracy and the overall accuracy at the cost of some non-fire class identification accuracy.And the fire risk map of the models for imbalanced datasets is more effective in identifying the position of the fire point during the middle fire risk season.As a result,employing the model for imbalanced datasets can effectively improve the accuracy of fire point recognition and the effect of fire risk maps in the middle fire risk season,which is helpful for forest fire risk assessment in the Great Xingan’s Mountain region.(3)Attempting to solve the issue of evaluating the applicability and stability of fire risk model and fire risk factors in a long time.Based on the Contents 1 and 2,thesis used CMIP6 climate data to compare the baseline fire risk values and true fire risk values from2000 to 2020 according to the monthly average,annual average and annual fire risk maps,and then to analyze the inter-year change of fire risk values and fire risk maps of the forecast period from 2020 to 2050 according to the annual average changes and fire risk maps.The results showed that the baseline fire risk and the true fire risk fit well on the temporal scale.The forecast fire risk showed that the extremely high fire risk year may have a big cycle of 40～50 years,and there will be a small cycle of 8～10 years in this big cycle.Therefore,we suggest that we should pay more attention to fire risk early warning,strengthen the monitoring and early warning of extreme fire weather in extreme fire risk years.It will reduce the probability and severity of forest fires and the losses of forest fires in the Great Xingan’s Mountain region by monitoring and early warning the extreme fire weather,deploying fire protection resources and clearing accumulated fuel in advance.