Method And Application Of Short-and Medium-term Fire Danger Early Warning In Forest And Grassland

In recent years,global forest and grassland fires have been rampant,posing significant threats to the safety of people’s lives,property,and ecological protection.Therefore,achieving precise forest and grassland fire danger warning and suppressing wildfires at their inception is crucial.With the rapid development of machine learning,Geographic Information System(GIS),and big data mining technologies,research on forest and grassland fire danger early warning based on multi-feature and complex models has become an important research focus in the field of wildfire prevention and control.Although the current short-term forest and grassland fire danger warning technologies for the upcoming week are relatively mature,research on precise short and medium-term warning for complex regions is relatively limited.These studies usually suffer from inadequate consideration of fire danger warning factors and a lack of coupled historical fire danger trends for short-term fire danger warning.Therefore,this study focuses on short and medium-term fire danger warning methods for forest and grassland fires,based on the mainstream fire danger factors combined with machine learning methods,to improve the accuracy of short and medium-term fire danger warnings.The specific research contents are as follows:(1)To address the current problem of insufficient coupling of short-term fire danger warning to historical fire danger trends,a dataset of forest and grassland fire cases in the Panxi region from 2003 to 2018 is constructed based on the current mainstream fire danger factors in the study area,and combined with the International GeosphereBiosphere Programme(IGBP)classification method to divide the study area into grassland and forest areas.Three short-term fire warning methods were compared: shortterm fire warning method based on GFS weather forecast data,short-term fire warning method based on historical fire data,and short-term fire warning method based on historical fire and GFS data in terms of warning accuracy.The results show that the receiver operating characteristic(ROC)curves,precision-recall(PR)curves,and F1-score(F1-score)of the three models show a good level of evaluation indexes,i.e.,all three methods can provide accurate warning for forest grassland However,the short-term fire danger warning method based on historical fire danger and GFS data has the highest accuracy,and the fire danger temporal distribution and fire danger change time series line graphs of actual fire cases can be analyzed to show that the short-term fire danger warning method based on historical fire danger and GFS data has the best effect.(2)To address the problems of unclear warning accuracy and lack of fire danger factor mining and analysis of different medium-term fire danger warning methods,this study takes the Chinese land area as the study area and firstly constructs the Fire Weather Index(FWI)by replacing weather stations with ERA-5 Land reanalysis data set,which reduces the accuracy error of weather station data interpolation.The FWI is produced on a large scale and for a long period,and the FWI is used to construct a severity index for the study area as a medium-term fire danger index for the Chinese region.Second,we constructed a spatial and temporal dataset of forest and grassland fire cases in the Chinese region based on monthly synthetic meteorological data,and compared the effects of the Random Forest(RF)model and the e Xtreme Gradient Boosting(XGBoost)model for forest and grassland fire danger modeling in the Chinese region,and selected the RF model to construct a medium-term fire danger warning model,and finely evaluated and analyzed its fire danger factors.Through the spatial and temporal distribution of monthly fire danger and the effect of Fire Density(FD)in the China region in 2010,it was found that the medium-term fire danger warning method based on monthly synthetic meteorological data could achieve more accurate medium-term fire danger warning.(3)Taking Cengong County,Guizhou Province as the research area,this study creatively adds two factors: vegetation type and Euclidean distance to cemeteries,based on the characteristics of fire cases in Guizhou Province and a detailed analysis of refined fire danger factors.By comparing the modeling accuracy and spatial and temporal distribution of fire danger for short and medium-term fire danger warning with and without adding the refinement factors,the accuracy of short and medium-term fire danger warning with the addition of refinement factors is further improved and the scope of fire danger warning is more accurate.It shows that the short-term fire danger warning model and the medium-term fire danger warning model with the addition of refinement factors can achieve more accurate fire danger warning in the refinement area,and provide more localized scientific guidance for fire prevention decisions.