Forest Fire Prediction Model And Risk Assessment In The Water Source Area Of The Central Line Of South-to-North Water Diversion

Significant climatic changes and weather extremes have resulted in frequent forest fire activity and even increased fire severity.The sudden and destructive nature of forest fires can also cause irreversible damage to the surrounding atmosphere,water bodies and soil.The ecological condition of the Danjiangkou Reservoir,as the culmination site of the water resources of the South-North Water Diversion Project,has attracted much attention.In recent years,most of the domestic forest fire risk research areas are concentrated in high forest fire-prone areas such as the northeast and southwest of China.Forest fire risk prediction studies in the Danjiangkou Reservoir have not yet been analyzed in detail.In order to clarify the objectives of water quality assurance in the reservoir area and to create awareness of environmental risks in the ecological reserve,this paper establishes and evaluates a predictive model of forest fire occurrence based on meteorological elements and fire weather indices as drivers by finding the relationship between forest fire activity and climatic elements.Based on the prediction results of the optimal model,a comprehensive assessment of the future forest fire risk in the Danjiangkou reservoir area is made in the context of climate change and "double carbon",with a view to providing a reference basis for environmental risk prevention to safeguard the water quality of the South-to-North Water Diversion Eco-region.Based on this,the paper focuses on the following:(1)Fire and meteorological data processing and analysis.By analyzing the characteristics of the variation of the monthly overfire area time series over a 37-year period(1982-2018)in the study area at spatial and temporal scales,three periods of high historical fire activity were identified,1985-1988,1997-2000 and 2016-2018.Fire season is summer and autumn(June-October),with slightly less activity in winter and occasional activity in spring;The high incidence areas were identified as study unit zones 4,8 and 10 by combining the total overfire area burned.Eighteen fire weather indices were calculated as index variables for forest fire prediction based on historical meteorological observations in the study area;The BCSD method was used to downscale the simulated data at different scales of the GCMs.The results of the CMIP6 climate model comparison and evaluation show that the Can ESM5-Can OE model has the best simulation capability and can effectively reflect the meteorological distribution of the Danjiangkou water source area during the historical period.(2)Forest fire occurrence prediction modeling and assessment.First,the set of predictor variables for the forest fire prediction model was selected from 22 variables using the Boruta method;then,the optimal parameters for the forest fire occurrence prediction model in the Danjiangkou reservoir area were determined based on three machine learning algorithms(Kernel k-Nearest Neighbor algorithm,C5.0 algorithm and logistic regression algorithm)by means of grid search;finally,five evaluation criteria(accuracy,precision,recall,F measure and AUC)to evaluate the model performance.The results show that four categories of meteorological variables(precipitation,air temperature,relative humidity and wind speed)and four index variables(Drought Code,Angstrom Index,Hot Dry Wind Index and Zdenko Index)were identified by feature selection as drivers of the forest fire dichotomous model.The mean performance results of the forest fire prediction model for each study unit area under five performance measures show that the KKNN algorithm exhibits a strong performance advantage over the other two algorithms.The highest accuracy in the training and test sets were 0.789 and 0.784 respectively,while the highest values of precision,recall and F-measure were 0.793,0.918 and 0.807 respectively,and the highest AUC value was0.839.The highest AUC value was 0.839.The models in order of "goodness" were KKNN,C5.0 and LR.(3)Assessment of future forest fire risk levels in the study area.The future trends of four types of meteorological elements in the Danjiangkou reservoir area are analyzed based on the results of two climate scenarios from the Can ESM5-Can OE model.Using the Jenks method,the probability of future fire occurrence in the study area is calculated based on the meteorological data from the two scenarios and classified into fire risk classes.The fire risk rating of each study area was analyzed for the last 13 years(2023-2035)under the goal of achieving "peak carbon" by 2030.The results of the analysis show a significant upward trend in future air temperature and a downward trend in precipitation,relative humidity and wind speed,with a more pronounced degree of trend in air temperature and precipitation changes.And there are stage troughs in air temperature,relative humidity and wind speed in 2030.The distribution of fire risk severity in the Danjiangkou reservoir area shows the most severe in the southeast(four to five),followed by the northeast(two to four),with the lightest fire risk forecast in the northwest(one to two).And with increasing radiative forcing,high fire danger levels are concentrated in the south-east(zones 7,9 and 10).The fire risk levels in Study Units 5,7,9 and 10 change over the three time periods when the “peak carbon dioxide emissions” target is achieved.In Zone 5,the fire risk levels show a “low-high-high” scenario in both scenarios,indicating that the fire risk levels in this zone are still increasing and not decreasing after 2030.Fire risk levels in Zones 7 and 9 show a “high-high-low” and“high-low-low” scenario under SSP585.Fire risk levels in Zone 10 show a “high-high-low”scenario under SSP245,suggesting that a “peak” fire risk situation is achieved in Zones 7,9and 10 by 2030.In response to climate change adaptation strategies and the ’double carbon’ goal,it is necessary to investigate the meteorological influences between wildfire and the carbon cycle.The study found that there are differences in the level of future forest fire risk in different study units of the Danjiangkou Reservoir,and that the control units need to be refined to achieve a “one area,one policy”.Establishing and improving the forest fire early warning and emergency response mechanism in the key risk unit areas of the reservoir to ensure water quality as a rigid constraint.Implement the main responsibility of local governments and the supervisory responsibility of watershed and forestry agencies,and adopt targeted forest fire contingency plans.