Identification Of Driving Factors Of Forest Fires In Fujian Province

Forest fire is one of the important influence factors to the forest ecosystem.It can cause the loss of forest resources,and also threaten people's property and lives.Fujian province is one of the main forest regions in the southern China occupying the highest forest coverage inChina;however,Fujian is also a high fire frequency region which the human-caused forest fire account for nearly 95%of the total forest fire.At present,the domestic research regarding forest fire and fire prediction are mainly concentrated in the northern China where the Daxing'an mountains and Xiaoxing'an mountains are located in and some achievements were obtained in the past decade;however,compared to the northern forest region,the study related to the forest fire prediction and fire risk distribution in Fujian are relatively rare.Although,few relevant studies that were conducted in Fujian were related to forest risk patterns and fire occurrence prediction,the approach or the factors that were considered in these studies were relatively sample,and the conclusion also had some limitations.Previous studies have shown that forest fire,especially those caused by human activities,was affected by climatic,vegetation,and topographic factors,therefore,a comprehensive analysis based on all the potential factors mentioned above is essential for the prediction of forest fire occurrence of Fujian.In this study,the raster conversion,layer overlay operation and other functions under Arcgis 10.0 environment were performed to extract the vegetation and topography factors.In addition,we used the daily weather factors from 22 national weather stations that were located in Fujian as the climatic factors during model fitting.Logistic regression model in SPSS 19.0 was used to fit the dataset to determine the driving factors on fire occurrence of Fujian.We divided the dataset into training dataset(60%)and validation dataset(40%).Besides,in order to decrease the biases in model parameter selection that caused by randomly dataset division,we made three repetitions on dataset division and three sub-models were obtained.The variables that were significant in at least two of three sub-models were selected to fit the final(complete dataset)model.The result showed that " distance to railway","elevation","daily maximum temperature","daily minimum temperature" and "sunshine hours"were significant in the final model(P<0.0001).The ROC test revealed that the AUC value was 0.847 and P<0.0001 meaning a good model fitting.In addition,according to the parameter estimation of model,we established the prediction model for Fujian province.A map of forest fire occurrence likelihood was generated based on Kriging interpolation under Arcgis environment.The map showed that the high fire risk was concentrated in the northern and north-eastern Fujian(around the junction area among Longyan,Quanzhou and Zhangzhou city);in contrast,the west of Fujian had low fire risk.One possible reason caused the variation of fire occurrence likelihood spatially was because the difference of forest coverage in spatial scale of Fujian.The west of Fujian had a lower forest coverage compared to north and northeast of Fujian due to the deforestation and soil erosion in the past few decades.Our study can provide some reference for the local forest fire prediction management.