Assessing Forest Disturbance, Post-fire Forest Recovery And Its Coupling Mechanism For Climate Change

The probability of forest fires is gradually increasing under the scenario of global warming.At the same time,post-fire vegetation recovery still depends on many on-site factors,such as the fire intensity,different management strategies for vegetation restoration,topography and local climate.Therefore,it is of great scientific and practical significance to study the forest restoration in burned areas and its response to the post-fire climate.Based on the 11 path/row tiles of landsat images,this study used the VCT algorithm to study the forest disturbance and restoration in the Daxing'anling area from 1987 to 2016.Based on the disturbance maps,Combining with the SVM classification algorithm,this study tried to differentiate those fire disturbance events with extra high severity from all types of disturbances mapped by in 1987,2000,2006,and 2010.Aiming at these extracted burning sites,their post-fire vegetation recovery mode was analyzed by using the Thiel-Sen estimation method.By using the stepwise regression,support vector machine and random forest algorithm,the statistical relationships between the vegetation recovery in the post-fire fifth year and the average maximum temperature,the average minimum temperature,the extreme maximum temperature,the extreme minimum temperature,the average annual precipitation,the average relative humidity were simulated and compared.The major findings of the current work were summarized as follows:(1)In the Greater Khingan Range,the monitoring results of VCT can still maintain a high spatial consistency,and the spatial consistency of most years is between 70%and 86%.(2)By combining VCT and SVM algorithms,fire disturbances can be well separated from all the types of disturbances mapped by VCT algorithm.Overall,this VCT-SVM method is very effective in extracting fire disturbances and can provide accurate information on fire disturbances,which lays a solid foundation for further studies regarding vegetation recovery on those fire scars.(3)From the Theil-Sen's estimations,it was concluded that among the four fires,vegetation recovery from the 1987 fire site was the fastest,and that for the other three fires was relatively slow.(4)Random forest algorithm could better explain the relationship between post-fire forest recovery and various predictor factors.Considering the modeling performance,the stepwise regression model had a R~2 value of 0.6441,with an average relative error of 0.0711,a root mean square error at 0.0584,while its validation R~2 at 0.5828,an average relative error of 0.0861,and a root mean square error at 0.0734.The random forest algorithm gained a R~2 at 0.9258(modeling),an average relative error at 0.0350,a root mean square error of 0.03,while its validation R~2 at0.7701,an average relative error at 0.0665,a root mean square error at 0.0581.The support vector regression(SVR)algorithm obtained a modeling R~2 of 0.9011,an average relative error at 0.0361,a root mean square error at 0.0329,while its validation R~2 at 0.7041,an average relative error of0.0724,and a root mean square error at 0.0581.(5)There was a relatively good coupling relationship between climatic factors and post-fire forest restoration.From the results of variable screening,it can be seen that precipitation,average relative humidity and extreme maximum temperature in meteorological factors have a significant impact on forest restoration.