Simulation Of Carbon Density On Forest Ground In Northeast China Based On Deep Learning And Its Response To Climate Change

It is of great significance to accurately investigate the distribution of carbon density on the forest floor in Northeast China,to monitor forest ecosystems in Northeast China,and to quantify the carbon cycle in forest ecosystems.However,at the current research stage,the estimation methods of carbon density and biomass on forest land are not fixed,and different methods have different prediction capabilities and uncertainties.Therefore,this paper takes the major forest areas in Northeast China as an example,explores the deep neural network model combined with multi-source remote sensing data to estimate the potential of carbon density distribution on the forest floor in the study area,and from the model point of view,the climate factors that affect the forest carbon density accumulation are analyzed.Discussion.The main work and findings are as follows:(1)DNN model of a deep neural network is established to reduce the dimensions and integrate the feature variables of 37 multi-source remote sensing data including remote sensing,climate and terrain,and to coordinate information of previously collected and collated forest inventory data To match it,it is adapted to the input variable form of the DNN model through different mathematical transformations.The verification of the model results shows that the deep neural network structure is feasible for the simulation of carbon storage on forest land.At the same time,the DNN model has a higher R2 value of 0.84 and a lower RMSE of 2.44 MgC ha-1.Compared with the other two machine learning models,SVM and RF algorithms,and the ANN model with only a shallow neural network structure,the verification results of DNN are excellent among the four models.This result shows that the prediction results of forest carbon storage using deep structure models must be accurate than non-deep structural models,which is of great significance for the accurate quantification of forest carbon storage.(2)Based on the distribution of forest carbon density in Northeast China,using the DNN model established by this research to estimate the forest carbon storage in Northeast China is 2.43 PgC,and the average forest ACD is about 47.27 MgC ha-1 The comparison of the above-ground carbon density results of forests in Northeast China simulated by Trendy data of other terrestrial carbon storage distributions shows that the two estimates are not much different.From the perspective of the generalization of the spatial distribution,the distribution characteristics of the prediction results of the DNN model roughly show that as the latitude increases from south to north,the value of carbon density on the forest floor decreases accordingly,and the carbon density on the forest floor in the southern Changbai Mountain area It is higher than the northern Xing'anling area.At the same time,the cumulative amount of broad-leaved forests is higher than that of coniferous forests throughout the Northeast China.This distribution feature also validates the spatial distribution of carbon density on the forest floor of the previously used terrestrial ecosystem model(Trendy),providing a basis for further evaluation.However,through the calculation of the spatial distribution of uncertainty,it was found that the DNN model has relatively high uncertainty in mountainous regions with higher altitudes.Through analysis,it is found that the cause of this phenomenon may be related to the rugged terrain affecting the forest carbon.The uncertainty of the density distribution is related.(3)From the perspective of the model,the importance of the characteristic variables of the model training involved in the determination of the prediction results of the model was evaluated.It was found that among the climate variables that affect the carbon density accumulation on forest,land,the importance of the precipitation variable in the wettest months the highest is 15.8%,and among the temperature-controlled climate variables,the annual average temperature is the most important variable at 10.81%.This result validates the conclusions calculated by previous scholars through mathematical models that average temperature and seasonal precipitation contribute the most to the distribution of forest ACD,and also proves the reliability of the DNN model's ability to identify feature variables.Based on the distribution of carbon density on the forest floor,which is controlled by both the precipitation and the annual average temperature in the wettest month,it is found that the two variables have a significant nonlinear interaction on the distribution of carbon density on the forest floor.And in the future climate change trend,when the temperature rises and the seasonal precipitation increases,the peak value of the carbon density accumulation in the forest floor of the Northeast region will show a downward trend.