Modelling And Forecasting The Effects Of Climate Change On The Distribution Of Chinese Forests Based On ANN And CA Methods

In this paper we firstly expatiate a comprehensive overview of the model which was used to explore the releationship of the vegetation and environment factors study of the relationship and the model's current situation and the latest progress, followed by a detailed introduction of artificial neural network,Arcgis Engine technology, the basic theory of cellular automata generalized additive modeling techniques and generalized linear modeling techniques, and the introduction of the general situation of geology, topography, climate, hydrology, soil, vegetation and other aspects of study area. And then four model approach (ie, artificial neural network technology and cellular automata (ANN+ CA), artificial neural network (ANN), generalized phase additive models (GAM), generalized linear model (GLM)) were used to builded the model of 12 forest vegetation types and environment factors, with Kappa coefficient and AUC (the receiver operating characteristic curve (ROC curve) under the area) to evaluate four kinds of model simulation accuracy, and then select the best model to predict the future distribution of 12 types of forest vegetation under future climate change.The vegetation were used in the text have two period of forest vegetation period, were respectively the year of 1980 and 2002,12 species of forest vegetation in China's distribution. In the establishment of model, the number of environmental variables (ie model independent variables) were 43, including four terrain variables (ie, digital elevation model, slope, east-west slope, north-south aspect), five distance variable (that is, each grid the distance to the nearest road, the distance to the nearest river, the distance to the nearest railway, the distance to the nearest lake, the distance to the nearest town),1 of soil variables, eight of climate variables (ie,1951-2004 annual average temperature, average summer temperature change rate, the average winter temperature change rate, the average winter minimum temperature change rate, annual mean temperature change rate, annual average precipitation, the average winter precipitation change rate, average summer precipitation change rate),13 forest vegetation variables and 12 forest vegetation neighborhood information variables.In this study we usede the 43 variables to reflect the environmental situation of each grid; the dependent variable of the model is a 12 second period forest vegetation types. Independent variable and dependent variable are recorded by the resolution of 1000×1000m i in the grid of 4000×4887. The the main idea of the modeling is to establish the relationships of 8 major climate change variables and two types of forest vegetation changes. The main conclusions of this paper were as follows:1. The AUC values of the four models of between 0.896 and 0.968, the ANN + CA model has the highest AUC value of 0.968, and the ANN model followed by value of 0.942, the AUC value of the GAM model was 0.940, the GLM model for the lowest AUC value of less than 0.9, whick was 0.896. The average AUC value of four modes was 0.942 model. By the AUC, the three froth methods of AUC values significantly higher than the GLM, ANN's AUC value was lower than the ANN + CA by the value of 0.26,whick dicating cellular automata (CA) can greatly improve the accuracy of the model.The Kappa value of the four models was between 0.482 and 0.631 with a mean of 0.577. ANN + CA model's Kappa value was the highest which was 0.631, ANN model's Kappa value was second with the value of 0.615,theAUC value of the GAM model was below 0.6, whick was 0.581, while the GLM model has the lowest Kappa value, whick was less than 0.5, with the value of 0.543.By the Kappa values of four models, the conclusions was consistent with the conclusions of AUC.No matter in the terms of AUC values or from Kappa value, the ANN + CA model (ie artificial neural network and cellular automata model) the highest accuracy, and therefore the ANN + CA model shoule be used to predict future vegetation changes under the future climate changes.2. the establish of ANN + CA ModelIn the establishment of ANN + CA model, we used the BP neural network technology, GIS (geographic information system) and the CA, we explore the how to choose the the CA's neighbor type, and how to obtain model input and output variables and how to derive the rule of change in CA. The results show that, by the BPANN and the 12 type of forest vegetation types of domain information, we can derive the rules of CA objectively,and make CA more easily. The ArcGISEngine9.2 combined the Visual Studio2005 platform simplify the process of the raster data.