|
This article first comprehensive overview of the technology at home and abroad with a variety of modeling methods to investigate the status of the study of the relationship between vegetation and environmental factors as well as the latest progress after detailed description of the artificial neural network (Artificial Neural Networks, ANN) Arcgis engine (Arcgis Engine AE) technology, cellular automata basic principles (Cellular Automata, CA), generalized additive modeling techniques (Generalized Additive Models, GAM) and generalized linear models (Generalized Linear Models, GLM), and describes the overview of all aspects of geology, topography, climate, hydrology, soils, vegetation, followed by four models (ie, artificial neural network technology combined with cellular automata (ANN CA), artificial neural network (ANN), generalized additive model (GAM), the generalized linear model (GLM)) of 12 kinds of Chinese forest vegetation type modeling, and Kappa coefficient and AUC (receiver operating characteristic curve (ROC curve) the area under) to simulation accuracy of the evaluation of four models, and then elected to the accuracy of the best model to predict the distribution of 12 forest vegetation types under future climate change conditions. Forest vegetation type two, respectively, for 1980 and 2002, 12 kinds of forest vegetation maps. Modeling environment variables (ie, model independent variable) is divided into 43, including four terrain variables (ie digital elevation model DEM, slope slop, aspect of things to aspectew north-south aspect aspectnw) 5 from distance variable (ie, each grid to the recent road distance dis2road of to nearest river distance dis2river, the to railway distance dis2railway, the distance to the nearest lake dis21ake, to the nearest town the dis2town), soil variables, eight of climate change variables (that is, from 1951 to 2004, the annual average temperature tavg average summer temperature trends t_summer, the average winter temperature trends t_winter average winter minimum temperature trends t_min_winter annual average temperature trends t_year, average annual precipitation pavg, the average winter precipitation trends p_winter, average summer precipitation changes the trend p_summer), 13 forest vegetation type variable (first of forest maps, and separated from the value of 0 and 12 the distribution of the forest) and 12 types of forest vegetation neighborhood variables, 43 variables to reflect the state of the environment of each grid; model the dependent variable for a second period of 12 kinds of forest vegetation type of separation of variables. Independent variables and the dependent variable based on a resolution of 1000 × 1000m record in 4000 × 4887 grid. The modeling ideas 4 topographic variables, 5 from the one of the variables, all soil variables as the background value, it is assumed that these variables during forest vegetation analog is unchanged, they are mainly in the modeling simulation the constrained background value of the forest vegetation type, four models to establish the relationship between the eight climate change variables and two changes in forest vegetation types. This article the main conclusions are: 1 four models AUC values ??ranged between 0.896 and 0.968, ANN CA model AUC highest, reaching 0.968 ANN model AUC followed, AUC = 0.942, GAM the AUC of the model is also more than 0.9, the value of 0.940, the AUC value of the GLM model minimum is less than 0.9, a value of 0.896. The four models of the average AUC value of 0.942. From the AUC, the AUC values ??of the first three methods to be significantly higher than the accuracy of the GLM, the ANN the AUC value than ANN CA low 0.26, cellular automata (CA) to improve the accuracy of the model. Kappa values ??ranged between 0.482 and 0.631, an average of 0.577. ANN CA model the highest Kappa values ??reached 0.631 ANN model Kappa value, followed by a value of 0.615, the GAM model kappa value less than 0.6, a value of 0.581, the GLM model lowest Kappa values ??of less than 0.5, value of 0.543. From the accuracy of the four models in the Kappa values ??of view, the conclusion is consistent with the AUC, are the ANN the CA gt; the ANN gt; GAM gt; the GLM. As can be seen, both from the AUC values ??or from the Kappa values ??of view, the four models ANN CA (ie, the highest accuracy of the artificial neural network combined with cellular automata model), so the vegetation change to predict future climate variables ANN CA model should be chosen. 2.ANN CA model in this paper, the BP neural network technology, GIS (geographic information system) and CA, CA neighbor types and how to get the conversion rules objectively correct definition and as well as how to obtain model ANN CA model The input and output variables are explored. The results showed that 12 areas of the forest vegetation types information as input variables of BPANN can more objectively the CA transformation rules, so that the the CA applications more simple; while based the Visual Studio2005 platform under the C # language and ArcGISEngine9 .2 model input and output variables extracted functional modules greatly simplify the raster data processing as well as the establishment of the database.
|