Forest Landscape Ecology Structure Analysis And Landscape Index Modeling Of Kang Ping County Based On GIS And ANN

Supported by GIS and combining data of forest resource secondary surveying and landscape ecological principles and methods, forest landscape constitutive construction , landscape heteroge- neities, scale and shape of landscape patches and landscape spatial pattern of Kangping county were analyzed, and on this basis, BP network model of fractal dimension and diversity index were established respectively by using Artificial Neural Network.The research shows the results as follows:(1) The result of landscape heterogeneities analysis showed that the study area was the typical broad-leaved forest landscape, woodland minors,suitable land for forestry and coniferous forestland whose landscape-scale just subordinate to broad-leaved forest, were important resources patches of the region.(2) Broad-leaved forest had the most patches in patches-scale of landscape elements, while nursery and auxiliary production forest land had the smallest. The limit margin of all types of patches was relatively high, showing that the size of patches differentiate more strongly.(3) The distribution of the grain classes construction of landscape element patches was dissymmetry,82.67% of total patches were identified as less than 10hm2 of area,16.29% of patches between 10hm2 and 50hm2,and only 1.04% of them were larger than 50hm2. Therefore, whole forest region patches size was belong to small grain structure .(4) Compared all kinds of forest landscape types, it shows that the patch shape of the scrublands are of more complication than ecological forest and non-standing land.(5) The result of forest landscape spatial distribution pattern showed that the auxiliary production forest land had the highest separation while coniferous forest land,woodland minors,suitable land for forestry and broad-leaved forest had the smaller, all below 0.5.Different types of patches made a great difference, showing that separation was closely related with human activities.(6) On the base of analysis of forest landscape ecology structure in Kangping county,the thesis set up two Back-Propagation neural models whose structure were both 3:S:1 , which took altitude,the number of residential area and distance to rivers and lakes as its input variables, and took fractal dimension,diversity index as its output variables.(7)The author selected 60 samples to train the models, the appropriate model structure were both 3:3:1 after repeated training and comparing and the networks were respectively named Fnet and Snet .The standard deviation of the training function mse were 0.00235547 and 0.0437107 and the fitting accuracy were 75.72% and 96.41% ,which showed that the fitting value closer to the actual value.(8) The constructed BP networks of fractal dimension and diversity index were respectively inspected by 17 test samples,the testing accuracy were 95.72% and 72.27%, the result showed the model had a good perfermence,and can simulate the ecological environment elements impact to the forest landscape pattern.