| Alien Species Invasion has now spread almost all the agricultural and natural ecosystems, which brings huge economic losses and ecological impact. Prediction of the potential distribution of alien invasive species is not only the essential part of risk analysis, but also provides theoretical basis for early prediction and preventive measures decision of biological invasion. Mikania micrantha is one of the weeds of most serious harm in tropical or sub-tropical zone, and it's universally recognized as one of the most dangerous plants for forestry. M.micrantha grows rapidly in its invaded area. At present, it has been rapidly proliferated and hard to control in Guangzhou, so that the prediction study of M.micrantha has been the matter of extreme imperative. But presently the studies on M.micrantha are largely concentrated in the aspects of harmful effects and chemical control, lacking of the prediction research on space distribution.The thesis selected actual survey data from Guangzhou area and corresponding quantitative relations of 14 environmental factors, in order to achieve some more appropriate models to predict the space distribution of M.micrantha. The main methods include:(1)Analysis of major constituent to assure the principal factors; and get the potential distribution of M.micrantha by interpolation and superposition method of GIS for establishing the main factor of threshold value division; (2)Utilizing BP neural networks'powerful ability for nonlinear, selecting effect environmental factors to establish a simple but feasible networks model, which will forecast M.micrantha''potential distribution in Guangzhou;(3)Utilizing ROC to implement accuracy test on GIS model and BP networks separately, and compare the accuracy between them. Main results as follow:(1)GIS model predicts:average annual temperature, average annual sunshine duration, mean annual precipitation, precipitation in the most wet and dry quarter, and altitude factors greatest affects the distribution of M.micrantha. Through the threshold value division of major factors, we found the temperature and sunshine condition are fairly good for the growing of M.micrantha. Its highly suitable regions are Huangpu and most areas of Luogang district, which accounting for 3.47% of Guangzhou's total area; Nansha, Tianhe district, Zengcheng, Conghua and parts of Panyu district are suitable regions of M.micrantha, which accounting for 19.64% of Guangzhou; Huadu, Baiyun, Conghua, the most area of Panyu and parts of Nansha, as M.micrantha's lowly suitable regions, account for 72.89% of Guangzhou. In non-normal regions——Liwan, Haizhu, parts of Panyu and Conghua, the altitude and precipitation become the main restrictive factors. But this regions is about 300 km2, accounting for 4% in Guangzhou, and all these illustrates the natural conditions of most area of Guangzhou are benefit for M.micrantha's growing and rapidly proliferation.(2)Results of BP Networks show that the distribution area of M.micrantha in Liwan, Yuexiu and Haizhu is minimal, which is almost close to 0, and therefore these districts are classified as non-normal regions. Conghua and Huadu are lowly suitable regions, where the distribution areas are 60 hm2 and 69.3 hm2. In Baiyun, Panyu, Zengchen and Nansha, the distribution area of M.micrantha account for a higher proportion of the regional area, where are suitable regions. Luogang, Tianhe and Huangpu are determined as highly suitable regions, where the situation of M.micrantha's distribution is most serious.(3)The predicting results of two models are almost same. ROC shows the effect of GIS model on M.micrantha's potential distribution is poor, the value of AUC under the curve is 0.771; but the AUC value in BP networks model is 0.834, which is even better fit for actual distribution. Obviously believed, BP networks model is precision higher than the former one.
|
PDF Full Text Request