基于MSVM-CA模型的区域土地利用演变模拟

Simulation of the Regional Land Use Evolution Based on MSVM-CA Model

研究目的:探索一种基于多类支持向量机自动获取地理元胞自动机非线性转换规则的方法(MSVM-CA),并将其用来模拟复杂的土地利用动态演化。研究方法:以长江口北岸启东市2002—2007年的土地利用演变模拟为例,采用1-A-1方法在每两类样本数据间训练一个分类器,将模拟土地利用变化的主要特征变量映射到Hilbert空间,通过多类支持向量机建立最优分割超平面,运用决策函数计算元胞到最优分割超平面的距离,并转换为元胞土地利用类型的转换概率。然后利用蒙特卡罗方法,确定元胞的转化状态,循环完成土地利用演变模拟,最后对模拟结果进行精度分析。研究结果:对照2007年遥感影像分类得到的实际土地利用分类图,实验模拟结果数量上正确率平均达到88.99%,模拟结果的总体分类精度为86.75%,Kappa系数为0.85。研究结论:基于多类支持向量机的非线性元胞自动机方法具有较高的模拟精度,可以应用于模拟多类土地利用类型之间的演变。

The purpose of this paper is to develop a MSCM-CA model to simulate the dynamics of multiple land use change, which is a new method for automatically getting nonlinear transition rule of geographic Cellular Automaton by integrating multi-class support vector machine. Method employed is to simulate changes by taking Qidong city between 2002 and 2007 in the north branch of the Yangtze River estuary as an example. The steps of simulation included that 1 ） to transform the data from nonlinear boundaries in the original space to linear boundaries in the Hilbert space and establish the optimal separating hyperplane by adopting one-against-one method to train a classifier in every two kinds of the sample dates; 2） to calculate the distance between the cells to the optimal partition hyperplane by the decision-making function and convert to land use conversion probability of the cells; 3 ） to determine the transformation of the cell state and to complete cycle of land use evolution simulation by using the Monte Carlo method; 4） and finally to analysis the precision of simulation results. By comparing the simulation map with the actual land-use map from remote sensing image, results indicate that the average simulation accuracy rate is 88.99%, spatially the simulation overall classification accuracy is 86.75%, and the Kappa coefficient is 0.85. It is concluded that the proposed model in the paper has high accuracy of simulating complex land use changes and can be used to simulate the multiple land use evolution.