Land Suitability Assessment Model Based On Weights Of Evidence Method And Its Application

After reviews and analyzes the methods used in land suitability assessment, the problems existing in those methods are revealed as follows:(1) Traditional process of land suitability assessment is always of some subjectivity, and involved anthropogenic interfere in different part of the process, such as in the step to determine the weight of evaluation indexes system and in the step to determine the benchmark of assessing the results. Those factors make the quantifying of assessment difficult. (2) Even though artificial intelligent algorithm is good at determining the weight of evaluation indexes system objectively, and solving the uncertainty and ambiguity of result, AI (artificial intelligent) algorithm has disadvantage as well. The most obvious shortage of AI algorithm is the "black box" used in the calculating process which makes the researchers getting the results without understanding the procedure and detail of the whole process. The shortage of "black box" always arouses the doubts of the rationality of indexes’weights, and it is hard to interpret the weights. Meanwhile, such algorithms are relatively difficult to comprehend and implement, which limits their popularity and applicability.In order to solving the problems mentioned above, the weights-of-evidence method is introduced into land suitability assessment. Based on the principle and calculating flow of the weights-of-evidence method, a model for land suitability assessment is built in this research, short as WOE land suitability assessment. The proposed method applies the evidence layers as suitability evaluation indexes system, and extended the traditional connotation of the training points in WOE method, which we extracts them by overlay the multi-year land use data to find the unchanged areas. In this research cokriging interpolation method is applied both to revise the extracted training points and to verify the rationality and feasibility of the proposed training points extraction model. And then we build a set of evaluation tools in ArcGIS platform by Model Builder which integrated the WOE method and cokriging interpolation to realize the facilitation of land suitability assessment process and model sharing.Basic farmland suitability assessment and urban growth boundary(UGB) in Jinan City are used as case study to test the WOE land suitability assessment model proposed in this research. The results are as follows:(1) Being a discrete multivariate statistical method, the weights-of-evidence method is effective and objective for land suitability assessment, which obtains the results of land suitability assessment in the form of posterior probability. The ROC Curve shows the accuracy is better than 90%. At the same time the superposition of other research results has achieved a well precision, so as to verify the feasibility and effectiveness of the model. (2) The weights of indexes are determined by using the WOE model considering the correlation between training points and evaluation indexes system. The value of the weights are calculated based on the importance determined by conditional probability, so the weights are objective and easy to interpret. (3) The revise and verification of training points are realized through cokriging interpolation. Along with the results of ROC Curve and the accumulative percentage of test points, the results show the WOE model could improve the accuracy of assessment results to a certain degree. And the interpolation result can also verify the rationality and feasibility of the training points extraction model through the overlay with training points.