Research On The Uncertainty Of Land Use Change Simulation Coupled Cellular Automata & Remote Sensing Imagery

In recent years, with the fast urbanization process in China, cities have been expanding rapidly. And land use pattern has been changed dramatically, which results in a lot of environmental problems. Therefore, it is very important to achieve a reasonable allocation of land use structure. For this, the accurate access to land use change simulation and prediction information is very useful, because it can provide a reference for the city manager to make land use policy. Cellular automata model can simulate the global change of complex systems through local interactions. It has been widely used to study land use change simulation. In the meantime, the rapid development of earth observation technology can provide abundant data, which greatly promotes the process of land use change simulation using cellular automata model.However, uncertainties exist in the process of land use change simulation by cellular automata model, including remote sensing classification uncertainty and cellular automata simulation uncertainty. This greatly affects the final results. On the one hand, the stochastic uncertainty and fuzzy uncertainty result in the classification uncertainty. On the other hand, the sensitivity of cellular automata and the error propagation of data source have serious impact on the simulation. Thus, it is crucial to effectively describe and analyze these uncertainties for improving the reliability of the simulation results.To solve these problems, firstly, the hybrid entropy model was proposed to evaluate the uncertainty of classification, from the pixel and category scale. Secondly, the response surface methodology was used to design experiments, analyze the sensitivity for each element, test the interaction between two elements, and find the optimal combination of scales. Then the data source error was produced based on hybrid entropy evaluation results. To explore the error propagation characteristic of the data source, the uncertainty was measured from three aspects:(1) the change of error size,(2) the change of overall spatial distribution, and(3) the change of spatial distribution for each land use type.The results show that the hybrid entropy model can express the stochastic uncertainty and fuzzy uncertainty of the classification result precisely. Moreover, it can intuitively reflect the spatial distribution of classification uncertainty. The response surface methodology was proposed to analyze the sensitivity of cellular automata model. It can recognize the interactions among cell size, neighborhood size and neighborhood type. And the regression equations between these factors and simulation accuracy were built. In addition, the optimal combination of scales was given. This contribution of this paper has improved the traditional analytical methods, which mainly aims at analyzing the sensitivity from the single factor. The hybrid entropy evaluation results were used to produce data source error. It has demonstrated that it is much closer to the real error distribution than stochastic distribution. The results indicate that the data close to the prediction time has a greater impact on the simulation accuracy of CA-Markov model.