基于元胞自动机的莫莫格湿地土地覆被预测模拟

Land-Cover Simulation and Forcast of Melmeg Wetland Using Cellular Automata

本文运用元胞自动机结合人工神经网络方法,利用1986年和2004年两时期的ETM遥感影像,对莫莫格国家级湿地自然保护区的土地覆被变化进行了动态的模拟和预测,探讨了元胞自动机模型在湿地土地覆被变化预测和湿地资源保护方面的重要作用,为湿地资源的动态监测与可持续发展提供参考和决策支持。在模型实现过程中,以正六边形元胞取代传统的四边形元胞,克服了传统四边形元胞各向异性的先天不足,减小了模型预测的误差。应用BP神经网络与元胞自动机相结合自动挖掘出元胞自动机模型的转换规则,不仅降低建立多种土地覆被类型预测模拟的难度,而且减少了人为的主观因素,提高了准确性。将模型应用到莫莫格湿地保护区,预测的结果与实际土地覆被类型比较得出预测精度达到约80%,通过对该地区未来34年土地覆被变化的预测,反映出了该地区耕地面积增大、湿地减少的生态环境变坏的趋势,为该保护区的管理和发展提供了参考。

This paper presents a model to simulate and forecast wetland land-cover changes with geographic cellular automata. As one of the three ecosystems, Wetland ecosystem has so many functions in regulating the climate and water circulation on earth and wetland is the important resource for human being. But in recent years, the function of wetland has been degenerating severely with the increased human activity. How to protect the wetland resource has become a hot issue in the wetland research domain. Cellular automata, a ＂Bottom-to-up＂ dynamic system with powerful function in simulation have been used to simulate complex phenomena in many domains. The article expounds the importance of cellular automata in protecting the wetland resource in many aspects, especially in dynamically forecasting and simulating the trend of wetland land-cover, which will provide the important reference for the management and development of wetland. Considering the traditional square cell of CA has intrinsic deficiency of anisotropy because of its geometry shape, the model replaces it with hexagon cell which has better character in isotropy. The model, which uses the back-propagation neural network to mine transition rules of cellular automata, has a lot of merits that not only reduces the difficulty to establish the transition rule because of many kinds of land-cover types, but also reduces subjective idea and then improve the simulation accuracy. Finally, this model is applied to the Melmeg National Reserve Wetland. The result, which owned accuracy of about 80%, shows that the model is suitable for simulating wetland land-cover change. The forecast and simulation is based on the hypothesis that the development velocity in study area is the same in a short period. By simulation, it＇ s clear that, in Melmeg wetland, the area of farmland will increase while wetland will decrease, and the grassland and forest may disappear in the coming years. The environment of wetland ecosystem of Melmeg will be worsened according to the simulation.