应用自组织映射方法的北京市2005-2013年土地利用时空演变分析

时间序列土地利用时空演变规律分析是当前的研究热点之一，通过应用自组织映射神经网络方法进行多时间序列土地利用变化时空一体化表达与演变规律分析，探索区域土地利用变化模式。基于北京市2005、2007、2009、2011、2013年5期土地利用遥感分类数据，构建自组织映射神经网络并利用其聚类和降维可视化功能对5个年份的土地利用数据同时进行训练输出，发现建设用地、耕地、林地、牧草地、园地的聚集模式，并通过对输出神经元进行二次聚类以及土地利用变化轨迹分析，获得北京市郊区5个监测时相土地利用变化的时空演变特征。结果揭示出北京市郊区2005-2013年土地利用变化具有明显的耕地型向建设用地型发展的平原区演变特征，以及向林地型发展的山区演变特征，且各区的发展具有时间上的顺序性；总体上形成6类土地利用演变轨迹。

Land use spatial-temporal evolution analysis using a self-organizing map in Beijing, 2005-2013

Multiple time series land using spatial-temporal evolution analysis is an important research area. In this study, we investigated the spatial-temporal integrated expression of multiple time series land use change. A self-organizing map (SOM) neural network was used to explore regional land use change modes and to analyze what has driven these changes. Remote sensing data for five land use classification data periods (2005, 2007, 2009, 2011, and 2013) for Beijing were used to train the network, and the outputs identified the aggregation modes for building land, farmland, forest land, grassland, and gardens by using the clustering, dimension-reducing, and visual functions of the SOM. Then we conducted second-step clustering to produce the neuron and build the land use change trajectories that are needed to analyze the spatial-temporal features of Beijing suburban land use changes during the five monitoring periods. The results revealed that there were two land use changes in the Beijing suburbs between 2005 and 2013. One was the development of buildings on farmland located on the plains and the other was the development of forest land in mountainous areas. Furthermore, development in each district had its own time sequences. This meant that we eventually obtained six land use change trajectories in total.