空间形态高度异质性城市LCZ地图制作方法研究

Study on Local Climate Zone Mapping in Highly Spatially Heterogeneous City

城市气候问题随着全球气候变化和城市化进程加速而日趋严峻。空间形态高度异质性发展的大中型城市,亟需精细化的气候分区以指导城市气候工作。局地气候分区(Local climate zone,LCZ)满足这一需求,而基于免费易获取影像的遥感影像法是制作LCZ地图的主流。以广州为例,本文系统对比了现有遥感影像法制作LCZ地图的性能表现,发现现有方法精度不高,总体精度在70%以下。为此,本文结合分层分类法和图像监督分类法的优势提出一种新方法。新方法综合使用影像原始波段DN值和若干归一化差值型指数,以及神经网络算法分类结果等多种数据信息。对比发现,新方法的总体精度提高,达到84%,Kappa系数为0.82,其中建成类型分类精度的提高最为显著。新方法可用于空间形态复杂城市的气候分区,为城市热岛、城市规划、建筑节能和室外热舒适等城市气候相关工作提供基础。

With the global climate change and the fast urbanization, urban climate problem has become increasingly serious.The refined climate zoning is urgently needed in the highly spatially heterogeneous cities to guide their urban climate related work.The Local Climate Zone( LCZ) scheme meets this requirement,and the remote sensing image method that utilizes the free and readily achieved images is dominant for mapping LCZ.Taking Guangzhou as an example,this study systematically compared the performances of current remote sensing methods in LCZ mapping,and found that existing methods are of low accuracy,with an overall accuracy of below70%.A novel method was accordingly proposed,by combining the advantages of Hierarchical Classification and Image Supervised Classification.This method synthesized the original band DN value and several normalized difference indices,as well as the classification results by the neural network algorithm and various data information.A comparative analysis found that this novel method performs much better,with the overall accuracy of 84%,Kappa of 0.82,and the most significant increase in accuracy of built type classification.The proposed mapping method can be applied to the highly spatially heterogeneous cities,and lay foundation for urban heat island,urban planning,building energy conservation,outdoor thermal comfort and other urban climate related work.