Ren, Chao (1);
Xu, Yong (2);
Cai, Meng (1);
Li, Xinwei (1);
Ng, Edward (1);
Pesaresi, Martino (3);
Corban, Christina (3);
Florczyk, Aneta (3);
Politis, Panagiotis (3) 1: The Chinese University of Hong Kong, Hong Kong S.A.R. (China);
2: Purdue University, (USA);
3: European Commission, Joint Research Centre (Italy)
A fast urbanization has started since the late 1970s in China and this urbanization trend may still need another 20-30 years before the whole process completes. Their urban residents are especially vulnerable to current and future climate hazards because of high population density, compact urban setting and infrastructure and social-economic activities in landscapes that are exposed to sea-level rise, warming trends, and other extreme weather events (WMO, 2003, IPCC, 2014, UN-Habitat, 2011). Thus, recently the central government of China has paid a high attention to environmental protection and recover and tried to improve the quality of citizens’ living environment. However, there is very limited land use land cover (LULC) open data available to access and there is an increasing interest in developing a national-scale database on Chinese cities for scientific inquiry and policy formulation.
According to the traditional way to classify the LULC, normally only three types for built-up areas, i.e. town-centre, sub-urban and rural areas. But this information cannot fully represent complicated urban morphological characteristics. In this study, the Local Climate Zone (LCZ) classification system (Stewart & Oke, 2012) with ten built-up types and seven natural land cover types was adopted. The study aims to evaluate the performance of the LCZ classification system for the entire country of China, and also to improve the original LCZ mapping workflow (Bechtel, et al., 2015) for generating a high accuracy LCZ data at the regional and the country levels.
A case study was conducted for the whole China. First, more than 700 Landsat-8 surface reflectance images from the year 2015 to 2016 were collected as input data and training samples of 17 types of LCZ were selected from six selected cities (Guangzhou, Shanghai, Xian, Wuhan, Tibet, Xining) which are evenly distributed in China. Secondly, by adopting the random forest classifier (Bechtel., 2015), all spectral information within the training samples from Landsat-8 data were used to train a standard classifier, and then the obtained classifier was applied to generate the LCZ mapping results using all images of China. Finally, all the LCZ mapping results were mosaicked to generate a complete LCZ map for China. Our experimental results indicated that the overall accuracy of developed LCZ map can achieve about 60-70%, except for some high-density cities as well as some western cities in China, with relatively low overall accuracy about 50-60%.
The developed national LCZ map of China can provide researchers, scientists and the practitioners with a useful dataset and spatial information platform of urban morphology and land cover. If linked other geo-referenced urban information, there are many possibilities for various applications such as climatic-sensitive planning, future land use prediction, and climate-change caused healthy impact analysis.
References:
WMO. Our Future Climate. Geneva, Switzerland: WMO, 2003
IPCC. Arc5: Impacts, Adaptation and Vulnerability: Summary for Policymakers. 2014
UN-Habitat. Global Report on Human Settlements 2011: Cities and Climate Change. London, Washington,DC: Earthscan; 2011.
Stewart ID, Oke TR. Local Climate Zones for Urban Temperature Studies. Bulletin of the American Meteorological Society. 2012; 93:1879-1900.