Comparison of three different methods to identify fractional urban signals for improving climate modelling

Urbanization has changed the properties of the Earth’s surface and resulted in modification of the biogeochemical cycle and possible climate feedback at global and regional scales. Such climate effects are especially evident locally over short periods in megacity areas. Climate model simulation and urbanization process analysis are often limited by poor accuracy of land-cover products that largely neglect mixed urban-surface information below certain thresholds. The present study compares three urban land identification methods (fractional cover, overlapping parabolic interpolation, and threshold) used in remote sensing and climate model parameterization with Landsat Thematic Mapper images and Moderate Resolution Imaging Spectroradiometer land-cover data sets in a systematic evaluation. We also analyse deviation induced by scaling effects and its influence on the urban radiation budget to better understand the implications of land-surface parameter deviation on regional climate analysis. A positive linear relationship is found between the spatial scale and urban-area deviation based on combined analysis of the three land identification methods, and deviation trends levelled off with an increase in the spatial scale. Coarse-resolution land-cover products could not capture well the urbanization process indicated by reference data from Beijing between 2000 and 2009, especially in urban fringe areas where major urban expansion was detected. Detailed sub-pixel information was possibly neglected by threshold methods, which resulted in strong deviation between land-cover products and actual conditions. The overlapping parabolic interpolation method used in climate models also produced deviation in surface parameter derivation during nested simulation work. This might further affect model performance at the regional scale and should be considered in climate model simulation.