Comparisons of urban-related warming for Shenzhen and Guangzhou

Urban-related warming in two first-tier cities (Guangzhou and Shenzhen) in southern China with similar large-scale climatic backgrounds was compared using the nested weather research and forecastin...


Reconstruction of annual urban gridded data
The general trends of the urban surface expansion and their spatial patterns over China were determined based on the integrated data from population information, multiple-source satellite images, and National Land Cover Datasets obtained from the Chinese Data Sharing Infrastructure of Earth System Science (Jia et al. 2014). At the same time, the nighttime light datasets from the Defense Meteorological Satellite Program-Operational Linescan System, which could reflect socioeconomic activities for urbanization and population growth, were also considered. The datasets showing the closest results in describing urban area fractions and the evaluations were then chosen and combined to construct five urban fraction images over China (1980China ( , 1990China ( , 2000China ( , 2010China ( , and 2016Jia et al. 2014;Hu et al. 2015) at the nested model resolutions of 30, 10, and 3.3 km, respectively, using 1-km urban surface data.
Fractional urban data for individual year from 1980 to 2016 were then reconstructed based on the five fractional images on each model grid cell. The increase in fractional urban areas was assumed to increase linearly during each time period (1980-1989, 1990-1999, 2000-2009, 2010-2016), for which the adoption of annual fractional urban areas could avoid unrealistic discontinuity-induced spurious values during long-term numerical integrations. The annual fractional urban data could offset the scarcity of satellite-based retrieved land use data in a certain degree.
However, differences between the reconstructed data and the real urban surface distributions still existed, which might induced error between the simulated results and observed values.
Annual land-use data for the coarse and nested domains, instead of the default land-use data in the WRF model, were obtained based on the reconstructed annual fractional urban area data and other land-use categories from the default land-use data.
For the WRF model before version 3.6 (here version 3.4 are adopted), only one dominant land-use category was assigned to each grid cell during the numerical integrations according to the methods by Guo and Chen (1994).

Driving data
The initial conditions and time-varying boundary conditions for the WRF model were provided by the National Centers for Environmental Prediction (NCEP) -the Department of Energy (DOE) Atmospheric Model Intercomparison Project (AMIP-II) reanalysis dataset during 1979 and 2016 (R-2, Kanamitsu et al. 2002). During the model integrations, the identical driving data, including sea surface temperatures and atmospheric data were used, and the forcing was only applied at the boundaries. The reanalysis data with a resolution of 2.5° × 2.5° were interpolated into the WRF model domain with the bilinear method and updated every six hours.

Experimental design
The central latitude and longitude of the simulated domain located at 35°N and 108.5°E, respectively, in which nested domains (30-10-3.3 km) and two-way feedback was adopted. In order to include urban effects in computing energy and water exchanges between the land surface and atmosphere, the unified Noah land-surface model (including a four-layer soil model and urban canopy model with the default urban-related parameters, Chen et al. 2006), was adopted in the integrations. Other physical parameterization schemes adopted in the integrations included the WRF single-moment six class graupel microphysics scheme, the Community Atmosphere Model shortwave and longwave radiation schemes, the Yonsei University boundary-layer scheme, and the Grell 3D ensemble cumulus scheme (for 30 km and 10 km-resolution integrations only).    1980 1983 1986 1989 1992 1995 1998 2001 2004 2007  Year 1980 1983 1986 1989 1992 1995 1998 2001 2004 2007