Content uploaded by Mengmeng Li
Author content
All content in this area was uploaded by Mengmeng Li on Jan 19, 2021
Content may be subject to copyright.
Improving mesoscale modeling using satellite-derived
land surface parameters in the Pearl River Delta
region, China
Mengmeng Li
1
, Yu Song
1
, Xin Huang
1
, Jianfeng Li
1
, Yi Mao
1
, Tong Zhu
1
, Xuhui Cai
1
, and Bing Liu
2
1
State Key Joint Laboratory of Environmental Simulation and Pollution Control, Department of Environmental Science,
Peking University, Beijing, China,
2
China National Environmental Monitoring Center, Beijing, China
Abstract Land surface parameters play an important role in mesoscale modeling by regulating the regional
heat flux and hydrological cycle. Recently, significant urbanization and afforestation occurred in the Pearl River
Delta (PRD) region, China, which exert an important effect on local meteorology and thermal circulation. But
previous studies failed to capture the complex changes of the surface characteristics in the PRD and thus were
difficult to accurately describe the land-atmosphere coupling. In this study, high-resolution Moderate
Resolution Imaging Spectroradiometer observations are used to specify the land cover type, green vegetation
fraction, and leaf area index in the Weather Research and Forecasting model. Comparisons with ground-based
observations during eight episodes, as well as satellite measurements, all indicate an improved model
performance when the satellite-derived land surface parameters are assimilated. Moreover, the remote sensing
data accurately reflect the surface inhomogeneity and successfully represent the intensity and spatiotemporal
characteristics of the urban heat island (UHI) effect. The UHI effect in turn modifies the local thermal circulation
by enhancing the urban-rural horizontal advection and initiating the urban heat island circulation, as well as
interacting with the sea/land breeze over the PRD. This work not only improves the understanding of local
meteorological simulation andforecasting but also sets the stagefor further research on the feedback between
air quality and meteorological responses due to land cover changes.
1. Introduction
The Earth’s surface has been and is continuing to be greatly modified by natural and human activities,
particularly in the most rapidly developing countries [Lambin et al., 2003; Liu et al., 2005; Liu and Tian, 2010].
These changes bring about numerous environmental consequences for the global biogeochemical cycle
[Houghton and Hackler, 2003; van der Gon, 2000] and climate feedback [Brovkin et al., 2004; Findell et al., 2007].
More importantly, as the fundamental interface for land-atmosphere interactions, several key components of
the land surface, such as land cover, vegetation, and soil texture, directly determine the surface physical
properties (e.g., albedo, emissivity, stomatal resistance, and roughness) and in turn modulate the regional
exchanges of energy, water, and momentum [Molders, 2001; Pielke, 2001; Wetzel and Chang, 1988]. Numerical
modeling provides an important tool for analyzing and understanding the role of land surface parameters in
land-atmosphere coupling. Molders [2001] explores the uncertainty of surface parameters (albedo, roughness,
heat capacity, conductivity, etc.) in mesoscale modeling and suggests that these parameters appreciably
affect water and energy fluxes, as well as cloud and precipitation formation. According to the study, domain-
specific and actual parameters are preferred in mesoscale modeling when available. Wetzel and Chang [1988]
investigate the relative importance of five land surface parameters in regional evapotranspiration. They find
that soil moisture, leaf area index, and fractional green vegetation cover are most important parameters,
followed by albedo and roughness length. All of these studies conclude that accurate representation of land
surface features is critical in land-atmosphere coupling and is of particular concern in mesoscale modeling.
Satellite remote sensing provides excellent and continuous monitoring of various land surface parameters
[Bartholome and Belward, 2005; Friedl et al., 2002; Loveland et al., 2000]. Remote sensing has been widely used
in numerical simulations to improve the model representation of the inhomogeneous landscape or to
understand the changes of local weather [Crawford et al., 2001; de Foy et al., 2006; Gutman and Ignatov, 1998;
Meng et al., 2009; Yucel, 2006]. de Foy et al. [2006] use high-resolution Moderate Resolution Imaging
Spectroradiometer (MODIS) satellite observations to specify the land cover, vegetation fraction, albedo, and
LI ET AL. ©2014. American Geophysical Union. All Rights Reserved. 6325
PUBLICATION
S
Journal of Geophysical Research: Atmospheres
RESEARCH ARTICLE
10.1002/2014JD021871
Key Points:
•MODIS-derived land cover and
vegetation parameters are assimilated
in WRF
•Comparisons with observations
reveal a significantly improved
model performance
•Urban heat island is generated and
interacts with the sea/land breeze
Supporting Information:
•Readme
•Text S1
•Figure S1a–S1f
•Figure S2a–S2f
Correspondence to:
Y. Song,
songyu@pku.edu.cn
Citation:
Li, M., Y. Song, X. Huang, J. Li, Y. Mao,
T. Zhu, X. Cai, and B. Liu (2014),
Improving mesoscale modeling using
satellite-derived land surface parameters
in the Pearl River Delta region, China,
J. Geophys. Res. Atmos.,119, 6325–6346,
doi:10.1002/2014JD021871.
Received 6 APR 2014
Accepted 9 MAY 2014
Accepted article online 16 MAY 2014
Published online 5 JUN 2014
surface temperature in the MM5 (Fifth-Generation Mesoscale Model) model. They yield improved
meteorological simulations for the wind circulation patterns and the urban heat island in Mexico City. Using
MM5, Yucel [2006] implements a new land cover map and albedo data set from MODIS observations and
reveals that changes in land cover specification or associated parameters affect surface wind, temperature,
and humidity, which in turn result in perceivable alterations in the planetary boundary layer evolution.
The Pearl River Delta (PRD) region, which is located on the southeastern coast of China facing the South China
Sea, is one of the largest metropolitan agglomerations and most densely populated areas in China. Currently,
this area covers approximately 41,700 km
2
and contains nine cities, in addition to Hong Kong and Macao:
Guangzhou, Shenzhen, Foshan, Dongguan, Huizhou, Jiangmen, Zhongshan, Zhaoqing, and Zhuhai. Over the
past two decades, continuous satellite observations revealed significant urban expansion and afforestation at
the expense of cropland due to the rapid economic development and policy adjustment [Seto et al., 2002;
Weng, 2002].
The meteorological responses to urbanization in the PRD region have received considerable attention and
are well documented by numerous observations [J. L. Chen et al., 2013; X. L. Chen et al., 2006; Fan et al., 2011] or
modeling studies [Lin et al., 2007; Lo et al., 2007, 2006; Wang et al., 2007, 2009]. For example, the 30 year
weather record reveals a 0.22°C/10 year increase in the annual mean temperature due to urban expansion
[J. L. Chen et al., 2013]. The intensive observations by Fan et al. [2011] during the summer of 2006 clearly
reveal discrepancies in the lower tropospheric wind regimes and thermodynamic structures among various
sites due to the urban expansion and urban heat island circulation. Lo et al. [2006, 2007] incorporate a
new land cover map from field investigations into MM5 and suggest that precise urban surface data are
critical to capture the major features of urban heat island effect and land-sea breeze circulation in the PRD.
Wang et al. [2007, 2009] investigate the effects of urbanization on local meteorological simulation and find
a typical warm island and generally stronger boundary layer mixture in the PRD.
However, all of the cited studies in the PRD only consider the replacement of land cover type in the mesoscale
model and thus fail to fully represent the other closely related land surface parameters, especially the
seasonal vegetation characteristics such as green vegetation fraction (GVF) and leaf area index (LAI). Several
studies indicate that the abundance of urban vegetation has a significant influence on the simulated near-
surface air temperature [Taha, 1996], as well as the vertical structure of the atmospheric boundary layer
[Pielke and Uliasz, 1998]. Betts et al. [1997] also note that the introduction of satellite-derived green vegetation
fraction into the land surface model could result in improved model forecasts of land surface fluxes and
planetary boundary layer structure. Thus, a combination of reliable estimates for all of the land surface
parameters is essential to produce good forecasts, and the inconsistency of land surface parameters in
previous simulations over the PRD may cause errors in the mesoscale model.
This study attempts to improve the mesoscale simulations by applying the high-resolution MODIS
observations to specify the main land surface parameters, including land cover type, green vegetation
fraction, and leaf area index. The model setup, land surface parameters, and experimental designs are
presented in section 2. Section 3 follows with the evaluation of the model performance, as well as a
quantitative assessment of the impacts on relevant meteorological variables and the local wind circulation.
Finally, a summary and conclusion are presented in section 4.
2. Methods and Data
2.1. WRF Model
The meteorological model used in this study is the National Center for Atmospheric Research Advanced
Research Weather Research and Forecast (WRF) model. WRF is a sophisticated three-dimensional
compressible and nonhydrostatic numerical weather simulation and prediction model that is widely used for
numerical weather prediction, hydrology, and air quality studies [Skamarock et al., 2008].
Here a two-way three-nested grid system (Figure 1), which is centered at 34.5°N, 108.9°E, is established. The
outmost domain covers the entire area of China to capture the synoptic-scale features, with a horizontal grid
resolution of 36 km and 170 ×124 grids. The second domain covers parts of South China, including
Guangdong, Hainan, Guangxi, Hunan, and Fujian, with a grid resolution of 12 km and 100 × 91 grids in the
horizontal direction. The innermost domain focuses on the PRD region and is designed to resolve the
Journal of Geophysical Research: Atmospheres 10.1002/2014JD021871
LI ET AL. ©2014. American Geophysical Union. All Rights Reserved. 6326
local-scale circulation features; it has a grid resolution of 4 km and 97 × 82 grids in the horizontal direction. All
the three domains have 28 vertical layers that extend from the surface to 5 hPa. The 6 h NCEP (National
Centers for Environmental Prediction) global final analysis data at 1° × 1° resolution are used as the initial and
boundary meteorological conditions.
In this study, the sophisticated Noah land surface scheme [Chen et al., 1996; Ek et al., 2003] is chosen to
describe the land-atmosphere interactions, which incorporates detailed land surface processes of
thermodynamics and hydrology, such as vegetation evapotranspiration, soil drainage and runoff, and
moisture diffusion. The other key physical parameterized options for the WRF modeling are as follows: the Lin
microphysics scheme [Lin et al., 1983] with the Kain-Fritsch cumulus parameterization [Kain and Fritsch,
1992] to describe the cloud and precipitation processes; the Yonsei University (YSU) boundary layer scheme
[Noh et al., 2003], the Goddard short wave radiation scheme [Chou and Suarez, 1999], the Rapid Radiative
Transfer Model (RRTM) long-wave radiation scheme [Gallus and Bresch, 2006], and the Monin-Obukhov
surface similarity scheme [Monin and Obukhov, 1954].
2.2. Satellite Land Surface Parameters
2.2.1. Land Cover
Land cover directly determines the secondary surface physical parameters and modifies the earth-atmosphere
heat and water fluxes. The standard WRF initial static field uses the 24-category U.S. Geological Survey (USGS)
global 1km land cover map (Figure 2a). The USGS land cover map is derived from the monthly Advanced Very
High Resolution Radiometer (AVHRR) normalized difference vegetation index (NDVI) observations spanning
April 1992 to March 1993 using the unsupervised clustering algorithms [Loveland et al., 2000], which could not
representthe real surface features due to the rapid land cover changes. On the other hand, Frolking et al. [1999]
report that the estimated croplandarea based on AVHRR data set is overestimated byabout 50%in South China
when compared with agricultural census data, which induces great errors in the WRF model.
Alternatively, WRF also allows the usage of the 20-class MODIS land cover data set for the Noah land surface
model (LSM), which contains 17 land cover types defined by the International Geosphere-Biosphere Program
(IGBP) [Friedl et al., 2002] plus three classes of tundra [Justice et al., 2002]. However, the original MODIS data
set in WRF is obtained for the base year 2001 and is designed to merely work with the Noah LSM. Therefore,
for future simulations or for simulations with other land surface schemes, it is needed to provide another way
of incorporating updated land cover data set. In this study, we construct a new land cover map in conjunction
with the MODIS land cover type product (MCD12Q1) for the year 2006 and water mask data (MOD44W) for
Figure 1. (a) Three-nested WRF modeling domains and (b) locations of the meteorological stations in the inner domain. The background shaded contour represents
the topography height (m); the red and blue symbols represent the urban and rural stations, respectively; the four up triangles represent the ground stations
selected for the time series analysis.
Journal of Geophysical Research: Atmospheres 10.1002/2014JD021871
LI ET AL. ©2014. American Geophysical Union. All Rights Reserved. 6327
the year 2000, both with a resolution of 500 m [Friedl et al., 2002]. The 17 MODIS land cover categories
defined by the IGBP are projected onto the existing 24 USGS categories as shown in Table 1. The coverage
fraction of each land cover type (“LANDUSEF”in WRF) is also recalculated based on the newly constructed
land cover map.
Figure 2 indicates that the terrestrial ecosystem demonstrates significant changes between the two data sets
and that the land surface heterogeneity is more accurately portrayed in the modified case. During the early
1990s, the entire PRD region was characterized by substantial amounts of cropland (5.2 × 10
4
km
2
) (Figure 2a).
Whereas in present day (Figure 2b), unprecedented urban or built-up land expansion occurred in the PRD
(totaling approximately 5.0 × 10
3
km
2
), especially in the central region along the Pearl River Estuary. The
urban expansion comprises two megacities (i.e., Guangzhou and Hong Kong), numerous medium-sized cities
(e.g., Shenzhen, Foshan and Dongguan), and other smaller cities (Figure 2). As cities developed, vegetation
(especially intensive agricultural land) was lost. Another noticeable difference between the two land data sets
is the large growth of forests and grassland in the neighboring regions of Huizhou and Jiangmen, along with
the removal of cropland (Figure 2). A detailed comparison emphasizes that over the past two decades, the
urban area, forest cover area, and grassland expanded by as much as 27 times, 133% and 80%, respectively, at
the expense of cropland (a 66% reduction).
2.2.2. Green Vegetation Fraction
Green vegetation fraction is defined
as the fraction of a grid cell for which
midday downward solar insolation is
intercepted by photosynthetically
active green canopy [Chen and
Dudhia, 2001]. In the Noah LSM, the
spatially and temporally varying GVF
is used to determine the surface
energy partition over the fractional
vegetation and bare ground
separately. The GVF parameter has a
large impact on the latent heat flux
because it controls evapotranspiration
[Chen et al., 1996; Ek et al., 2003;
Jacquemin and Noilhan, 1990].
By default, WRF determines the
climatological vegetation fraction
based on AVHRR NDVI observations
collected from 1985 through 1990 at
a resolution of 20 km [Gutman and
Table 1. Conversion of the MODIS IGBP Land Cover Classes to
USGS Classifications
MODIS USGS
0 Water 16 Water bodies
1 Evergreen Coniferous Forest 14 Evergreen Coniferous forest
2 Evergreen Broadleaf Forest 13 Evergreen Broadleaf forest
3 Deciduous Coniferous Forest 12 Deciduous Coniferous forest
4 Deciduous Broadleaf Forest 11 Deciduous Broadleaf forest
5 Mixed Forest 15 Mixed Forest
6 Closed Shrubland 8 Shrubland
7 Open Shrubland 9 Mixed Shrubland/Grass
8 Woody savanna 10 Savanna
9 Savanna 10 Savanna
10 Grassland 7 Grassland
11 Permanent wetland 17 Herb. Wetland
12 Cropland 5 Crop/Grass Mosaic
13 Urban and Built-up 1 Urban
14 Cropland/Natural Vegetation Mosaic 6 Crop/Wood Mosaic
15 Snow and Ice 24 Snow or Ice
16 Barren or Sparsely Vegetated 19 Barren or Sparsely Vegetated
254 Unclassified 25 No Data
Figure 2. Land cover map for the inner domain in (a) WRF default and (b) WRF modified.
Journal of Geophysical Research: Atmospheres 10.1002/2014JD021871
LI ET AL. ©2014. American Geophysical Union. All Rights Reserved. 6328
Ignatov, 1998]. In this study, the monthly 1 km NDVI product (MOD13A2) for 2006, which is produced from
visible and near-infrared data acquired by the MODIS sensor, is used to estimate the current distribution of
green vegetation fraction [Justice et al., 2002; Purevdorj et al., 1998].
Serious underestimation of rural vegetation cover and overestimation of urban vegetation cover are
observed in the AVHRR-based data set. Take October, for example; Figure 3 present the distributions of green
vegetation fraction in the inner domain for the default and modified cases in October. The coarse resolution
and gradual spatial variations of the default data set fail to capture the spatial patterns of green vegetation
fraction. Low vegetation cover (below 0.6) is generally observed across the entire inner domain (Figure 3a). By
comparison, the MODIS GVF composite (Figure 3b) exhibits a more mixed spatial structure and significantly
higher vegetation cover (over 0.8) is found over the surrounding forested regions. Furthermore, less
vegetation cover is observed in the urban area (below 0.1) and adjacent cropland as compared to the
AVHRR data.
2.2.3. Leaf Area Index
LAI is defined as the one-sided green leaf area per unit ground area (m
2
/m
2
). In WRF, leaf area index is
dependent on vegetation type and is a basic parameter to determine the vegetation stomatal resistance of
latent heat exchange in the Jarvis-type approach [Jacquemin and Noilhan, 1990; Jarvis, 1976]. By default, it is
simply calculated online using the tabulated value assigned to each land/vegetation type in conjunction with
the climatological green vegetation fraction data. In this study, the modified LAI distribution is directly
substituted with 8 day MODIS satellite measurements (MCD15A2) in 2006 at 1 km resolution and resampled
on a monthly basis [Knyazikhin et al., 1998].
Figure 4 presents the distributions of leaf area index for the default and modified cases in October. As pointed
out by Carlson and Ripley [1997], the values of LAI have a strong dependence on NDVI and GVF and thus show
a similar spatial pattern as GVF. As detected by the MODIS sensor, the LAI high-value centers (approximately
5.0) coincide with the locations of forests, whereas the urban LAI generally drops below 1.5–2.0 (Figure 4b).
The comparisons in Figure 4 also reveal a systematic overestimation of default LAI based on look-up tables,
and this phenomenon is also found for other seasons. The default values exceed 5.0 over nearly the entire
study domain, which are roughly 2.0 greater than the MODIS measurements.
2.3. Experiment Designs
Two parallel WRF simulations that are respectively designated as “WRF default”and “WRF modified”are
conducted, with identical physical parameterization schemes, meteorological initial conditions, and lateral
boundary conditions (as described in section 2.1); however, the simulations have different representations for
Figure 3. Green vegetation fraction in October for the inner domain in (a) WRF default and (b) WRF modified.
Journal of Geophysical Research: Atmospheres 10.1002/2014JD021871
LI ET AL. ©2014. American Geophysical Union. All Rights Reserved. 6329
the land surface characteristics. The WRF-default experiment is defined as a control simulation with the static
geographical field input set to the default values. The modified case (WRF modified) is conducted with
updated land surface parameters (i.e., land cover, green vegetation fraction, and leaf area index) that are all
produced from the MODIS observations as described in section 2.2. Both cases use the same 17-category soil
texture map from the 5 min United Nations/Food and Agriculture Organization database. The remaining
secondary surface parameters (e.g., albedo and roughness length) are assigned by tabulated values in the
similar way according to land cover type, green vegetation fraction, and soil index in WRF.
Numerous observations [Wang et al., 1990; Weng and Yang, 2004] indicated that during the dry seasons
(September–December and January–March) characterized by calm and clear conditions, the heat/moisture
budget at the land surface exerts greater effects on local meteorology, and stronger urban heat island was
observed for Guangdong. While during the rainy seasons (April–August), the large cloud cover, frequent
precipitation, and unstable weather system accelerate the heat exchange with the surrounding areas and
suppress the development of urban heat island.
Thus, in this study, eight cases are run to represent the model improvements of different seasons, including
six dry season episodes (17–23 September, 25–29 September, 4–8 October, 10–13 October, 7–10 November,
and 16–21 December) and two wet season episodes (19–24 July and 12–17 August), which are all
characterized by clear sky, no precipitation, and long sunshine duration. Particularly, during the dry season
episode of 4–8 October, South China is controlled by the maritime high-pressure system over the Pacific
(Figure 5) and is not affected by strong synoptic system. Therefore, weak synoptic forcing, calm conditions,
and clear skies are typical for the PRD region during this period, which provide optimal conditions for the
development of relatively strong thermal circulation and to explore the impacts of surface features on the
local meteorology.
In the following sections, detailed statistical evaluations of model results against ground observations will be
discussed for all the eight episodes (section 3.1). Results of urban heat island effects (section 3.2) and local
thermal circulation (section 3.3), however, will be provided for the 5 day simulation during 4–8 October.
3. Results and Discussions
3.1. Model Evaluation
The model results for the eight episodes are compared with hourly ground observations that are obtained
from 57 automatic meteorological stations throughout the inner domain (Figure 1). We define a monitoring
station as a typical urban station if it is located in an urban grid in the innermost domain as identified by the
Figure 4. Leaf area index in October for the inner domain in (a) WRF default and (b) WRF modified.
Journal of Geophysical Research: Atmospheres 10.1002/2014JD021871
LI ET AL. ©2014. American Geophysical Union. All Rights Reserved. 6330
MODIS land cover data set; otherwise, the monitoring station is regarded as a rural station. In total, the
automatic stations are classified as 27 urban stations and 30 rural stations (Figure 1b).
First, the simulated near-surface meteorological variables are compared with observations in terms of
performance statistics, including mean bias (MB), root-mean-square error (RMSE), fractional bias (FB), fractional
error (FE), and index of agreement (IOA) (Table 2). For the vector variable (i.e., wind direction), only the statistical
results for mean bias and root-mean-square error are given in Table 2. The values of the statistical measures for a
perfect model would be 0.0 for MB, RMSE, FB and FE, and 1.0 for IOA. The statistical metrics for all the eight
episodes demonstrate visible model improvements attributed to the updated land surface parameters.
Due to the coarse representation of land surface characteristics, obvious underestimation of 2 m air
temperature is generated over both the urban and rural regions in the WRF-default simulation whencompared
with the actual observations, resulting in a domain-average cold bias from 0.67°C to 1.78°C for the eight
episodes (Table 2). The inclusion of MODIS measurements leads to different degrees of model improvement for
the eight cases, with the mean bias over the entire inner domain reduced by 0.6°C to 1.4°C and the index of
agreement reaching approximately 0.95. Particularly, it is also noticed that the default case seriously
underestimates the urban temperature (Table 2). For example, during 4–8 October the temperature bias
reaches 2.05°C at the urban sites, which is sharply reduced to 0.19°C after the inclusion of
MODIS measurements.
Figure 5. Korean Meteorological Administration (KMA) weather map at 500 hPa at 0800 LT, 6 October (downloaded from the Korea Meteorological Administration
website: http://web.kma.go.kr/).
Journal of Geophysical Research: Atmospheres 10.1002/2014JD021871
LI ET AL. ©2014. American Geophysical Union. All Rights Reserved. 6331
Table 2. Quantitative Performance Statistics for the Meteorological Simulations in WRF Modified and WRF Default (in Parentheses)
Variable Index
Episode 1 (19–24 July) Episode 2 (12–17 August)
Urban Rural Total Urban Rural Total
2 m temperature (°C) MB
a
0.34(1.79) 0.84(1.26) 0.62(1.50) 0.06(1.40) 0.53(0.85) 0.27(1.09)
RMSE
b
1.33(2.23) 1.71(1.92) 1.55(2.06) 1.31(1.92) 1.58(1.70) 1.47(1.80)
FB
c
0.01(0.06) 0.03(0.04) 0.02(0.05) 0.01(0.05) 0.02(0.03) 0.01(0.04)
FE
d
0.03(0.06) 0.05(0.06) 0.04(0.06) 0.03(0.06) 0.04(0.05) 0.04(0.05)
IOA
e
0.94(0.85) 0.93(0.91) 0.94(0.89) 0.94(0.88) 0.94(0.92) 0.94(0.91)
2 m relative humidity (%) MB 3.61(13.17) 9.38(13.81) 5.60(13.44) 3.37(12.46) 10.29(13.69) 5.73(12.88)
RMSE 10.66(16.14) 12.14(15.66) 11.19(16.02) 10.97(16.07) 13.53(16.12) 11.90(16.09)
FB 0.04(0.18) 0.14(0.19) 0.07(0.18) 0.04(0.17) 0.14(0.19) 0.08(0.17)
FE 0.13(0.19) 0.15(0.20) 0.13(0.19) 0.13(0.18) 0.16(0.19) 0.14(0.19)
IOA 0.85(0.72) 0.85(0.77) 0.85(0.74) 0.83(0.70) 0.79(0.75) 0.82(0.71)
10 m wind speed (ms
1
) MB 0.36(0.55) 0.53(0.60) 0.48(0.57) 0.53(0.55) 0.85(0.92) 0.54(0.56)
RMSE 1.44(1.55) 1.56(1.65) 1.50(1.60) 1.39(1.52) 1.52(1.61) 1.46(1.56)
FB 0.07(0.23) 0.21(0.22) 0.14(0.22) 0.16(0.24) 0.20(0.21) 0.19(0.22)
FE 0.52(0.59) 0.65(0.68) 0.59(0.64) 0.50(0.57) 0.62(0.66) 0.56(0.61)
IOA 0.55(0.52) 0.53(0.51) 0.54(0.52) 0.61(0.60) 0.59(0.57) 0.61(0.59)
Wind direction (degree) MB 13.16(7.49) 8.76(6.87) 10.71(7.15) 1.40(7.41) 6.60(7.39) 4.30(7.40)
RMSE 73.52(75.03) 85.78(86.44) 80.59(81.60) 71.01(74.89) 84.53(85.19) 78.82(80.78)
Episode 3 (17–23 September) Episode 4 (25–29 September)
Variable Index Urban Rural Total Urban Rural Total
2 m temperature (°C) MB 0.48(1.73) 0.42(1.14) 0.01(1.41) 0.19(2.03) 0.90(1.57) 0.41(1.78)
RMSE 1.36(2.35) 1.58(2.08) 1.49(2.21) 1.11(2.59) 1.70(2.22) 1.46(2.40)
FB 0.02(0.07) 0.02(0.04) 0.01(0.06) 0.03(0.08) 0.03(0.06) 0.02(0.07)
FE 0.04(0.08) 0.05(0.07) 0.04(0.07) 0.01(0.08) 0.05(0.07) 0.04(0.08)
IOA 0.93(0.84) 0.95(0.90) 0.94(0.88) 0.97(0.85) 0.95(0.91) 0.96(0.89)
2 m relative humidity (%) MB 3.81(8.60) 4.69(9.04) 2.10(8.73) 2.96(9.03) 6.97(10.35) 0.03(9.43)
RMSE 11.15(12.79) 11.29(13.49) 11.20(13.00) 8.99(12.94) 11.70(13.95) 9.89(13.25)
FB 0.07(0.15) 0.09(0.16) 0.03(0.15) 0.06(0.12) 0.11(0.15) 0.01(0.13)
FE 0.15(0.18) 0.15(0.19) 0.15(0.18) 0.12(0.15) 0.15(0.17) 0.13(0.16)
IOA 0.79(0.79) 0.82(0.76) 0.80(0.79) 0.89(0.82) 0.84(0.78) 0.87(0.81)
10 m wind speed (ms
1
) MB 1.12(1.35) 1.25(1.56) 1.18(1.45) 0.84(1.05) 0.95(1.15) 0.90(1.09)
RMSE 1.74(2.07) 1.92(2.20) 1.83(2.14) 1.58(1.92) 1.73(1.92) 1.65(1.92)
FB 0.45(0.48) 0.49(0.56) 0.47(0.52) 0.36(0.39) 0.40(0.46) 0.38(0.42)
FE 0.54(0.57) 0.59(0.64) 0.57(0.60) 0.56(0.62) 0.65(0.68) 0.60(0.65)
IOA 0.58(0.56) 0.59(0.56) 0.59(0.56) 0.51(0.44) 0.49(0.45) 0.50(0.45)
Wind direction (degree) MB 24.98(24.71) 26.61(26.90) 25.91(25.97) 39.87(43.88) 39.39(37.85) 38.77(41.44)
RMSE 55.32(56.27) 68.14(68.43) 62.57(63.15) 66.38(69.19) 76.96(76.43) 72.01(73.51)
Episode 5 (4–8 October) Episode 6 (10–13 October)
Variable Index Urban Rural Total Urban Rural Total
2 m temperature (°C) MB 0.19(2.05) 0.93(1.31) 0.60(1.64) 0.45(0.99) 0.23(0.53) 0.08(0.73)
RMSE 1.38(2.65) 1.79(2.20) 1.62(2.41) 1.39(1.75) 1.42(1.49) 1.41(1.61)
FB 0.01(0.08) 0.04(0.05) 0.02(0.06) 0.02(0.04) 0.01(0.02) 0.01(0.03)
FE 0.04(0.09) 0.06(0.07) 0.05(0.08) 0.04(0.06) 0.04(0.05) 0.04(0.05)
IOA 0.93(0.81) 0.93(0.90) 0.94(0.87) 0.92(0.89) 0.95(0.94) 0.94(0.92)
2 m relative humidity (%) MB 1.10(12.02) 5.75(9.49) 2.40(11.31) 1.72(7.29) 4.56(7.87) 0.18(7.47)
RMSE 11.50(16.44) 13.24(15.30) 12.01(16.13) 9.17(11.39) 9.96(11.90) 9.41(11.55)
FB 0.02(0.18) 0.08(0.14) 0.03(0.17) 0.04(0.09) 0.06(0.11) 0.01(0.10)
FE 0.16(0.22) 0.17(0.20) 0.16(0.21) 0.11(0.13) 0.12(0.14) 0.11(0.13)
IOA 0.86(0.76) 0.80(0.74) 0.84(0.76) 0.88(0.83) 0.88(0.83) 0.88(0.83)
10 m wind speed (ms
1
) MB 0.03(0.07) 0.15(0.17) 0.09(0.05) 0.05(0.08) 0.14(0.15) 0.02(0.05)
RMSE 1.21(1.36) 1.33(1.40) 1.27(1.38) 1.11(1.20) 1.21(1.22) 1.16(1.21)
FB 0.02(0.14) 0.03(0.02) 0.01(0.06) 0.04(0.09) 0.08(0.09) 0.01(0.02)
FE 0.53(0.61) 0.61(0.63) 0.57(0.62) 0.56(0.60) 0.63(0.63) 0.59(0.61)
IOA 0.58(0.57) 0.59(0.58) 0.58(0.58) 0.54(0.51) 0.51(0.50) 0.53(0.51)
Wind direction (degree) MB 4.98(3.39) 7.27(6.87) 6.25(5.31) 6.26(11.67) 3.45(8.56) 4.72(9.96)
RMSE 79.89(80.41) 87.04(88.54) 83.91(85.00) 85.57(88.21) 91.17(91.95) 88.69(90.28)
Journal of Geophysical Research: Atmospheres 10.1002/2014JD021871
LI ET AL. ©2014. American Geophysical Union. All Rights Reserved. 6332
In terms of the 2 m relative humidity, the default data set has high biases in urban soil moisture and LAI,
which result in a high evaporation bias (7.29 to 13.17%) and RMSE (11.39 to 16.44%) at the urban sites. After
the implementation of the new land surface parameters in the WRF-modified experiment, the model
performance significantly improves over the urban regions in terms of both mean bias (3.81 to 3.61%) and
RMSE (8.99 to 13.08%). A strong moist bias of 6.39–13.81% is also observed over the rural areas in the default
case, which is reduced to 3.97–10.29% in the modified simulation. Overall, the modified case yields a domain-
average humidity bias of 2.10 to 5.73% and a root-mean-square error of 8.73 to 11.90%.
Wind speed and wind direction at the surface are the most variable and difficult parameters to simulate.
Generally, overestimation of wind speed is observed over the urban regions (0.55–2.64 m s
1
) and the rural
regions (0.60–2.68 m s
1
) in WRF default during most episodes. An obvious reduction of the wind speed error
by 0.020.95 m/s at the urban sites and 0.03–0.88 m/s at the rural sites is achieved by implementing the
newly developed land surface parameters. For the simulation of wind directions, a slightly refined simulation
is also found across the inner domain (Table 2).
Figures 6 and 7 give an example of the time series comparisons for the observations and simulation results
during two dry season episodes (4–8 October and 25–29 September) and the two wet season episodes
(19–24 July and 12–17 August). Time series are extracted for the above near-surface meteorological variables
at four representative observational sites (marked in Figure 1 with up triangles) that have changed from
cropland/grassland to metropolis: namely, HZ (23.08°N, 114.4°E), FS (22.85°N, 113.25°E), ZS (22.50°N, 113.35°E),
and DG (22.98°N, 113.98°E). The meteorological elements at all four stations are significantly influenced by land
surface changes, especially during the two dry season episodes that the model results show an obvious
improvement in the temperature and humidity variations. The temperature variation follows a strong diurnal
Table 2. (continued)
Episode 7 (7–10 November) Episode 8 (16–21 December)
Variable Index Urban Rural Total Urban Rural Total
2 m temperature (°C) MB 0.27(1.16) 0.11(0.27) 0.06(0.67) 0.09(1.86) 0.78(1.18) 0.41(1.48)
RMSE 1.39(1.87) 1.57(1.79) 1.49(1.82) 1.29(2.52) 2.11(2.47) 1.81(2.50)
FB 0.01(0.05) 0.00(0.01) 0.00(0.03) 0.00(0.16) 0.06(0.09) 0.03(0.12)
FE 0.04(0.07) 0.06(0.06) 0.05(0.06) 0.07(0.17) 0.16(0.18) 0.12(0.18)
IOA 0.95(0.93) 0.96(0.95) 0.96(0.94) 0.96(0.88) 0.93(0.91) 0.94(0.90)
2 m relative humidity (%) MB 0.57(7.63) 4.19(6.39) 0.72(7.29) 3.08(9.10) 3.97(7.67) 0.96(8.77)
RMSE 10.74(13.17) 10.99(12.39) 10.81(12.96) 13.08(13.99) 13.96(15.52) 13.58(14.63)
FB 0.02(0.12) 0.07(0.11) 0.00(0.12) 0.07(0.22) 0.13(0.21) 0.01(0.22)
FE 0.15(0.16) 0.15(0.17) 0.15(0.16) 0.26(0.28) 0.28(0.32) 0.27(0.29)
IOA 0.92(0.90) 0.92(0.90) 0.92(0.90) 0.58(0.68) 0.47(0.47) 0.56(0.64)
10 m wind speed (ms
1
) MB 0.38(0.77) 0.61(0.90) 0.49(0.83) 1.69(2.64) 1.80(2.68) 1.76(2.68)
RMSE 1.23(1.52) 1.39(1.68) 1.31(1.60) 2.51(3.44) 2.53(3.31) 2.52(3.38)
FB 0.19(0.34) 0.31(0.41) 0.25(0.37) 0.52(0.66) 0.50(0.64) 0.51(0.65)
FE 0.55(0.59) 0.62(0.66) 0.58(0.62) 0.61(0.71) 0.57(0.67) 0.59(0.69)
IOA 0.56(0.54) 0.56(0.53) 0.57(0.54) 0.46(0.43) 0.54(0.48) 0.50(0.45)
Wind direction (degree) MB 27.41(30.18) 18.76(19.80) 22.6(24.45) 19.11(17.68) 22.00(21.63) 20.87(20.06)
RMSE 78.68(78.63) 85.41(84.79) 82.47(82.09) 44.43 (43.71) 58.34(58.66) 52.97(52.91)
a
MB ¼1
NX
N
i¼1
simiobsi
ðÞ:
b
RMSE ¼ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
X
N
i¼1
simiobsi
ðÞ
2=N
s:
c
FB ¼2ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
simiobsi
ðÞ=simiþobsi
ðÞ
p=N:
d
FE ¼ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
simiobsi
jj
=simiþobsi
ðÞ
2
q=N:
e
IOA ¼1NRMSE2
X
N
i¼1
obsiobs þ
jj
simiobs
2
;where the term sim and obs refer to the simulated and observed meteorological values, respectively; N
represents the number of data pairs. In this study, Nranges from 5472 to 8208 depending on the simulation period.
Journal of Geophysical Research: Atmospheres 10.1002/2014JD021871
LI ET AL. ©2014. American Geophysical Union. All Rights Reserved. 6333
Figure 6. Time series of observed and simulated 2 m temperature, 2 m relative humidity, 10 m wind speed, and wind direction during the two dry season episodes at
four representative stations: (a) HZ, (b) FS, (c) ZS, and (d) DG.
Journal of Geophysical Research: Atmospheres 10.1002/2014JD021871
LI ET AL. ©2014. American Geophysical Union. All Rights Reserved. 6334
Figure 7. Same as Figure 6 but for the two wet season episodes.
Journal of Geophysical Research: Atmospheres 10.1002/2014JD021871
LI ET AL. ©2014. American Geophysical Union. All Rights Reserved. 6335
pattern and is reasonably well simulated in both experiments. However, the default model tends to strongly
underpredict the afternoon and midnight minimum values by 2–4°Cinautumn(Figure6)and1–3°C in
summer (Figure 7). In contrast, the modified run is warmer in the near-surface layer and matches the
observed temperature very closely. The 2 m relative humidity follows a very regular pattern that is
opposite to that of temperature: the minimum occurs at midday (less than 60%). Systematically, positive
bias (10–20%) of the modeled humidity is noted in the default simulation, and the modified simulation
yields greatly reduced humidity bias and better uniformity; however, slight overestimations are still
apparent during the two summer episodes (Figure 7). The wind speed is characterized by large diurnal
fluctuations and generally nocturnal minimums in both simulations. The wind distributions are very similar,
and no systematic differences between the two simulations are identified, except that at the ZS and DG
sites the default simulation slightly overestimates the wind speed at the beginning of the September
episode (Figure 6). In terms of wind direction, both cases could successfully capture the diurnal and
seasonal wind shift during the modeling period.
The above analyses reveal that the assimilation of MODIS observations in WRF results in large differences
compared with AVHRR-based simulations and outperforms the default simulation in terms of statistical
results (Table 2) and temporal variations (Figures 6 and 7).
Figure 8. Simulated surface skin temperature (°C) for (a) WRF default and (b) WRF modified. (c) MODIS-observed land surface temperature (°C).
Figure 9. Simulated 2 m air temperature (°C) for (a) WRF default and (b) WRF modified.
Journal of Geophysical Research: Atmospheres 10.1002/2014JD021871
LI ET AL. ©2014. American Geophysical Union. All Rights Reserved. 6336
3.2. Urban Heat Island
The phenomenon of urban heat island is generated over the PRD region during all the eight episodes in the
modified simulation. Figures 8 and 9 presented the simulated urban heat island during 4–8 October in terms
of both surface skin temperature (TSK) (Figure 8) and near-surface air temperature (Figure 9).
As the urban size increases, the enhanced surface radiation absorption (Figures S1a and S1b in the supporting
information) due to the low albedo of the urban surface and suppressed evaporation over impervious building
materials (Figures S1e and S1f) lead to an elevated surface skin temperature (Figure 8b), which is approximately
3.6°C higher on average compared with the default case (Table 3). Over the rural regions, a significantly
negative correlation is observed between skin temperature and green vegetation coverage (Figure 3) due to
the synergistic cooling effects from shade and evaporation loss. The lowest skin temperature is observed
successively in forest, grassland, and cropland (Figure 8b). The MODIS sensors onboard the Terra and Aqua
satellites provide 1 km measurements of land surface temperature (LST) 4 times a day (each satellite passes over
once per night and day). The LST data are obtained using the generalized split-window algorithm and are
calibrated for cloud cover [Wan, 2003]. A further comparison with the simultaneously observed LST (Figure 8c)
reveals a high-temperature center (over 31°C) that is located near the urban core and a gradual decreasein the
rural temperature, which agrees well with the WRF-modified simulation (Figure 8b).
The distributions of 2 m air temperature in the two simulations also exhibit distinct patterns (Figure 9).
Compared with the original simulation (Figure 9a), urbanization causes a remarkablewarming effect (Figure 9b)
that averages 1.1°C in the daytime and 2.3°C in the nighttime for WRF modified (Table 3). This positive forcing is
associated with the enhanced surface heating via upward sensible heat flux from the warmer land surface
(Figures S1c and S1d) within cities during the day (117.0 W m
2
) and the massive release of daytime ground
heat storage (130.5 W m
2
) overnight [Oke, 1982]. In accord with the modeling of skin temperature, the newly
urbanized region also experiences extremely high air temperature (commonly over 27°C) and a sharp cooling
transition relative to the rural surroundings, which is generally referred to as the urban heat island
(UHI) (Figure 9b).
Table 3. Averages and Differences for Each Meteorological Variable in WRF Default and WRF Modified During
4–8October
Variable
a
WRF Default WRF Modified Difference
Urban
b
Rural
b
Urban Rural Urban Rural
All Day
GRDFLX 7.0 4.1 35.4 3.2 28.4 0.9
HFX 46.0 38.1 112.4 41.1 66.4 3.0
LH 93.7 107.0 0.7 116.0 93.0 9.0
TSK 27.6 25.7 31.2 25.3 3.6 0.3
T2 25.3 24.0 27.1 23.9 1.8 0.1
Q2 14.5 13.6 13.4 13.4 1.1 0.2
Daytime
c
GRDFLX 49.1 41.1 130.5 33.3 81.4 7.8
HFX 97.4 85.0 214.4 92.6 117.0 7.6
LH 180.9 207.6 0.4 227.7 180.5 20.1
TSK 33.1 31.0 36.7 30.9 3.6 0.1
T2 27.8 26.5 28.9 26.6 1.1 0.1
Q2 14.4 13.9 12.7 13.8 1.7 0.1
Nighttime
c
GRDFLX 35.1 32.9 59.8 27.0 24.7 5.9
HFX 5.4 8.9 10.3 10.4 15.7 1.5
LH 6.4 6.4 1.0 4.4 5.4 2.0
TSK 22.0 20.3 25.8 19.7 3.8 0.6
T2 22.9 21.5 25.2 21.2 2.3 0.3
Q2 14.6 13.3 14.1 12.9 0.5 0.4
a
GRDFLX: ground heat flux (W m
2
); HFX: sensible heat flux (W m
2
); LH: latent heat flux (W m
2
); TSK: surface skin
temperature (°C); T2: air temperature at 2 m (°C); Q2: water vapor mixing ratio at 2 m (gkg
1
).
b
The inner domain (excluding the ocean) is classified into urban and rural regions based on the MODIS land
cover map.
c
Daytime lasts from 0800 to 1900 (LT) and nighttime lasts from 2000 to 0700 (LT).
Journal of Geophysical Research: Atmospheres 10.1002/2014JD021871
LI ET AL. ©2014. American Geophysical Union. All Rights Reserved. 6337
To distinguish the isolated impacts of land cover and vegetation parameters in the characteristics of urban
heat island, another experiment (“WRF Land Cover (WRF LC)”) is also conducted. The experiment designs and
results of WRF LC are explained in the supporting information (Text S1). It is found that although land
cover plays a dominant role in the generation of urban heat island, the inclusion of satellite-derived GVF and
LAI further enhances the cooling and warming effects due to its important role in surface energy partition.
The formation of UHI is clearly the result of urban-rural energy balance differences at the land surface interface,
which are mainly induced by the different surface thermal properties and soil moisture availability. Figure S1,
respectively, presents the average modeled heat fluxes for the default and modified runs. The results reveal that
the inclusion of MODIS satellite retrievals causes significant differences in the surface energy partitioning and
successfully represents the distinctions of energy budgets between vegetation and bare soil (Figure S1). Over
the central cities, the dominant portionof radiative energy is realized as ground heat flux (35.4 W m
2
) (Figure
S1b) and sensible heat flux (112.4 W m
2
) (FigureS1d). Thus, more energy goes into heating thenear-surface air
and land rather than as latent heat to evaporate (Figure S1f). In contrast, over the rural areas, the latent heat
exchange (116.0 W m
2
(Table 3)) dominates due to the enhancement of the plant transpiration caused by
abundant soil water availability and high vegetative cover (Figure 3).
The urban heat island is present day and night. The magnitude of the UHI intensity, which is expressed as the
difference of domain-average urban-rural air temperature at 2 m, exhibits a marked diurnal variation as
shown in Figure 10. The UHI intensity ranges from 1.5 to 4.8°C (Figure 10c), depending on the variations of the
urban-rural energy balance difference (Figures 10a and 10b), which produce different rates of near-surface
warming and cooling. The urban surface bears a greater soil thermal conductivity and heat capacity than
vegetation [Ek et al., 2003]; thus, it is able to store more energy during the daytime (130.5 W m
2
) and
release heat faster during nocturnal hours (59.8 W m
2
) (Figure 10a) than rural areas (Figure 10b). At night,
when the daytime energy storage assumes a more important role in the total energy balance, the release of
stored heat is especially favorable for maintaining a warmer urban temperature. Thus, a generally amplified
nocturnal UHI (4.0°C), but a weaker daytime effect (2.3°C) (Table 3), is clearly observed attributed to the
diverging urban/rural cooling rates and the weak nocturnal turbulent mixing. The UHI intensity normally
024681012141618202224
400
300
200
100
0
-100
-200
-300
024681012141618202224
-100
0
100
200
300
400
024681012141618202224
18
20
22
24
26
28
30
32
Local time (hr)
Local time (hr)
(c)
(b)
Heat flux (W m-2)
Heat flux (W m-2)
Local time (hr)
Ground heat flux
Sensible heat flux
Latent heat flux
(a) Groundheat flux
Sensible heat flux
Latentheat flux
Urban temperature
Rural temperature
0
1
2
3
4
5
UHI intensity
Figure 10. Simulated diurnal variations of (a) urban surface heat flux (W m
2
), (b) rural surface heat flux (W m
2
), and
(c) urban heat island intensity (°C) in WRF modified.
Journal of Geophysical Research: Atmospheres 10.1002/2014JD021871
LI ET AL. ©2014. American Geophysical Union. All Rights Reserved. 6338
exhibits a sharp increase around sunset due to the stronger cooling rate in the rural area and then reaches its
maximum a few hours (typically 3–5 h) later (approximately 5°C ). Thereafter, the cooling rates in the urban and
rural areas become similar and the UHI intensity remains nearly constant at 4°C until around sunrise (at 6:00 A.M.).
As the mixed layer builds, a weak daytime UHI (about 1.5°C) is gradually established due to sensible heating.
The influence of urban surface heating extends into the overlying atmosphere and gradually diminishes with
height (Figure 11). The vertical development of urban heat island is generally opposite to that observed for the
near-surface heat island (Figure 10). The vertical extent of urban heat island is stronger during the daytime,
especially in the late afternoon. In the morning (1000 LT), the UHI is much shallower below approximately 600 m
(Figure 11a). By noon (1200 LT ), the influence of land cover on the boundary layer warming can reach as high
as over 1 km. With the growth of the boundary layer,the urban warming layer deepens further until late afternoon
(around 1800 LT) (Figures 11c and 11d), which is clearly evident below 1.5 km where the temperature is warmer
than the surrounding countryside by 1–2°C. Compared with the greater-than-1 km depth during the day, the
vertical extent of nocturnal UHI is mainly restricted to the layer below 200 m.
In addition to the changes of thermal structure, the regional hydrological cycle is also modified by land
surface characteristics. The total evapotranspiration is a sum of direct surface water evaporation via bare soil
layer and plant canopy transpiration, which are controlled by vegetation fraction, soil moisture, atmospheric
conditions of demand (solar radiation, air temperature, and vapor pressure deficit), and leaf area index [Chen
et al., 1996; Ek et al., 2003]. Well-established easterly winds frequently bring moist air from the ocean; thus,
high moisture (over 15 g kg
1
) exists along the southeastern coast and decreases progressively inland
(Figure 12a). Comparisons between the default and modified simulations indicate that the sharp reduction in
soil moisture, enhanced by the vegetation removal (Text S1), act to suppress the daytime urban hydrological
cycle and dry the urban atmosphere by as much as 1.7 g kg
1
(Table 3). This is consistent with the
considerable shift from latent heat to sensible heat fluxes in the urban energy budget (Figure S1). In the rural
regions, the MODIS forest classification features more efficient transpiration due to a higher vegetation
fraction (Figure 3) and stronger root uptake of soil water than that of the cropland in the default case; thus,
Figure 11. Cross sections of simulated UHI (°C) along the line AA′marked in Figure 9b at (a) 1000 LT, (b) 1200 LT, (c) 1400 LT, and (d) 1600 LT on 6 October in WRF
modified. The black-shaded areas represent the terrain; the red bar represents the extent of the urban area.
Journal of Geophysical Research: Atmospheres 10.1002/2014JD021871
LI ET AL. ©2014. American Geophysical Union. All Rights Reserved. 6339
the model tends to enhance local latent heat exchange and moisten the local climate by 1 gkg
1
in large
portions of the forested regions (Figure 12b). The role of vegetation parameters in near-surface humidity is
explained in the supporting information (Text S1).
3.3. Local Wind Circulation
The PRD region has a very complex wind regime that comprises apparent interactions among the monsoon, sea-
land breeze circulation, and urban heat island circulation. The initiation and development of the local thermal
circulation are mainly driven by the increase in the mesoscale baroclinicity that is associated with the thermal
structure (i.e., land-sea temperature contrast and UHI effects) [Bach, 1970; Bornstein and Johnson, 1977].
3.3.1. Urban Heat Island Circulation
The simulated horizontal wind field at 10 m and air temperature at 2 m over the inner domain at selected
times on 6 October are depicted in Figure 13. In the land, the flow patterns are modified according to the
distribution of the urban land cover type.
As describedin section 3.2, the UHI intensity is relatively weak around early morning (Figure 10), so the effect of
the urban heat island on local wind field is not evident by that time. Afterward, the UHI intensity gradually
increases (Figure 10). By noon (1200 LT), the wind speed around the urban region slowly increases with the
inflow of the cooler rural air (Figures 13a and 13b). The enhancement of the urban breeze is a result of the
increased mesoscale baroclinicity and decreased stability caused by urban-rural temperature difference.
Figures 14a and 14b present the cross sections of wind components at 1200 LT along the line AA' that
crosses the isolated urban centers (Foshan and Guangzhou). Associated with this urban warming zone are areas
of increased upward velocities (0.2m s
1
) (near 112.9°E–113.4°E). In the afternoon (1400 LT) (Figures 13c and
13d), the temperature over much of the land areas surrounding the urban centers increases, thereby enhancing
the urban/rural temperature contrast (Figure 10). This temperature gradient generates several obvious
convergent zones in Foshan, Shenzhen, and eastern Jiangmen in the WRF-modified simulation (Figure 13d).
In the late afternoon (1800 LT), when the urban heat island intensity significantly increases (over 4°C)
(Figure 10) and relatively low wind conditions exist, enhanced urban convergence is observed in Foshan and
northern Dongguan city (Figure 13f). The urban-rural horizontal wind speed is approximately 2 m s
1
faster
than the default case (Figure 13e). In the vertical direction, two well-defined closed urban heat island
circulations are activated (Figures 14c and 14d). The most prominent circulation zone corresponds to the
urban classification that is located over eastern Foshan City (112.9°E–113.4°E). Areas of enhanced thermal
updraft (up to 0.6 m s
1
) above the urban area, which are associated with the circulation zone, are observed.
The urban updraft eventually entrains air from the northeastern and southwestern rural regions, which
Figure 12. (a) Simulated 2 m water mixing ratio (g kg
1
) during the daytime for WRF default and (b) difference between WRF modified and WRF default.
Journal of Geophysical Research: Atmospheres 10.1002/2014JD021871
LI ET AL. ©2014. American Geophysical Union. All Rights Reserved. 6340
Figure 13. Simulated surface wind field at (a and b) 1200 LT, (c and d) 1400 LT, and (e and f) 1800 LT on 6 October for WRF default (Figures 13a, 13c, and 13e) and WRF
modified (Figures 13b, 13d, and 13f). The shaded areas represent the simulated 2 m air temperature (°C); the circles mark the surface convergent zones.
Journal of Geophysical Research: Atmospheres 10.1002/2014JD021871
LI ET AL. ©2014. American Geophysical Union. All Rights Reserved. 6341
results in a convergent inflow into the urban areas at the lower levels and a divergent flow at the upper level
(horizontal wind velocity is approximately 1–2ms
1
). Meanwhile, relatively weak vertical descending motion
(around 0.1 m s
1
) occurs in the surrounding rural areas to form a closed vertical circulation. The extension of
this circulation zone can reach up to 40 km in the horizontal dimension and 1.6 km in the vertical dimension.
As described above, as the UHI develops, the abrupt temperature and pressure gradient over the urban-rural
border produces sharp pulses of cooler country air that is oriented toward the warmer city to form several
surface urban convergent zones (Figure 13). Meanwhile, the strong urban updraft is common during the day
(from around 1000 LT to 2000 LT). Afterward, the urban upward flow and the UHI circulation begin to
dissipate, which are indicative of weak mixing in the stable boundary layer.
3.3.2. Effects on Sea/Land Breeze Circulation
The evolution of the sea/land breeze is modified by the development of the urban heat island. By noon
(1200 LT), the land surface grows slightly warmer than the ocean; hence, a sea breeze is dominant over most
of the coastal regions (Figures 13a and 13b). Both WRF experiments generate similar large-scale background
prevailing easterly winds over the ocean and depict the sea breeze dominating the southeastern coastal
region and, to some extent, the Pearl River Estuary. A distinct divergent zone develops over the mouth of
Pearl River Estuary, and the sea breeze penetrates into the coastal portions of Zhongshan and Shenzhen
cities. Nevertheless, the urban heat island intensity is relatively weak so that the sea breeze does not
penetrate farther inland due to the weak land surface forcing. In the western coast of Shenzhen city, where
there is large urban expansion and slightly higher temperature, a weak convergent zone is initiated that is
likely related to the combined effects of the UHI and coastal sea breeze (Figure 13b).
Another interesting feature is the increased wind velocity over the Pearl River Estuary. Figure 15 presents the
cross sections of wind components along 22.6°N, which crosses a typical coastal city (Shenzhen) under a
complex wind regime. At 1200 LST, both simulations capture the divergent zone and subsidence flow above
the Pearl River Estuary (113.5°E–113.9°E) up to 1 km. In the modified case, the opposite recirculation cells at
the surface over the Pearl River Estuary are not symmetric; the westerly sea breeze toward Shenzhen is
Figure 14. Cross sections of wind vectors along the line AA' marked in Figure 9b at (a and b) 1200 LTand (c and d) 1800 LT on 6 October for WRF default (Figures 14a
and 14c) and WRF modified (Figures 14b and 14d). The black-shaded areas represent the terrain; the red bar represents the extent of the urban area.
Journal of Geophysical Research: Atmospheres 10.1002/2014JD021871
LI ET AL. ©2014. American Geophysical Union. All Rights Reserved. 6342
stronger due to the increased land-sea temperature contrast at Shenzhen. Additionally, compared to the
control run, WRF modified produces a stronger buoyant flow near 140°E, which is induced by local heating
near Shenzhen, to form part of the closed sea/land breeze circulation (114.0°E–114.5°E).
At 1400 LT (Figures 13c and 13d), the temperature contrast between the land and sea is enhanced by the land
warming. This leads to a well-developed sea breeze that penetrates inland along the coast at quite a long
distance of 30 km. Furthermore, the enhanced land-sea temperature difference in the modified case caused
by the UHI accelerates the sea breeze by 2–3ms
1
(Figure 13d) when compared with the scenario without a
city (Figure 13c). Afterward, the sea breeze along South China is fully developed and lasts until the early
evening. The sea breeze over the Pearl River Estuary is continuously enhanced by the urban heat island effect.
During the nighttime and early morning, the land surface quickly loses heat and becomes cooler than the ocean.
The sea breeze circulation gradually dissipates and is replaced by a land breeze circulation. However, as a result
of urbanization, the surface UHI can still be quite significant during the evening (Figure 10), which leads to a
more complicated wind field over the PRD. For example, at 0400 LT, the land breeze is well established along
the coast (Figure 16) and the urban regions remain relatively warmer than the adjacent regions by 3–5°C
(Figure 16b). In the default simulation, a strong convergent zone forms over the Pearl River Estuary (Figure 16a).
However, the localized hot spots around Shenzhen that are created in the modified run destroy the surface
convergent center and weaken the land breeze directed from the western coast of Shenzhen (Figure 16b).
Figure 16. Simulated surface wind field at 0400 LT on 6 October for (a) WRF default and (b) WRF modified. The circles mark the surface divergent zones.
Figure 15. Cross sections of wind vectors along 22.6°N at 1200 LT on 6 October for (a) WRF default and (b) WRF modified. The black-shaded areas represent the
terrain; the red bar represents the extent of the urban area; the blue bar represents the extent of the ocean.
Journal of Geophysical Research: Atmospheres 10.1002/2014JD021871
LI ET AL. ©2014. American Geophysical Union. All Rights Reserved. 6343
4. Conclusions
Evidence from theoretical and observational studies indicates that changes in land surface parameters play
an important role in the weather system. The present work seeks to reduce the uncertainties and the
dependence on prescribed look-up tables for important surface parameters in the WRF model by using high-
resolution MODIS observations to specify the land cover type, green vegetation fraction, and leaf area index.
Eight episodes spanning July to December are selected for verification of model improvements. Comparisons of
all the eight cases against ground observations reveal improved simulations of near-surface meteorological
elements as a result of a superior representation of the land surface and spatial heterogeneity.
The 5 day simulation during 4–8 October is selected for further analysis of urban heat island effects and
thermal circulation. Noticeable differences in the surface heat flux, land/air temperature, moisture and wind
circulation are produced by the application of MODIS measurements. As the urban size increases, a typical
urban heat island is generated as the result of the urban-rural energy balance differences. The UHI is
characterized by higher skin temperature and 2 m air temperature, and drier air (1.1 g kg
1
). Additionally,
the local wind circulation is modified due to the complex interactions between the urban heat island and the
local sea-land breeze in the PRD. For example, the UHI effect significantly enhances the urban-rural
momentum exchange and initiates the UHI circulation. During the daytime, the sea breeze over the Pearl
River Estuary is strengthened due to the increased land-sea temperature contrast, whereas during the night
and early morning the UHI effect weakens the land breeze.
Some issues still exist in the mesoscale modeling. At present, although some complex urban canopy modules
have been developed for mesoscale applications, the detailed processes involved in urban climates are not
considered in this study because of the absence of relevant urban geometry and energy parameters (e.g.,
urban size, urban population, and anthropogenic heat). Molders [2001] suggests that subgrid inhomogeneity
has a significant impact on the Bowen ratio and soil moisture. Other studies [Grossman-Clarke et al., 2005;
Stefanov et al., 2001] use the Landsat thematic mapper data to describe the heterogeneity of the urban
surface and found a significant impact on the turbulent heat flux and boundary layer evolution, as well as
improved simulation of the diurnal temperature cycle in the urban area. These studies highlight a new use for
advanced urban modules.
Further use can be made of other existing satellite data sets. For example, surface emissivity, deep soil
temperature, and albedo can be used to refine the land surface model input and to provide necessary
constraints for the model [de Foy et al., 2006]. The specification of tabulated soil moisture values may cause a
large uncertainty in the latent heat flux and moisture estimation, which should be further examined by
additional simulations and validations. The relationship between vegetation fraction and surface radiant
temperature has been used to estimate soil moisture availability from satellite observations [Carlson et al., 1990;
Owen et al., 1998], which has the potential to refine the analysis of the surface latent heat budget in the model.
Recently, numerous studies have focused on the feedback between regional air quality and meteorological
variability that is caused by the alterations of land surface characteristics [Avise et al., 2009; Jiang et al., 2008;
Lo et al., 2006; Yu et al., 2012]. Yu et al. [2012] noted a distinct increase (up to 20 ppb) in daytime ozone
concentrations in the Beijing-Tianjin-Hebei region and a 5 ppb increase in the Yangtze River Delta region in
China due to urbanization. Lo et al. [2006] found that urbanization enhances cross-city air pollutant transport
and accumulation via the coastal and urban land-sea breeze circulation. To mitigate the rapidly deteriorating
air quality in the PRD, it is critical to determine the effects of land cover changes on regional air quality,
especially the complex interactions between the local land-sea breeze circulation and the urban
environment. In addition to an improved understanding of the role of land surface parameters in regional
meteorology, this work also sets the stage for further investigation of the interactions between the
atmospheric physical and chemical processes in the PRD region.
References
Avise, J., J. Chen, B. Lamb, C. Wiedinmyer, A. Guenther, E. Salathe, and C. Mass (2009), Attribution of projected changes in summertime US
ozone and PM2.5 concentrations to global changes, Atmos. Chem. Phys.,9(4), 1111–1124.
Bach, W. (1970), An urban circulation model, Arch. Met. Geoph. Biokl. Ser. B,18, 155–168.
Bartholome, E., and A. S. Belward (2005), GLC2000: A new approach to global land cover mapping from Earth obser vation data, Int. J. Remote
Sens.,26(9), 1959–1977, doi:10.1080/01431160412331291297.
Acknowledgments
This study is supported by National
Natural Science Foundation of China
(41275155 and 41121004), the Public
Welfare Project for Environmental
Protection (201309009), the project
2010CB428501 supported by Chinese
Ministry of Science and Technology, and
the special fund of State Key Joint
Laboratory of Environment Simulation
and Pollution Control (14Y01ESPCP).
The MODIS land products are provided
by Land Process Distributed Active
Archive Center (LP-DAAC).
Journal of Geophysical Research: Atmospheres 10.1002/2014JD021871
LI ET AL. ©2014. American Geophysical Union. All Rights Reserved. 6344
Betts, A. K., F. Chen, K. E. Mitchell, and Z. I. Janjic (1997), Assessment of the land surface and boundary layer models in two operational
versions of the NCEP Eta Model using FIFE data, Mon. Weather Rev.,125(11), 2896–2916, doi:10.1175/1520-0493(1997)125<2896:
Aotlsa>2.0.Co;2.
Bornstein, R. D., and D. S. Johnson (1977), Urban-rural wind velocity differences, Atmos. Environ.,11(7), 597–604, doi:10.1016/0004-6981(77)
90112-3.
Brovkin, V., S. Sitch, W. von Bloh, M. Claussen, E. Bauer, and W. Cramer (2004), Role of land cover changes for atmospheric CO
2
increase and
climate change during the last 150 years, Global Change Biol.,10(8), 1253–1266, doi:10.1111/j.1365-2486.2004.00812.x.
Carlson, T. N., and D. A. Ripley (1997), On the relation between NDVI, fractional vegetation cover, and leaf area index, Remote Sens. Environ.,
62(3), 241–252, doi:10.1016/S0034-4257(97)00104-1.
Carlson, T. N., E. M. Perry, and T. J. Schmugge (1990), Remote estimation of soil-moisture availability and fractional vegetation cover for
agricultural fields, Agr. Forest. Meteorol.,52(1-2), 45–69, doi:10.1016/0168-1923(90)90100-K.
Chen, F., and J. Dudhia (2001), Coupling an advanced land surface-hydrology model with the Penn State -NCAR MM5 modeling system. Part I:
Model implementation and sensitivity, Mon. Weather Rev.,129(4), 569–585, doi:10.1175/1520-0493(2001)129<0569:Caalsh>2.0.Co;2.
Chen, F., K. Mitchell, J. Schaake, Y. K. Xue, H. L. Pan, V. Koren, Q. Y. Duan, M. Ek, and A. Betts (1996), Modeling of land surface evaporation by
four schemes and comparison with FIFE observations, J. Geophys. Res.,101(D3), 7251–7268, doi:10.1029/95JD02165.
Chen, J. L., Y. D. Du, and W. G. Sun (2013), Impact of urbanization on air temperature change in Pearl River Delta, Adv. Clim. Change Res.,9(2),
123–131.
Chen, X. L., H. M. Zhao, P. X. Li, and Z. Y. Yin (2006), Remote sensing image-based analysis of the relationship between urban heat island and
land use/cover changes, Remote Sens. Environ.,104(2), 133–146, doi:10.1016/j.rse.2005.11.016.
Chou, M. D., and M. J. Suarez (1999), A shortwave radiation parameterization for atmospheric studies, NASA/TM-104 606, vol. 15, 40 pp.
Crawford, T. M., D. J. Stensrud, F. Mora, J. W. Merchant, and P. J. Wetzel (2001), Value of incorporating satellite-derived land cover data in MM5/
PLACE for simulating surface temperatures, J. Hydrometeorol.,2(5), 453–468, doi:10.1175/1525-7541(2001)002<0453:Voisdl>2.0.Co;2.
de Foy, B., L. T. Molina, and M. J. Molina (2006), Satellite -derived land surface parameters for mesoscale modelling of the Mexico City basin,
Atmos. Chem. Phys.,6, 1315–1330.
Ek, M. B., K. E. Mitchell, Y. Lin, E. Rogers, P. Grunmann, V. Koren, G. Gayno, and J. D. Tarpley (2003), Implementation of Noah land surface model
advances in the National Centers for Environmental Prediction operational mesoscale Eta model, J. Geophys. Res.,108(D22), 8851,
doi:10.1029/2002JD003296.
Fan, S. J., Q. Fan, W. Yu, X. Y. Luo, B. M. Wang, L. L. Song, and K. L. Leong (2011), Atmospheric boundary layer characteristics over the Pearl
River Delta, China, during the summer of 2006: Measurement and mode l results, Atmos. Chem. Phys.,11(13), 6297–6310, doi:10.5194/acp-
11-6297-2011.
Findell, K. L., E. Shevliakova, P. C. D. Milly, and R. J. Stouffer (2007), Modeled impact of anthropogenic land cover change on climate, J. Clim.,
20(14), 3621–3634, doi:10.1175/Jcli4185.1.
Friedl, M. A., et al. (2002), Global land cover mapping from MODIS: Algorithms and early results, Remote Sens. Environ.,83(1-2), 287–302,
doi:10.1016/S0034-4257(02)00078-0.
Frolking, S., X. M. Xiao, Y. H. Zhuang, W. Salas, and C. S. Li (1999), Agricultural land-use in China: A comparison of area estimates from ground-
based census and satellite-borne remote sensing, Global Ecol. Biogeogr.,8(5), 407–416, doi:10.1046/j.1365-2699.1999.00157.x.
Gallus, W. A., and J. F. Bresch (2006), Comparison of impacts of WRF dynamic core, physics package, and initial conditions on warm season
rainfall forecasts, Mon. Weather Rev.,134(9), 2632–2641, doi:10.1175/Mwr3198.1.
Grossman-Clarke, S., J. A. Zehnder, W. L. Stefanov, Y. B. Liu, and M. A. Zoldak (2005), Urban modifications in a mesoscale meteorological model
and the effects on near-surface variables in an arid metropolitan region, J. Appl. Meteorol.,44(9), 1281–1297, doi:10. 1175/Jam2286.1.
Gutman, G., and A. Ignatov (1998), The derivation of the green vegetation fraction from NOAA/AVHRR data for use in numerical weather
prediction models, Int. J. Remote Sens.,19(8), 1533–1543, doi:10.1080/014311698215333.
Houghton, R. A., and J. L. Hackler (2003), Sources and sinks of carbon from land-use change in China, Global Biogeochem. Cycles,17(2), 1034,
doi:10.1029/2002GB001970.
Jacquemin, B., and J. Noilhan (1990), Sensitivity study and validation of a land surface parameterization using the Hapex-Mobilhy data set,
Bound-Lay Meteorol,52(1-2), 93–134, doi:10.1007/Bf00123180.
Jarvis, P. G. (1976), Interpretation of variati ons in leaf water potential and stomatal conductance found in canopies in field, PhilosTrans. R.Soc.
B,273(927), 593–610, doi:10.1098/rstb.1976.0035.
Jiang, X. Y., C. Wiedinmyer, F. Chen, Z. L. Yang, and J. C. F. Lo (2008), Predicted impacts of climate and land use change on surface ozone in the
Houston, Texas, area, J. Geophys. Res.,113, D20312, doi;10.1029/2008JD009820.
Justice, C. O., J. R. G. Townshend, E. F. Vermote, E. Masuoka, R. E. Wolfe, N. Saleous, D. P. Roy, and J. T. Morisette (2002), An overview of MODIS
Land data processing and product status, Remote Sens. Environ.,83(1-2), 3–15, doi:10.1016/S0034-4257(02)00084-6.
Kain, J. S., and J. M. Fritsch (1992), The role of the convective trigger function in numerical forecasts of mesoscale convective systems,
Meteorol. Atmos. Phys.,49(1–4), 93–106, doi:10.1007/Bf01025402.
Knyazikhin, Y., J. V. Martonchik, R. B. Myneni, D. J. Diner, and S. W. Running (1998), Synergistic algorithm for estimating vegetation canopy leaf
area index and fraction of absorbed photosynthetically active radiation from MODIS and MISR data, J. Geophys. Res.,103(D24),
32,257–32,275, doi:10.1029/98JD02462.
Lambin, E. F., H. J. Geist, and E. Lepers (2003), Dynamics of land-use and land-cover change in tropical regions, Annu. Rev. Environ. Resour,28,
205–241, doi:10.1146/annurev.energy.28.050302.105459.
Lin, W. S., C. H. Sui, L. M. Yang, X. M. Wang, R. R. Deng, S. J. Fani, C. S. Wu, A. Y. Wang, S. K. Fong, and H. Lin (2007), A numerical study of the
influence of urban expansion on monthly climate in dry autumn over the Pearl River Delta, China, Theor. Appl. Climatol.,89(1-2), 63–72,
doi:10.1007/s00704-006-0244-6.
Lin, Y. L., R. D. Farley, and H. D. Orville (1983), Bulk parameterization of the snow field in a cloud model, J. Climate Appl. Meteorol.,22(6),
1065–1092, doi:10.1175/1520-0450(1983)022<1065:Bpotsf>2.0.Co;2.
Liu, J. Y., H. Q. Tian, M. L. Liu, D. F. Zhuang, J. M. Melillo, and Z. X. Zhang (2005), China’s changing landscape during the 1990s: Large-scale land
transformations estimated with satellite data, Geophys. Res. Lett.,32, L02405, doi 10.1029/2004GL021649.
Liu, M. L., and H. Q. Tian (2010), China’s land cover and land use change from 1700 to 2005: Estimations from high-resolution satellite data
and historical archives, Global Biogeochem. Cycles,24, GB3003, doi:10.1029/2009GB003687.
Lo, J. C. F., A. K. H. Lau, J. C. H. Fung, and F. Chen (2006), Investigation of enhanced cross-city transport and trapping of air pollutants by
coastal and urban land-sea breeze circulations, J. Geophys. Res.,111, D14104, doi 10.1029/2005JD006837.
Lo, J. C. F., A. K. H. Lau, F. Chen, J. C. H. Fung, and K. K. M. Leung (2007), Urban modification in a mesoscale model and the effects on the local
circulation in the Pearl River Delta region, J. Appl. Meteorol. Climatol.,46(4), 457–476, doi:10.1175/Jam2477.1.
Journal of Geophysical Research: Atmospheres 10.1002/2014JD021871
LI ET AL. ©2014. American Geophysical Union. All Rights Reserved. 6345
Loveland, T. R., B. C. Reed, J. F. Brown, D. O. Ohlen, Z. Zhu, L. Yang, and J. W. Merchant (2000), Development of a global land cover charac-
teristics database and IGBP DISCover from 1 km AVHRR data, Int. J. Remote Sens.,21(6-7), 1303–1330, doi:10.1080/014311600210191.
Meng, X. H., S. H. Lu, T. T. Zhang, J. X. Guo, Y. H. Gao, Y. Bao, L. J. Wen, S. Q. Luo, and Y. P. Liu (2009), Numerical simulations of the atmospheric
and land conditions over the Jinta oasis in northwestern China with satellite-derived land surface parameters, J. Geophys. Res.,114,
D06114, doi 10.1029/2008JD010360.
Molders, N. (2001), On the uncertainty in mesoscale modeling caused by surface parameters, Meteorol. Atmos. Phys.,76(1-2), 119–141,
doi:10.1007/s007030170043.
Monin, A. S., and A. M. Obukhov (1954), Basic laws of turbulent mixing in the surface layer of the atmosphere, Tr. Akad. Nauk SSSR Geophiz.
Inst.,24(151), 163–187.
Noh, Y., W. G. Cheon, S. Y. Hong, and S. Raasch (2003), Improvement of the K-profile model for the planetary boundary layer based on large
eddy simulation data, Boundary Layer Meteorol.,107(2), 401–427, doi:10.1023/A:1022146015946.
Oke, T. R. (1982), The energetic basis of the urban heat island, Q. J. R. Meteorol. Soc.,108(455), 1–24, doi:10.1002/qj.49710 845502.
Owen, T. W., T. N. Carlson, and R. R. Gillies (1998), An assessment of satellite remotely-sensed land cover parameters in quantitatively
describing the climatic effect of urbanization, Int. J. Remote Sens.,19(9), 1663–1681, doi:10.108 0/014311698215171.
Pielke, R. A. (2001), Influence of the spatial distribution of vegetation and soils on the prediction of cumulus convective rainfall, Rev. Geophys.,
39(2), 151–177, doi:10.1029/1999RG000072.
Pielke, R. A., and M. Uliasz (1998), Use of meteorological models as input to regional and mesoscale air quality models—Limitations and
strengths, Atmos. Environ.,32(8), 1455–1466, doi:10.1016/S1352-2310(97)00140-4.
Purevdorj, T., R. Tateishi, T. Ishiyama, and Y. Honda (1998), Relationships between percent vegetation cover and vegetation indices, Int. J.
Remote Sens.,19(18), 3519–3535, doi:10.1080/014311698213795.
Seto, K. C., C. E. Woodcock, C. Song, X. Huang, J. Lu, and R. K. Kaufmann (2002), Monitoring land-use change in the Pearl River Delta using
Landsat TM, Int. J. Remote Sens.,23(10), 1985–2004, doi:10.1080/01431160110075532.
Skamarock, W. C., J. B. Klemp, J. Dudhia, D. O. Gill, D. M. Barker, M. G. Duha, X.-Y. Wang, W. Wang, and J. G. Powers (2008), A description of the
advanced research WRF version 3, Tech. Note NCAR/TN-475+STR, Natl. Cent. for Atmos. Res., Boulder, Colo.
Stefanov, W. L., M. S. Ramsey, and P. R. Christensen (2001), Monitoring urban land cover change: An expert system approach to land cover
classification of semiarid to arid urban centers, Remote Sens. Environ.,77(2), 173–185, doi:10.1016/S0034-4257(01)00204-8.
Taha, H. (1996), Modeling impacts of increased urban vegetation on ozone air quality in the South Coast Air Basin, Atmos. Environ.,30(20),
3423–3430, doi:10.1016/1352-2310(96)00035-0.
van der Gon, H. D. (2000), Changes in CH
4
emission from rice fields from 1960 to 1990s—1. Impacts of modern rice technology, Globa l
Biogeochem. Cycle,14(1), 61–72.
Wan, Z. M. (2003), Monitoring thermal status of ecosystems with MODIS land-surface temperature and vegetation index produc ts,
Remote Sens. Agric. Ecosyst. Hydrol. IV,4879, 280–288, doi:10.1117/12.462411.
Wang, W. C., Z. Zeng, and T. R. Karl (1990), Urban heat islands in China, Geophys. Res. Lett.,17(12), 2377–2380, doi:10.1029/GL017i013p02377.
Wang, X. M., W. S. Lin, L. M. Yang, R. R. Deng, and H. Lin (2007), A numerical study of influences of urban land-use change on ozone distri-
bution over the Pearl River Delta region, China, Tellus B,59(3), 633–641, doi:10.1111/j.1 600-0889.2007.00271.x.
Wang, X. M., Z. Y. Wu, and G. X. Liang (2009), WRF/CHEM modeling of impacts of weather conditions modified by Urban expansion on
secondary organic aerosol formation over Pearl River Delta, Particuology,7(5), 384–391, doi:10.1016/j.partic.2009.04.007.
Weng, Q. H. (2002), Land use change analysis in the Zhujiang Delta of China using satellite remote sensing, GIS and stochastic modelling, J.
Environ. Manage.,64(3), 273–284, doi:10.1006/jema.2001.0509.
Weng, Q. H., and S. H. Yang (2004), Managing the adverse thermal effects of urban development in a densely populated Chinese city, J.
Environ. Manage.,70(2), 145–156, doi:10.1016/j.jenvman.2003.11.006.
Wetzel, P. J., and J. T. Chang (1988), Evapotranspiration from nonuniform sur faces—A 1st approach for short-term numerical weather
prediction, Mon. Weather Rev.,116(3), 600–621, doi:10.1175/1520-0493(1988)116<0600:Efnsaf>2.0.Co;2.
Yu, M., G. R. Carmichael, T. Zhu, and Y. F. Cheng (2012), Sensitivity of predicted pollutant levels to urbanization in China, Atmos. Environ.,60,
544–554, doi:10.1016/j.atmosenv.2012.06.075.
Yucel, I. (2006), Effects of implementing MODIS land cover and albedo in MM5 at two contrasting US regions, J. Hydrometeorol.,7(5),
1043–1060, doi:10.1175/Jhm536.1.
Journal of Geophysical Research: Atmospheres 10.1002/2014JD021871
LI ET AL. ©2014. American Geophysical Union. All Rights Reserved. 6346
