Effects of Updated RegCM4 Land Use Data on Near-Surface Temperature Simulation in China

Yulong REN, Yaohui LI*, Zhaoxia PU, Tiejun ZHANG, Haixia DUAN, and Wei WANG

1 Institute of Arid Meteorology of China Meteorological Administration (CMA), Key Laboratory of Arid Climatic Change and Disaster Reduction of Gansu Province and CMA, Lanzhou 730020, China

2 Department of Atmospheric Sciences, University of Utah, Utah 84112, USA

ABSTRACT Biogeophysical effects of land use and land cover (LULC) changes play a significant role in modulating climate on various spatial scales. In this study, a set of recent LULC products with a spatial resolution of 500 m was developed in China for update in RegCM4 (regional climate model version 4). Two sets of comparative numerical experiments were conducted to study the effects of LULC changes on near-surface temperature simulation. The results show that after LULC changes, areas of crops and mixed woodlands as well as urban areas increase over entire China, accompanied with greatly expanded mixed farming and forests/field mosaics in southern China, and reduced areas of 1) irrigated crops and short grasses in northern China and the Tibetan Plateau, and 2) semi-desert and desert in northwestern China. Improvements in the LULC data clearly result in more accurate simulations of the near-surface temperature. Specifically, increasing latent heat and longwave albedo due to enhanced LULC in certain areas lead to reduction in land surface temperature (LST), while changes in shortwave albedo and sensible heat also exert a great influence on the LST. Overall, these parameter adjustments reduce the biases in near-surface temperature simulation.

Key words: land use and land cover, RegCM4 (regional climate model version 4), temperature simulation

1. Introduction

Near-surface temperature is an important factor in climate change studies. In recent years, aridification caused by increased near-surface temperature has had a significant effect on ecosystems, the hydrological cycle, and agricultural production (Tang et al., 2016; Guan et al.,2017; Ren et al., 2017). Previous studies have shown that the regions with significant temperature increases in the last 130 years are located in mid–high-latitude zones in Asia (Stocker et al., 2013), including northeastern China,northern China, and the Tibetan Plateau (Wang et al.,2012; Liu et al., 2016; Guan et al., 2017; Yuan et al.,2017). Climate change has been linked to an increasing number of significant drought events, and the risks associated with drought continue to intensify in China (Dai,2011; Yang et al., 2011; Chen et al., 2015), resulting in increased long-term effects on socioeconomics and agricultural production (Zhang et al., 2012, 2014; Li and Li,2017; Li and Lyu, 2017; Wang et al., 2017; Zhang et al.,2017). Thus, increased attention is being paid to temperature change, and its simulation has become one of the main indices by which to measure the performance of regional climate models.

The simulation results of near-surface temperatures are closely related to the land-surface parameters of land surface models (LSMs). Fang et al. (2010) optimized the albedo, length of roughness, and soil thermal properties in an LSM, CoLM (Common Land Model), and found that although surface temperature is sensitive to albedo throughout the year, it is most sensitive to albedo during the spring and summer. Reijmer et al. (2004) studied the effects of the length of roughness on climate in the Antarctic and showed that decreased length of roughness can result in increased near-surface wind speed and in turn,decrease near-surface temperatures with increased atmospheric temperature. Using the NOAH LSM, He et al.(2014) compared the effects of land-surface data with varying resolution on the winter near-surface temperature modeling results for Lanzhou and found that the results were quite sensitive to the resolution of the land-surface data used. Moreover, Yu and Xie (2013) studied the effects of variation in regional land cover on regional climate and concluded that variation in regional LULC(land use and land cover) can change the land-surface energy, water balance, and large-scale circulation of an area, thus significantly affecting near-surface temperatures. Furthermore, Zhang et al. (2009) conducted sensitivity experiments using two sets of land-surface data and showed that variations in land-surface parameters can result in significant changes in precipitation and temperature; in addition, the variation in land-surface parameters is a potential driver of prolonged drought in midwestern China.

RegCM is a regional climate model that evolved out of the expansion and modification of the radiation scheme, convection parameterization scheme, and physical land-surface processes within the mesoscale model MM4, originally developed by Dickinson et al. (1989)and Giorgi and Bates (1989). Later, Giorgi et al. (1993a,b) further improved the schemes for the associated physical processes of the mesoscale model and produced RegCM2 and RegCM3. RegCM4 is the most recent version of the RegCM series and has been widely applied to studies of regional climate throughout the years. In RegCM4, the default land-surface model is the biosphere–atmosphere transfer scheme (BATS), which describes the momentum, energy, and water vapor exchange between vegetation, the land surface, and the atmosphere (Dickinson et al., 1986, 1993). Similar to other LSMs, LULC change greatly affects the ability of BATS to simulate temperature. In the most recent BATS version (Giorgi et al., 2003), i.e., BATS1e, 21 different types of vegetation are considered, and the model calculates variations in topology and land-surface type on a sub-grid scale using the mosaic method (i.e., by adopting regular small-scale, table-level grids for each coarse grid). Relevant studies have shown that compared to more complicated LSMs, the land-surface processes in BATS are physically well defined, and running them occupies a relatively small proportion of computing resources. This results in fast computing speeds that support research based on climate modeling and simulations(Meng and Fu, 2009; Yang et al., 2016). However, the default LULC data in BATS1e are based on the USGS(Geological Survey) GLCC (Global Land Cover Characteristics Database), which is relatively out of date. Previous studies have shown that the accuracy of the data in China is lower than the global average; therefore, large uncertainties exist in the data that affect the model’s ability to accurately simulate climate in China (Gong, 2009;Ran et al., 2010). Thus, it is necessary to update the land use data to improve the ability of RegCM4 to simulate temperatures in China.

In this study, we improve the land use data in BATS1e by creating a new set of regional LULC products at a spatial resolution of 500 m in China that can optimize the land-surface parameters in RegCM4, which is of paramount importance if we are to improve the model’s ability to simulate near-surface temperature in China.

2. Data and methods

2.1 LULC data

The default LULC data in BATS1e are the USGS GLCC. To produce more accurate and precise LULC data for China, 1:100,000 land survey data for China in 2000 and 2005 were used to assess the classification accuracy of four LULC products in terms of type, area, and spatial consistency. The four widely used LULC products are (1) the Global Land Cover Dataset (IGBP DISCover), which was developed by the USGS for the International Geosphere–Biosphere Program; (2) the Global Land Cover Dataset of the University of Maryland (UMD); (3) the Global Land Cover Data Products in 2000 and 2005 (GLC2000/05) of the European Union Joint Research Center (JRC) Institute of Space Applications (SAI); and (4) the MODIS Land Cover Data products (MOD12Q1) in 2000 and 2005. Evaluation results show that the MODIS and IGBP land cover data have the highest classification accuracy (78.34%) among the four products. On this basis, by using spatial analysis and the discrimination algorithm module of the ArcGIS software, the two sets of high-precision data were merged. Then, MODIS water-body masks and other related products were used to classify the fused land classification data for category fusion and judgment. Finally,according to the LULC classification in BATS1e (Table 1),a set of LULC products with a spatial resolution of 500 m in China was created. Before conducting the simulation test, the percentage of different land cover types for each model grid was calculated, and then, the vegetation typewith the largest percentage was selected as the LULC type of the grid, finally forming the LULC data that could be read and written by RegCM4 directly.

Table 1. LULC classification in the land surface model BATS1e

Figures 1a and 1b compare the spatial distributions of the regional LULC in China before and after the update.To analyze the variation in characteristics of different regions, we defined four main regions (Fig. 1c) in China according to the related literature (Huang, 1989) as follows: (1) northern China, which comprises the northern portion of the monsoon climate zone; (2) southern China,which is located south of Qinling–Huaihe River, east of the Tibetan Plateau and adjacent to the east, and the South China Sea; (3) northwestern China, which is located primarily west of the Greater Khingan Range, north of the Great Wall, and the Kunlun Mountains–A’erjin Mountains; and (4) the Tibetan Plateau, which includes the Qinghai–Tibetan Plateau, with the highest altitude in the world, and the transitional area from the Tibetan Plateau to the Loess Plateau, which is the largest area in China influenced by complex terrain.

Fig. 1. Spatial distributions of land-surface types (a) before and (b) after the land use data update, and (c) locations of the four natural regions in China.

Fig. 2. LULC type variations after updating the LULC data in Northwest China, North China, South China, and the Tibetan Plateau.

Figure 2 shows the variations in land-surface type after the LULCC in the four regions. The northwestern region is characterized by a relatively large decrease in semi-deserts, short grasses, and deserts, which declined by 11.1%, 6.5%, and 4.1%, respectively. In comparison,crops and mixed woodland increased by 6.4% and 8.9%,respectively, and urban areas increased from 0% to 0.4%.In the northern region, crops/mixed farming and mixed woodland increased by 11.7% and 11.2%, respectively,while short grasses and irrigated crops decreased by 11%and 5.8%, respectively; in addition, urban areas increased from 0% to 1.1%, which is considered a rapid rate. In the southern region, mixed woodland, forest/field mosaic, and crops/mixed farming increased by 20.5%,12.2%, and 15.6%, respectively; these were the largest increases identified in this region. Urban areas also increased from 0% to 0.4%. The greatest change on the Tibetan Plateau was found for crops/mixed farming coverage, which increased by 2.6%, followed by evergreen needle-leaf forest (0.5%) and urban areas (0.5%). In contrast, the semi-desert, irrigated crop, and short-grass coverage decreased by 2.2%, 0.6%, and 0.9%, respectively.These results demonstrate that the updated land-surface data vary significantly compared to the original data, particularly in their representation of urban areas; the updated data show that urban areas cover a larger proportion of all areas compared to the original data, thus showing the increased effect of human activities on regional climate change and more accurately representing the actual situation.

2.2 CRU temperature data

The near-surface temperature data used in this study are derived from the Climatic Research Unit (CRU). The data were processed by thin plate smoothing interpolation splines at the Climate Research Department of the University of East Anglia, which interpolated the observed data from thousands of meteorological stations around the world to the corresponding longitudes and latitudes to obtain a temperature sequence that was further improved by a series of steps. Previous studies have shown that compared to similar data, the satellite-combined CRU data can be processed rapidly and have high spatial and temporal resolution. In addition, the temporal sequence of the CRU data is also relatively long, and the CRU temperature sequence is currently widely applied in global climate change research (Wen et al., 2006).

2.3 Experimental design

In this study, the central grid for the numerical simulation region is 37.39°N, 103.48°E, with a horizontal resolution of 40 km, resulting in a 150 × 130 cell grid along the east–west and south–north axes. For the horizontal level, Arakawa–Lamb B staggered grids, based on the Lambert projection, are used. The vertical level contains 18 inhomogeneous layers. A hybrid coordinate is used for these layers; the pressure at the top layer is 50 hPa.The MIT–Emanuel cumulus convection scheme was selected for cumulus parameterization, the Zeng scheme was selected for sea flux parameterization, BATS1e was selected for land-surface parameterization, and the NCAR CCM3 radiation transfer scheme was chosen. The lateral boundary condition was based on the NCEP reanalysis data (2.5° × 2.5°), which are updated every 6 h; the sea surface temperature (SST) data were based on the OSSST monthly average data from NOAA. Table 2 outlines the model schemes in detail.

To analyze the effect of improved LULC data on nearsurface temperature simulations, we designed two sets of experiments: Test 1 and Test 2. Test 1 uses the original LULC data in the model while Test 2 uses the updated LULC data.

The results from the two sets of experiments were integrated over 3 yr (January 2008 to December 2010).Considering the spin-up time in the model, we analyzed the simulation results from 2009 to 2010.

Table 2. Primary physical process schemes

3. Results

3.1 Effect of LULC changes on near-surface temperature simulation in China

Figure 3 shows the annual, winter, and summer temperature biases between the simulated near-surface temperatures generated by the two tests and the CRU temperature data. From Figs. 3a–f, we can see that in Test 1,the simulated average near-surface temperature is 0.5°C higher than the average CRU temperature in some local regions of northeastern China, southwestern China, the Tibetan Plateau, and southern China; however, it is generally lower than the average CRU temperature in most regions. In contrast, in Test 2, the total area of positive bias is clearly smaller due to the reduced positive bias values; in the other regions, the simulated temperatures increase, and thus, the bias values decrease.

Figures 4a–c compare the average regional near-surface temperatures obtained from CRU, Test 1, and Test 2. Figure 4a shows that the simulated annual temperatures from Tests 1 and 2 are lower than the actual values in northwestern China, southern China, and the Tibetan Plateau but are higher for regions in northern China. By analyzing the annual temperature bias, we observe that the average temperature bias from Test 2 is clearly smaller than that from Test 1; in Test 2, the average temperature bias in all regions except that the Tibetan Plateau is quite small because the simulated annual average temperatures are quite close to the actual CRU values. Figure 4b shows that the average winter temperatures simulated in Tests 1 and 2 are higher than the actual values for northwestern China, northern China, and southern China but are lower for the Tibetan Plateau. However, the simulated values from Test 2 are closer to the actual temperatures, with the smallest bias in the northwestern China and approximately equal biases in northern and southern China. Figure 4c further shows that the average summer temperatures simulated by Test 1 are higher than the actual values, while the average summer temperatures simulated by Test 2 are lower than the actual values for southern China. However, the biases for the average temperatures simulated by Test 2 are clearly smaller than those simulated by Test 1.

Table 3 shows that the temperature biases for data simulated in Test 1 reach –1.0°C (northwestern region),1.1°C (northern region), 1.5°C (southern region), and–1.2°C (plateau region). In contrast, the corresponding biases for data simulated in Test 2 are clearly smaller,with the exception of those simulated for the Tibetan Plateau, which have slightly larger biases (–0.2, 0.2,–0.1, and –0.9°C for northwestern, northern, southern,and plateau regions, respectively); the simulated average temperatures for northwestern, northern, and southern regions are quite close to the CRU values. In winter, the simulated temperatures for the Tibetan Plateau are smaller than the CRU values, while those for northwestern,northern, and southern regions are all greater than the CRU values. In addition, the biases illustrated by Test 2 are smaller than those by Test 1; the corresponding biases from the CRU values are 0.4 (Test 1), 0.1 (Test 2);1.7 (Test 1), 0.7 (Test 2); 2.7 (Test 1), 1.3 (Test 2); and–0.9 (Test 1), –0.4°C (Test 2) for northwestern, northern,southern, and plateau regions, respectively. The simulated summer temperatures in northwestern and northern regions are greater than the CRU values, while the simulated southern and plateau values are greater in Test 1 but smaller in Test 2. The respective temperature biases in northwestern, northern, southern, and plateau regions are 1.4 (Test 1), 0.3 (Test 2); 2.2 (Test 1), 0.4 (Test 2); 0.8(Test 1), –0.2 (Test 2); and 0.8 (Test 1), –0.1°C (Test 2).

3.2 Change in major land surface parameters

In BATS1e, the land surface parameters affecting land surface temperature (LST) are vegetation coverage,roughness, albedo, and stomatal resistance. After the LULC data were updated (Table 4), the vegetation coverage in the four regions increased to different degrees. In southern region, vegetation coverage increased more than in any of the other regions (by 0.24 m), and in northwestern and northern regions, it increased by 0.1 and 0.06 m,respectively, while the coverage in the plateau region did not change much. Corresponding to the increase in vegetation, surface roughness in southern region increased the most for all regions, by 0.26 m. Surface roughness in northwestern and northern regions increased by 0.14 and 0.12 m, respectively, while that in the plateau region changed only slightly. In northern and southern regions,vegetation albedo for wavelengths ＜ 0.7 μm (shortwave)decreased, and for wavelengths ＞ 0.7 μm (longwave) it increased. The albedo in the plateau region changed only slightly. In the south, stomatal resistance decreased more than in any other region, reaching 27.8 s m–1, followed by the northern and plateau regions. The plateau region had the smallest change of all regions, with a decrease of 4.38 s m–1.

3.3 Major influencing factors

LST is one of the principal factors that determine change in near-surface temperature. Variations in land surface parameters change the near-surface temperature by affecting the LST. To analyze the major factors influencing LST, the following LST change was derived from the surface radiation balance equation (Chen and Dirmeyer, 2016):

Fig. 4. Average simulation temperature (°C) in Tests 1 and 2 in northwestern, northern, and southern, and plateau regions for (a) winter, (b)summer, and (c) annual.

Table 3. Simulated annual, winter, and summer temperature biases (°C) for northwestern China, northern China, southern China, and the Tibetan Plateau for data simulated in Tests 1 and 2

Table 4. Variation of some land surface parameters affecting LST

In Eq. (1), Tsis the LST, SWinis incoming shortwave radiation, LWinis incoming longwave radiation, σ is the Stephan–Boltzmann constant (with a value of 5.67 × 10–8W m–2K–4), H is the sensible heat flux, LE is the latent heat flux, and G is the ground heat flux. On the left side of Eq. (1), △Tsis the LST change. SWin△αsis the surface albedo change; Eq. (1) indicates that when the surface albedo increases (decreases), the surface temperature decreases (increases). The item (1 – αs)△SWinrepresents the variation in incident shortwave radiation, and Eq. (1) shows that when the incident shortwave radiation increases (decreases), the surface temperature also increases (decreases); △LWinrepresents the variation in incident longwave radiation, which is proportional to the change in LST; △LE is the change in latent heat, △H is the change in sensible heat, and △G is the change in surface heat flux. These three terms are inversely proportional to the change in LST. Surface emissivity is ignored in Eq. (1). In this paper, the average annual winter and summer data of the model output were used to calculate the various factors of Eq. (1), and the principal factors causing the annual, summer, and winter temperature changes were analyzed.

The annual average change of each factor indicates(Table 5) that changes in LH reduced the LST in the northwestern, northern, southern, and plateau regions by 0.5, 1.2, 2.1, and 0.1°C, respectively, while the SH increased the LST of the four regions by 0.9, 0.2, 0.3, and 0.2°C. Additionally, incident shortwave radiation increased the LST by 1.0, 0.5, 0.2, and 0.3°C. The change in albedo increased the LST by 0.2, 0.2, 0.1, and 0.1°C.Incident shortwave changes increased the LST by 1.0,0.5, 0.2, and 0.3°C. Incident longwave radiation reduced the LST by 0.1, 0.6, 0.4, and 0.1°C. The comprehensive effect of these factors caused the LST in the four regions to change by 1.2, –0.9, –1.9, and 0.4°C, respectively.

Table 5. Annual, winter, and summer average △Ts contributions of each item (°C) in Eq. (1)

The winter average change in each factor indicates(Table 5) that variations in LH caused the LST in the northwestern, northern, southern, and plateau regions to decrease by 0.7, 0.3, 1.4, and 0.1°C, respectively, while the SH increased the LST values by 0.6, 0.6, 0.1, and 0.2°C. In addition, incident shortwave radiation increased the LST by 0.3, 0.7, 0.7, and 0.5°C. Albedo increased the LST by 0.1, 0.3, 0.7, and 0.5°C. Incident longwave radiation reduced the LST by 0.5, 0.3, 0.5, and 0.1°C. The comprehensive effect of various factors caused the LST in the four regions to change by –0.2,0.8, –1.0, and 0.6°C, respectively.

The summer average change in each factor indicates(Table 5) that the LH reduced the LST in the northwestern, northern, southern, and plateau regions by 1.3, 1.4,1.2, and 1.1°C, respectively, while the SH increased it by 0.2, 0.3, 0.1, and 0.3°C. Incident shortwave radiation increased the LST by 0.2, 0.3, 0.3, and 0.2°C. In addition,albedo increased the LST by 0.3, 0.4, 0.5, and 0.2°C. Incident longwave radiation reduced the LST by 0.5, 1.0,0.4, and 0.6°C. The comprehensive effect of these factors caused the LST in the four regions to change by –1.1,–1.4, –0.7, and 1.0°C, respectively.

A comparison of the comprehensive effects of each factor on the LST (Table 5) and the near-surface air temperature (Table 6) shows that the LST is directly proportional to the near-surface air temperature. Since the simulation tests change only the LULC, the near-surface air temperature changes were caused principally by these factors. From the variation in the national average LST caused by various factors, LH is found to have the most significant effect, causing a significant decrease in theLST. The vegetation coverage in each region increased and stomatal resistance decreased after the LULC data were updated, which led to increased evapotranspiration and in turn to a decrease in the LST. Simultaneously, the LST decreased slightly because of the increase in longwave albedo. Although there was a slight decrease in shortwave albedo, this decrease caused a small increase in the LST. The SH changes also had a great influence on the LST, which increased the LST of each region. Finally, these parameter adjustments reduced the LST simulation bias.

Table 6. Differences between average temperatures (°C) simulated by Tests 1 and 2

4. Summary and conclusions

In this study, we updated the existing LULC data in China by using a grid with a spatial resolution of 500 m to replace the one currently being used in RegCM4.Then, the effects of the update in LULC data on the nearsurface temperature simulation were assessed, and the causes of the improvement were discussed.

The updated land use data better reflect the land-surface characteristics in the study regions. In northwestern region, the semi-desert, short-grass, and desert coverages are greatly reduced, while the crop, mixed woodland, and urban coverages are increased. In northern region, the crop/mixed farming, mixed woodland, and urban coverages increase in the model, while the short-grass and irrigated crop coverages decrease; in southern region, the mixed woodland, crop/mixed farming, and forest/field mosaic coverages greatly increase when us-ing the new data, and the coverage of urban areas increases to a lesser extent. On the Tibetan Plateau, the crop/mixed farming coverage increases, while the irrigated crop, short-grass, and semi-desert coverages decrease.

Based on our analysis of the average annual, winter,and summer temperatures and their associated biases, we can see that the model that relied on the improved land use data better simulates temperature variations in the regions studied. This greatly minimizes the previous model’s drawback of simulating large areas with positive biases; it also reduces the overall simulated average temperature biases. For regions that originally had large negative temperature biases, the improved model also simulates smaller biases because the simulated temperatures are increased. Our analysis of the four study regions in China shows that the simulated average annual, summer,and winter temperature biases all decrease to varying degrees in the four regions.

Latent heat caused the greatest decrease in LST because of increasing evapotranspiration, which was induced by an increase in vegetation coverage and a decrease in stomatal resistance in each region when the LULC data were updated. Simultaneously, the LST decreased slightly because of the increase in longwave albedo. Sensible heat changes also had a great influence on the LST, which increased the LST in each region.Moreover, a change in shortwave albedo induced a slight increase in the LST. Overall, the adjustment of these parameters reduced the near-surface temperature simulation bias.