Present day bias and future change signal of temperature over China in a series of multi-GCM driven RCM simulations
- 223 Downloads
Simulation of surface air temperature over China from a set of regional climate model (RCM) climate change experiments are analyzed with the focus on bias and change signal of the RCM and driving general circulation models (GCMs). The set consists of 4 simulations by the RCM of RegCM4 driven by 4 different GCMs for the period of 1979–2099 under the mid-range RCP4.5 (representative concentration pathway) scenario. Results show that for present day conditions, the RCM provides with more spatial details of the distribution and in general reduces the biases of GCM, in particular in DJF (December–January–February) and over areas with complex topography. Bias patterns show some correlation between the RCM and driving GCM in DJF but not in JJA (June–July–August). In JJA, the biases in RCM simulations show similar pattern and low sensitivity to the driving GCM, which can be attributed to the large effect of internal model physics in the season. For change signals, dominant forcings from the driving GCM are evident in the RCM simulations as shown by the magnitude, large scale spatial distribution, as well as interannual variation of the changes. The added value of RCM projection is characterized by the finer spatial detail in sub-regional (river basins) and local scale. In DJF, profound warming over the Tibetan Plateau is simulated by RCM but not GCMs. In general no clear relationships are found between the model bias and change signal, either for the driving GCMs or nested RCM.
KeywordsBias Change signal Temperature Regional climate model China
At present, general circulation models (GCMs) are the most commonly used tool in climate change simulations and projections. Undoubtedly, they produce useful results on the global scale. However, they tend to exhibit a relatively low performances in describing the present day climate in regional scale due to their coarse resolutions, which may lead to a less reliable projected future changes (e.g. Gao et al. 2001; Giorgi et al. 2001; Zhou and Li 2002; Kang and Yoo 2006; Christensen et al. 2007; Xu et al. 2010; Flato et al. 2013; Kirtman et al. 2013; Woldemeskel et al. 2015; Jiang et al. 2016; Rummukainen 2016).
Thus, to meet also the increasing demand for fine scale regional and local climate information usable for impact assessment studies, regional climate models (RCMs) with higher horizontal resolution have been widely used nowadays (e.g. Déqué et al. 2005, 2012; Giorgi et al. 2009; Mearns et al. 2013; Niu et al. 2015; Gao and Giorgi 2017). Application of RCMs over East Asia is in particular important due to its complex topography and unique weather and climate systems (Hirakuchi and Giorgi 1995; Gao et al. 2001, 2006; Lee and Hong 2014; Yu et al. 2010, 2015; Zou and Zhou 2013; Niu et al. 2015; Shi et al. 2017). In addition to the finer scale topographically induced structures in the climate change signals, the RCMs may project some significantly different patterns of change over the region (Gao et al. 2008, 2012; Tang et al. 2017).
A thorough assessment of models’ biases firstly before they are applied for projections is important (Tebaldi and Knutti 2007; de Elia and Cote 2010; Christensen and Boberg 2013; Flato et al. 2013; Hui et al. 2018). In the meantime, as well known, credibility of RCMs depends largely on the quality of the GCMs providing the large-scale boundary conditions (van Oldenborgh et al. 2009; Déqué et al. 2012; Diaconescu and Laprise 2013; Tang et al. 2017; Tamara et al. 2019). Then to what extent the model biases affect the climate change signal, how and to what extent the GCM bias/signal may transfer (inherit) into the nested RCM? Efforts have been devoted in addressing the issues.
Of the first concern within the context of global warming is the change in temperature. Giorgi and Coppola (2010) found that for temperature in the regional scale, essentially no relationship is found between the bias and change signal based on their analysis on multi-GCM simulations. To be more specific, the projected future changes in temperature depend more on climate sensitivity of the models rather than corresponding regional biases as indicated by Giorgi (2008).
Di Luca et al. (2012) analyzed the added value of RCMs over North America. They found that the climate change signals for seasonal temperature were similar to those in the driving GCMs, and the spatial detail gained by downscaling was not large in the regional scale. Boberg and Christensenc (2012) suggested that over Mediterranean, climate models (regional and global) with systematic biases tend to overestimate regional amplification of global warming. Eum et al. (2015) investigated the relationship between climate change signals and biases for RCM/AOGCM simulations to assess the dependence of future projections on regional model biases for extreme temperature over southern Québec (Canada). They found that changes of extreme temperatures could be significantly affected by a large-scale forcing from boundary conditions, rather than systematic biases in the RCMs.
Three RCMs were applied over south-east Australia by Olson et al. (2016). The analysis shows that the RCMs tend to be biased cold than the GCMs during warm seasons. The downscaled projections are broadly consistent with the GCMs but to a weaker warming over land. Sørland et al. (2018) investigated the bias patterns and climate change signals from a set of RCM simulations under the EURO-CORDEX (the International COordinated Regional climate Downscaling Experiment, Giorgi et al. 2009) framework (Jacob et al. 2014). They found that the two RCMs systematically reduced the biases and modified climate change signals of the driving GCMs, most noticeably by lowering the warming of the driving GCMs.
Of the limited studies been conducted over East Asia, Gao et al. (2012) compared temperature changes simulated by a RCM and 2 driving GCMs during the monsoon season of May–September. They found that the large-scale boundary forcing from GCMs dominates the warming magnitude and broad patterns of the RCM over the region. Based on multi-RCM simulations driven by one GCM, Tang et al. (2016) reported a less warming projected compared to the GCM, in particular during the cold seasons. Park et al. (2016) found the systematic RCM biases, such as a cold bias along the coastline and a warm bias in the northern China, are mostly influenced by the GCM forcings. However, a systematic assessment of the bias–change relationship is still insufficient so far over the region, despite the many RCM simulations conducted here.
Recently, a new and unprecedented set of RCM (RegCM4) 21st century climate change experiments have been conducted over East Asia domain. The RCM is driven by 4 different GCMs at a 25 km grid spacing with the period from 1979 to 2099 under RCP4.5 emission pathway. Based on the simulations, the projected changes in haze pollution potential in China from the air environment carrying capacity perspective (Han et al. 2017), the future climate changes over the Xiong’an District (Wu et al. 2018), and future changes in thermal conditions over China by using the index of effective temperature (Gao et al. 2018), have been previously reported.
The experiments also allow us for a systematic comparison of the present day simulations and biases, and future change signals among the GCMs and RCM, as well as between the nested RCM and driving GCM. In the present study, we focus our analysis on temperature, aim at answering the above questions of how the driving GCM affect the RCM simulations, and what are the agreements and disagreements of them over China and its sub-regions.
The paper is organized as follows. A short description of the model, data and experimental designs are provided in Sect. 2. This is followed by result in Sect. 3. Finally we present the conclusion and discussion in Sect. 4.
2 Model, data and experimental designs
The Abdus Salam International Centre for Theoretical Physics (ICTP) RegCM Version 4.4 (http://gforge.ictp.it/gf/project/regcm/), with further developments and implementation of physical processes carried out from the first version of RegCM4 (Giorgi et al. 2012), is employed in this study. We use the same physical configuration as Gao et al. (2016, 2017) for a better performance of the model over the region. To be more specific, the land surface process used is the CLM3.5 (Oleson et al. 2008) and the convection is selected as Emanuel scheme (Giorgi et al. 2012). The other model physics includes radiation package from the NCAR Community Climate Model CCM3 (Kiehl et al. 1998), the non-local formulation of Holtslag et al. (1990) for the planetary boundary layer, and resolvable scale precipitation represented by the SUBEX parameterization (Pal et al. 2000). In addtion, the land cover data is updated following Han et al. (2015) based on the vegetation and vegetation regionalization maps of China instead of the default one in CLM. Note the RCM uses same model configuration, including the model physics, treatment of boundary conditions, and buffer zones, etc., for all simulations in the study, except the driving GCMs.
The four CMIP5 GCMs used to drive RegCM4 are CSIRO Mk3.6.0 (Rotstayn et al. 2010), EC-EARTH (Hazeleger et al. 2010), HadGEM2-ES (Collins et al. 2011), and MPI-ESM-MR (Stevens et al. 2012; Jungclaus et al. 2013). They are hereinafter referred to as CSIRO, EC, Had, and MPI, respectively. Horizontal resolutions of their atmospheric component are spectral T63 (1.875° × 1.875°, longitude × latitude), T159 (1.125°), N96 (1.875° × 1.25°) and T63, respectively. The models are selected due to their high resolutions to match the 25 km grid spacing RCM, data availability, and their good performances over the region (Jiang et al. 2016).
The gridded observational dataset of CN05.1 (Wu and Gao 2013) is applied to validate the simulations, with the resolution of 0.25° × 0.25°. The model outputs are interpolated bilinearly to the CN05.1 grid to facilitate the comparison. The same weighted average is used for the ensemble mean of the four global models and four RegCM4.4 simulations, referred to as ensG and ensR, respectively.
The analysis is focus on DJF and JJA over the China region and its 10 main river basins (MWRC 1981) as shown in Fig. 1, namely Songhuajiang River Basin (SRB), Liaohe Basin (LRB), Haihe Basin (HaiRB), Yellow River Basin (YLB), Huai River Basin (HRB), Yangtze River Basin (YRB), Zhujiang River Basin (ZRB), Southeast Rivers Basin (SERB), Southwest Rivers Basin (SWRB), and interior rivers in the Northwest Basin (NWRB). Definition of the 10 main river basins in general follows the major rivers in eastern China with dense population and more developed economic conditions. For western China, the rivers in southwest entering Southeast or South Asia are generalized as SWRB, while the rivers in northern Tibetan Plateau and northwest China as interior rivers in the NWRB. The definition is officially recommended by The Ministry of Water Resources of China and used widely in the scientific communities (e.g. water resources, hydrological, and climate, etc.). In addition, the river basins also agree roughly with the major climatic zones in China.
3.1 Biases in present day simulation
3.1.1 Whole of China
As shown in the table, although large differences in the values can be found for GCMs’ biases (Fig. 3; Table 1), the CORs across them are high, with values in the range of 0.39–0.77, all statistically significant at 95% confidence level. As shown in Fig. 3a, high level of agreements are found over plains and basins (cold bias), as well as mountain ranges (warm bias). Thus at least part of the large CORs of GCM biases can be attributed to their coarse resolution to describe the smaller scale topography and consequently the temperatures there.
Regional mean temperature bias (model simulation minus observation) during 1986-2005 and the change during 2080–2099 (relative to 1986–2005) in DJF and JJA for GCM/RCM and (RCM minus GCM) over China (unit: °C)
− 2.6/− 1.2 (1.3)
− 0.8/− 1.3 (− 0.5)
− 2.6/− 0.4 (2.2)
0.7/− 0.4 (− 1.1)
− 1.3/− 0.8 (0.5)
0.7/− 0.3 (− 1.0)
− 1.2/− 2.0 (− 0.8)
0.7/− 0.5 (− 1.2)
− 0.3/− 1.5 (− 1.2)
− 0.0/− 1.1 (− 1.1)
3.4/3.0 (− 0.4)
2.0/1.9 (− 0.1)
3.8/3.2 (− 0.6)
2.6/1.8 (− 1.8)
3.0/2.5 (− 0.5)
3.3/2.4 (− 0.9)
3.1/3.0 (− 0.1)
2.7/2.4 (− 0.3)
As indicated by the cross symbols in Fig. 4b for model agreements, biases of RCM show more consistencies among the runs characterized by a cold bias in northern part of China, eastern coast areas, and the Tibetan Plateau, and a large warm bias over the deserts in the Northwest. It is noted that the above biases show agreements with the RCM driven by re-analysis, except a dominant cold bias along the coast region of Yellow Sea. Similar to reported by Park et al. (2016), this is largely due to the prevailing cold SST bias simulated by the GCMs (figure not shown for brevity).
Regional mean biases for RCM are in general 1 °C lower compared to the driving GCMs with the ensR as − 1.1 °C (Table 2). It is interest to note that the regional mean differences of bias between RCM and driving GCM show close value (~ 1 °C) to the RCM when driven by re-analysis (Gao et al. 2017). In the meantime, influence from driving GCM can be also observed. For the warmer ones of CSIRO and Had, CdR and HdR are also warmer compared to EdR and MdR although values are still negative.
3.1.2 River basins
In the meantime over basins where RCM bias is not so systematic, the correlation is in general not significant. Exceptions are found over, e.g. YRB and ZRB, where negative correlations exist. However the differences of the GCM and RCM biases are small there and can be at least partly considered as a random behavior of the model. This further confirms the complex relationships existed between the RCM and driving GCM.
Different from DJF, correlations between RCM and GCM biases are more evident in JJA. With general cold bias of the RCM (see Fig. 4b), it tends to lower the temperatures. In the northern basins in eastern China (SRB, LRB, HaiRB, YLB, HRB and SERB), a kind of shift to the cold direction can be found in the RCM simulations. Again in basins with no systematic bias, the biases of RCM show less correlation with GCM (e.g. in YRB).
3.2 Change signals in the end of 21st century
3.2.1 Whole of China
Magnitude of the warming simulated by RCM in general follows but smaller than the driving GCM. Regional mean changes are in the range of 1.8–3.2 °C with ensemble mean as 2.5 °C (Table 2), half degree lower than GCM. The largest difference is − 1.8 °C between MdR and MPI with the least of − 0.1 °C between EdR and EC. In northern basins, the warm bias of the RCM may lead to less snow coverage and consequently a weaker snow-albedo feedback, and finnally a weaker warming. However in the warm south, the mechanism behind is much more complex and need to be further explored in future studies.
The broad pattern of warming in RCM in general agrees well with the driving GCM as shown in Fig. 7. CORs between RCM and the driving GCM are in the range of 0.43–0.64, all significant at 95% level (Table 4). However, besides more spatial details in RCM, noticeable differences are found over the Tibetan Plateau where a dominant warming exists in 3 out of the 4 simulations except EdR. But still greater warming can be observed in eastern part of the Plateau in EdR. The profound warming in RCM may be related to a stronger feedback effect of warming and reduction in surface albedo due to melting of snow cover (Giorgi et al. 1997). However, further analysis in the future is needed to identify clearly the reasons of the different model responses.
Magnitude of the regional mean warming in JJA from RCM is close to but slightly lower than the driving GCM, except a larger difference of 0.9 °C for CSIRO and CdR (Table 2). The later contributes largely to the 0.3 °C lower value of ensR. A greater warming (2.4–3.0 °C) is found along 40°N and over eastern part of the Tibetan Plateau in ensR (Fig. 8b). As in DJF, the spatial pattern of RCM is consistent with the driving GCM in a broad scale. CORs for the pairs are all significant and range from 0.25 to 0.67. It is of interesting to note that although the RCM is not so responsive to biases of GCMs (Table 3), it responses the climate change signal of GCMs well, either for the regional mean value or the broad scale spatial distribution of the changes.
In the meantime, as shown in Fig. 8d, f, h j, spatial patterns across the RCM simulations are not so consistent as in DJF. For CORs among RCM simulations, only 3 out of 6 are positive and significant (Table 5).
3.2.2 River basins
Spatial correlation coefficients (CORs) of temperature change between different pair of model simulations in DJF/JJA over the 10 major river basins for 2080–2099 (relative to 1986–2005)
− 0.38*/− 0.14
− 0.56*/− 0.70*
Spatial correlation coefficients (CORs) of temperature change between ensG and ensR in DJF/JJA over China and the 10 major river basins for early (2016–2035) mid (2046–2065) and end of the 21st century
3.3 Temporal evolution
It is noted that the year to year changes of nested RCM follows driving GCM well. The detrended CORs for all model pairs and over all basins in both DJF and JJA are statistically significant (not shown for brevity). Thus the inter-annual variability in the nested RCM is dominated by the driving GCM.
4 Conclusion and discussion
In the present paper, we report analysis on temperature simulations from a set of 4 RCM 21st century climate change experiments over China. The biases in the present day simulation and change signal in the future projection are presented and inter-compared with the driving GCM, with the focus on the seasons of DJF and JJA.
For present day simulation, the GCMs reproduce well the climatology, while the RCM provides with more spatial detail of the distributions. Biases of the GCMs or RCM show consistencies among themselves, both in spatial distribution and magnitude in the two seasons. The biases of RCM in DJF, at least to some extent, are transferred from the driving GCM. In JJA, internal physics of the RCM may play a larger role, with almost no correlations found for bias distributions between the RCM and driving GCM. Furthermore, the RCM tends to add its systematic biases which lead to a larger bias in its ensemble compared to GCM. The improvement of RCM is more significant in DJF, with the largest added value as a better description of temperature over places with complex terrains.
Concerning future changes, the pattern and magnitude show large differences across the GCMs. It is difficult to conclude a correlation between the GCM bias and change signal. The two models with greatest warming both in DJF and JJA, CSIRO and Had, are the ones with largest climate sensitivity among the 4 (Myhre et al. 2013; Chen and Gao 2019). This further confirms previous findings that regional temperature change signal depends much more on the GCM sensitivity rather than local processes (e.g. Giorgi 2008; Giorgi and Coppola 2010).
The magnitudes and broad regional scale distribution of the warming in RCM in general follows the driving GCMs, indicating dominance of the warming pattern from GCM as reported by Gao et al. (2012). Thus similar to GCM, almost no relationship has been found between the RCM biases and projection. The most significant added value in RCM is the much finer spatial detail in sub-regional (river basin) and local scales. The largest difference between RCM projection and GCM is found over the Tibetan Plateau in DJF with a much profound warming in RCM. This leads to the less correlation of the warming patterns between RCM and GCM in the sub-regional scale (the river basins). Furthermore, the correlation varies in the decadal scale.
The change patterns among RCM simulations show good consistencies in DJF when large scale circulation dominates, but not in JJA, the monsoon season with medium and small weather and climate systems prevail.
It is of interest to find that a general reduced warming is projected by the RCM, which is consistent with findings in Australia, Europe, and East Asia (Olson et al. 2016; Tang et al. 2016; Sørland et al. 2018). However, whether this is region or model dependent needs to be further explored. Further analysis is also needed to understand the physical mechanisms behind. For example over Europe, the lowing of warming is significant in JJA, which may relate to the drying feedback over the region (Sørland et al. 2018). However in our simulations, the less warming is more evident in DJF instead, indicating other processes play major roles.
Due to the limited scope of the study, we reported the model behaviors on temperature simulation and projection only in the present paper. In-depth investigation on the physical mechanisms behind requires more special design experiments and need to be conducted in the future. Two sets of experiments are planned so far in the next stage. The first one will be focused on the role of model resolution in projecting future changes, as did by Gao et al. (2006) for present day climatology. In the experiments, resolutions of the RCM will decrease gradually from 25 km to a coarse resolution comparable to the driving GCM, aiming to identify how the model resolution affects the projected signal. The second set is on model physics. Simulations will be conducted using different convection parameterization schemes, e.g. the Tiedtke (Tiedtke, 1989) implemented recently in RegCM4, to investigate dependence of the change signal on model physics.
As indicated by Gao et al. (2012), precipitation change over the region shows larger uncertainties during the monsoon season. Thus the bias and change analysis will be expanded to precipitation. Finally, climate change simulations conducted over this large domain are still very limited so far compared to other CORDEX regions. Further completion of large multi-GCM/RCM ensembles within the CORDEX framework are needed to better address the issues.
This research was jointly supported by the National Key Research and Development Program of China (2018YFB1502803), the Strategic Priority Research Program of the Chinese Academy of Sciences (Grant No. XDA20060401), the Science and Technology Program of Yunnan “Impact assessments and monitor-forecasting technology of meteorological disasters for Yunnan plateau characteristic agriculture under climate change” (Grant No. 2018BC007), and National Natural Science Foundation of China (41805074, 41375104).
- Chen N, Gao XJ (2019) Climate change in the 21st century over China: The projections by a RCM and the driving GCM. Atmos Ocean Sci Lett (accepted)Google Scholar
- Christensen JH, Boberg F (2013) Temperature dependent climate projection deficiencies in CMIP5 models. Geophys Res Lett 39:L24705Google Scholar
- Christensen JH, Hewitson B, Busuioc A et al (2007) Regional climate projections. In: Solomon S, Qin D, Manning M et al (eds) Climate change 2007: the physical science basis. Contribution of Working Group I to the 4th assessment report of the Intergovernmental Panel on Climate Change. Cambridge University Press, Cambridge, pp 847–940Google Scholar
- Flato G, Marotzke J, Abiodun B et al (2013) Evaluation of climate models Climate Change 2013: the physical science basis. Contribution of Working Group I to the Fifth Assessment Report of the Intergovernmental Panel on Climate Change. Cambridge University Press, Cambridge, pp 741–866Google Scholar
- Giorgi F, Jones C, Asrar G (2009) Addressing climate information needs at the regional level: the CORDEX framework. WMO Bull 58:175–183Google Scholar
- Han Z, Gao XJ, Shi Y, Wu J, Wang ML, Giorgi F (2015) Development of Chinese high resolution land cover for the RegCM4/CLM and its impact on regional climate simulation. J Glaciol Geocryol 37:857–866 (in Chinese) Google Scholar
- Kirtman B, Power SB, Adedoyin JA et al (2013) Near-term climate change: projections and predictability. Climate Change 2013: the physical science basis. Contribution of Working Group I to the Fifth Assessment Report of the Intergovernmental Panel on Climate Change. Cambridge University Press, Cambridge, pp 953–1028Google Scholar
- MWRC (The Ministry of Water Resources of China) (1981) Preliminary assessments of water resources in China. The Ministry of Water Resources of China, Beijing (in Chinese) Google Scholar
- Myhre GD, Shindell FM, Bréon W et al (2013) Anthropogenic and natural radiative forcing. Climate Change 2013: the physical science basis. Contribution of Working Group I to the Fifth Assessment Report of the Intergovernmental Panel on Climate Change. Cambridge University Press, Cambridge, pp 659–740Google Scholar
- Rotstayn LD, Collier MA, Dix MR et al (2010) Improved simulation of Australian climate and ENSO-related rainfall variability in a global climate model with an interactive aerosol treatment. Int J Climatol 30:1067–1088Google Scholar
- Tamara J, McGrath F, Macadam I, Jones R (2019) High-resolution climate projections for South Asia to inform climate impacts and adaptation studies in the Ganges–Brahmaputra–Meghna and Mahanadi deltas. Sci Total Environ 650:1499–1520. https://doi.org/10.1016/j.scitotenv.2018.08.376 CrossRefGoogle Scholar
- Wu J, Gao XJ (2013) A gridded daily observation dataset over China region and comparison with the other datasets. Chin J Geophys 56:1102–1111 (in Chinese) Google Scholar
- Wu J, Gao XJ, Xu Y (2018) Climate change projection over Xiong’an District and its adjacent areas: an ensemble of RegCM4 simulations. Chin J Atmos Sci 42(3):696–705 (in Chinese) Google Scholar
Open AccessThis article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made.