Changes in building climate zones over China based on high-resolution regional climate projections

Based on the up-to-date observations and five sets of high-resolution climate projections under RCP4.5 over East Asia using a regional climate model, this study defines building climate zones over China and assesses their past and future changes according to an established climate classification scheme. The five-model ensemble captures the observed general distribution of climate zones, with a distinct latitudinal pattern over the densely populated eastern China, a subarctic zone in the western mountains including the Tibetan Plateau and a cool dry zone in the Tarim and Turpan basins; large disparities are found in northwestern China due to warm and dry biases. Observational data identifies a recent northward shift of most climate zones in eastern China and a shrinking of the subarctic climate zones over the Tibetan Plateau, which caused a building climate zone change for 18% of the country’s land area. As the warming continues in the future, the five-model ensemble projects additional climate zone changes influencing 43% and 55% of the country by mid- and late-century, respectively. In addition, the total area of the subarctic zone is projected to decrease from 16.6% of the country in present-day climate to 9.2% and 7.4% in the mid- and late-century, respectively, and that of the hot or very-hot climate zone is projected to expand from 4.2% to 9.1% and 11.3%, respectively. These changes should be taken into consideration in long-term development planning related to urbanization, energy efficiency, and environmental sustainability.


Introduction
The Fifth Assessment Report (AR5) of the Intergovernmental Panel on Climate Change (IPCC) found that 'warming of the climate system is unequivocal' . Global average of surface temperature shows a warming of 0.85 [0.65-1.06] • C over the period of 1880 to 2012, with faster warming in recent decades. Further warming which is estimated to be in the range of 0.3 • C-4.8 • C averaged in the global by the end of the century (2081-2100) relative to the present day (1986)(1987)(1988)(1989)(1990)(1991)(1992)(1993)(1994)(1995)(1996)(1997)(1998)(1999)(2000)(2001)(2002)(2003)(2004)(2005) is projected with continued emission of greenhouse gases (GHGs) (IPCC 2013). One of the potentially more significant effects of climate change is the alteration of the distribution, speciation and life cycle of vegetated ecosystems (IPCC 2013). As a simple method to combine the climate variables and vegetation, climate classification such as the Köppen climate classification (Köppen 1936) and its modified versions Köppen-Geiger (K-G, Kottek et al 2006), Köppen-Trewartha (K-T, Trewartha and Horn 1980), and the cluster analysis (Zhang and Yan 2014, Netzel and Stepinski 2016) has been frequently used in a broad range of topics concerning climate and climate change impact as they related to biogeography, hydrology, horticulture, agriculture, and engineering design (e.g. Kottek et al 2006, de Castro et al 2007, Gao and Giorgi 2008, Ruble and Kottek 2010, Baker et al 2010, Jacob et al 2012, Shi et al 2012, Chen and Chen 2013, Feng et al 2014, Zhang et al 2017, Fermandez et al 2017, He et al 2019. The choice of classification often depends on the application purpose. For example, in the U.S., the USDA Hardiness zone is used for horticulture and agriculture (USDA Plant Hardiness Zone Map 2012), and the building zone classification for energy efficiency is based on temperature, precipitation, and heating and cooling degree days (HDD and CDD) (ANSI/ASHRAE Standard 169 2013).
Similar to the Köppen climate classification, climate zones in China are defined according to a set of rules (e.g. NMA 1979, 1994, 2002, EBCAS 1985. The climate classification for building design started in the 1980s, and the latest standard of climate regionalization for architecture was released in 1993 (MCC 1993). This building design standard is defined according to the mean temperature in January and July, the cumulative days of daily temperature higher or lower than a certain threshold, the annual total precipitation, and the maximum wind. Following the 1993 standard, the spatial distributions of the building climate zones have been investigated based on observed climate (e.g. Fu et al 2008, Bai andYang 2018). With the modernization of China's infrastructure, building energy efficiency has become increasingly important, and new approaches towards the development of building climate zones for China have emerged. For example, Bai et al (2020) investigated the applicability of the ANSI/ASHRAE Standard to climate in China and proposed additional zones to identify high elevation as the cause for cool, cold, and very cold zones in regions such as the Tibetan Plateau. In this study we examine how future climate change may influence the building climate zones according to the ANSI/ASHRAE Standard.
In a warming world, climate zones are expected to migrate with the changes of temperature and precipitation. To investigate the response of the climate system to increases in GHG concentrations and to project future climate, the primary tools are coupled global atmosphere-ocean general circulation models (GCMs). Based on the simulations from the global models, changes of climate zones have been analyzed by many researchers. For example, using the K-G classification, Rubel and Kottek (2010) generated a series of digital world maps for the extended period 1901-2100 and showed the largest shifts between the main classes of equatorial climate (A), arid climate (B), warm temperate climate (C), snow climate (D) and polar climate (E) on global land areas are estimated to be 2.6%-3.4% (E to D), 2.2%-4.7% (D to C), 1.3-2.0 (C to B) and 2.1%-3.2% (C to A). Feng et al (2014) indicated that the K-T climate types would shift toward warmer and drier climate types from the current climate distribution based on CMIP5 models. Belda et al (2016) found a substantial decline in ice cap, tundra, and boreal climate in the warming world, accompanied by an expansion of temperate climates, dry climates, and savanna, nearly unanimous within the CMIP5 ensemble. Generally, a model with higher resolution performs better compared to its lower-resolution counterpart (Gao et al 2006, Shi et al 2018. With the rapid increase of computational capacity, high-resolution models are becoming increasingly common in Phases 5 and 6 of the Coupled Model Intercomparison Project (CMIP5, Taylor et al 2012;CMIP6, Eyring et al 2016). But the resolution is still too coarse to capture the effects of local and regional forcing, especially in regions with complex topography such as China, leading to model performance challenges (e.g. Jiang et al 2005, Xu et al 2010, Yu et al 2011. From the application side, coarse resolution global models cannot resolve the spatial heterogeneity needed for developing local and regional climate adaptation strategies. Nested high-resolution regional climate modeling (RCM) has been used to improve simulation fidelity and to better represent the spatial heterogeneity over China (e.g. Gao et al 2001, 2006, Qian and Leung 2007, Zou and Zhou 2013. However, few high-resolution studies on climate classification have been conducted in general and for building energy efficiency purpose over China in particular.
In this paper, climate classifications used for building code and energy efficiency will be analyzed based on five sets of high-resolution climate change simulations under RCP4.5 scenario over East Asia using the regional climate model  The initial and boundary conditions at six hourly intervals are needed to drive the RegCM4.4 model which is usually derived from GCMs. The ratio of the RCM to GCM resolutions should be less than 10, and preferably in the range of 3-5. Considering model resolution, performance over the domain of interest, model climate sensitivity, and data availability for desired RCPs, we selected five CMIP5 global models to include in this study, CSIRO-MK3-6-0, EC-EARTH, HadGEM2-ES, MPI-ESM-MR and NorESM1-M (table 1). Note that NorESM1-M, MPI-ESM-MR, HADGEM2-ES are considered the highest priority of GCMs for inclusion in the CORDEX CORE simulations (https://cordex.org/experimentguidelines/cordex-core/cordex-core-simulation/). In addition, previous studies have shown that the five global climate models have a good performance over China (Zhou et al 2014, Jiang et al 2016.
Of the 120 years' simulation, the period of 1986-2005 is considered as the historical reference, and changes relative to the historical reference are defined for both the middle-century (2046-2065) and late-century (2080-2099), respectively. The observational temperature and precipitation dataset (CN05.1) developed by Wu and Gao (2013) at 0.25 • × 0.25 • (latitude by longitude) resolution has been updated to cover the period 1961-2018, and is employed in this study to validate model performance and to quantify past changes of climate zones. It is worth to be mentioned that the performance of the RCM has been evaluated in previous studies (Han et al 2017, Gao et al 2018, Shi et al 2018, Wu and Gao 2020. Therefore, in this paper, we focus on the simulations and future changes of climate zones based on results from the RCM without presenting the model validation for climate simulation.

Bias correction
Just like global models, climate from RCMs need to be corrected for biases to obtain more realistic estimates of relevant variables to support adaptation assessment (Rahimi et al 2019). Therefore, prior to their application in building climate zone assessment, the RCM outputs went through a two-step bias correction procedure (Han et al 2019) making use of the climatological annual cycle derived from the daily data of CN05.1. Specifically, following a similar approach of Kim et al (2016), the model-based anomalies were first interpolated to the CN05.1 grid and added to the CN05.1 annual cycle to derive an intermediate dataset that has the same climatology as CN05.1 and same anomalies as the model. Quantile delta mapping (QDM, Cannon et al 2015) was then employed to correct the systematic biases in the RCM output, especially in the cumulative distribution function of the simulated variable, using the CN05.1 data as the observational reference. Briefly, the method is based on calculating the relationship between model simulations and observations during the historical period and applying the derived correction factors to future climate projections.
As reported by previous studies, the conventional quantile mapping (QM) method effectively removes the model biases in mean temperature and precipitation, interannual variability, and extreme events (Tong et al 2017, Han et al 2018 and the biascorrected results has been used in the projection of the thermal comfort conditions over China (Gao et al 2018). Furthermore, Tong et al (2020) compared the results after bias correction using the methods of QDM and QM with the original output from the RCM and the results show that both the QM and QDM methods are effective in removing the systematic model biases. It should be noted that the QDM method conserves the future change signal in quantiles simulated by the RCM while the QM does not.

Climate zones classification scheme
As well known, climate has a great impact on the energy use of residential buildings. In different climate zones, the requirements, such as wall insulation are different (Athalye et al 2016). The building climate zone classification approach from the U.S. Department of Energy Pacific Northwest National Laboratory (PNNL) (https://eere.energy.gov/buildings/publications/pdfs/ building_america/climate_region_guide.pdf) is based on temperature, precipitation, as well as HDD and CDD due to their relevance to buildings' energy consumption and efficiency. The zones range from 1 in the hottest to 8 in the coldest regions and most are further divided into subzones (table 2). The HDD18 • C and CDD10 • C in the criteria for climate zones are defined as follows: HDD18 • C: the sum of the absolute difference between 18 • C and the daily mean temperature for 1 d when the mean temperature is less than 18 • C. Mixed-Humid (4 A)/Dry(4B) CDD10 • C ≤ 2500 and HDD18 • C ≤ 3000 4C Mixed-Marine(4 C) 2000 < HDD18 • C < 3000 5 Cool-Humid (5 A)/Dry(5B)/Marine(5 C) 3000 < HDD18 • C ≤ 4000 6 Cold-Humid (6 A)/Dry(6B)/Marine(6 C) 4000 < HDD18 • C ≤ 5000 7 Very cold 5000 < HDD18 CDD10 • C: the sum of the absolute difference between the mean temperature for 1 d and 10 • C when the mean temperature is more than 10 • C.
Within the U.S., these climate zones were set along county boundaries, making it easier for builders to determine the climate zone building code that a specific neighborhood is subject to; in this study, similar to other climate classifications used in China, boundaries of municipal administration units are not considered. To define climate zones based on the five-model ensemble, the HDD and CDD are first estimated separately for each model and their averages among the five models are then used to classify the climate zones.

Climate zones and past changes
As the observational data is available during 1961-2018, we divide the record into two periods of approximately 30 years, 1961-1990 and 1991-2018. The spatial distribution of the climate zones over China in the two periods and the difference between them are shown in figure 1. Note that changes of all the climate types are analyzed, with a focus on the major 13 groups; climate zones 1B, 1C, 2B, 2C, 5C and 6C each occupies less than 0.1% of the domain and are not shown in the figure.
Distinct latitudinal distribution of climate zones over East China can be found in figure 1(a) and (b), with the hot climate zone (2A) in the south, warmhumid climate zone (3A) in much of the Yangtze River Basin, mixed-humid climate zone (4A) in the Huanghuai River Basins, cool-humid climate zone (5A) along the Taihang Mountain, cold-humid climate zone (6A) extending from the south of the northeast to east of Tibetan Plateau, and the very cold (7) and subarctic climate zones (8) in northeastern China. In western China, subarctic climate zone is dominant in the mountains including the Tibetan Plateau while cool dry climate zone (5B) is found in the Tarim and Turpan basins. Comparison between the two periods indicates that warming during the past several decades has caused a general northward shift of the climate zones (including 1A, 2A, 3A, 4A, 5A, 6A and 7 in eastern China) with an expansion of the spatial coverage for most of them and the greatest decrease in spatial coverage of the zone 8 (subarctic) (figures 1(c) and 2). The climate zone 8 in some parts of the Tibetan Plateau has been replaced by zone 7, and zone 4B in the western basins has increased. The spatial coverage of 4B in the two periods is 0.1% and 1.3%, respectively (figure 2). Taking all changes together, over the past several decades, about 18% of the country has experienced a shift of building climate zone. As the GHG-induced warming continues in the future, more areas of the country will be influenced.

Model validation and future projections
The climate zones based on the five-model ensemble are compared against the observed data from CN05.1 in figure 3. Different from the 30-year interval used in section 3.1, here the period 1986-2005 is used as the historical reference, following the IPCC AR5 convention (IPCC 2013). The five-model ensemble mean reproduces the spatial distribution of the climate zones over China reasonably well (figures 3(a) and (b)). And it also performs generally better compared to the ensemble mean without bias correction, especially in Northwest China, Southwest China and the Tibetan Plateau (figures not shown). Compared with the observation, the largest differences are found mainly in northwestern China where the model features a greater coverage of 4B while observations feature a greater coverage of 5B, which results from a combination of warm and dry biases of the RCM (Gao et al 2018, Shi et al 2018). The fivemodel ensemble mean also well captures the spatial coverage of the climate zones over China, characterized by the maximum percentage (>20%) of very cold  1961-1990 (a) and 1991-2018 (b), and the differences between the two periods (c). climate zone, and then the subarctic, the warm, the cold, the cool, the mixed, the hot and the very-hot climate zones ( figure 4) The spatial pattern of climate zones and their projected changes by the mid-and late-century based on the multi-model ensemble mean are shown in figures 3(c)-(f). Relative to the historical reference, significant decrease of the subarctic climate zone and increase of the very-hot and hot climate zones are projected, with greater changes in the late-century than mid-century (figures 3(c) and (d)). Specifically, the fractional coverage of the subarctic zone is projected to decrease from 16.6% to 9.2% and 7.4% by the mid-and late-century, respectively, while that of the very-hot (hot) climate types is projected to increase from 0.3% (3.9%) to 1.0% (8.1%) and 1.5% (9.8%), respectively (figure 4). The overall fractional coverage of the very cold climate zone would decrease from a historical reference of 24.7% to 23.1% and 22.4% in mid-and late-century, respectively, but greater    2046-2065(Mid) and 2080-2099 and climate zone changes by the mid-and late-century in the three major economic regions of Beijing-Tianjin-Hebei (JJJ), Yangtze River Delta (YRD) and Pearl River Delta (PRD). JJJ  YRD  PRD   3A  4A  5A  6A  7A  2A  3A  4A 1A  2A  RF  0  44  24  20  12  0  84  16  0  100  Mid 41  13  25  18  3  3  97  0  47  53  End  48  9  25  17  0  differences can be found in the spatial distribution, with a decrease in northeast China and an increase in the Tibetan Plateau (figures 3(c) and (d)). The projected warming would cause a widespread change of the building climate zones associated with a northward shift of the boundaries for climate zones including 1A, 2A, 3A, 4A, 5A and 6A in eastern China, and an extension of 3B in the southern part of Northwest China (figures 3(e) and (f)). These widespread changes towards warmer and more humid building climate zones are projected to influence 43% of the country by the mid-century and 55% by the latecentury. The climate zone changes are especially striking in the densely populated regions in eastern China with rapid urbanization and widespread construction (table 3). For example, over the Jing-Jin-Ji region, climate zone 3A is absent in the present day, but is projected to cover 41% and 47% of the region by the mid-and late-century, respectively (table 3). The increase is mainly caused by the change from 4A to 3A due to warming. Other changes from cold to hot climate zones are also found, including from 7A to 6A, 6A to 5A and 5A to 4A. Summing up all changes together, 72% (mid-century) or 83% (late-century) of the region are projected to experience a change of building climate zone. Similarly, 37% of the Yangtze River Delta region is projected to experience a building climate zone change by late-century, when the area of climate zone 2A would increase from zero in the present day to 21% of the region and the climate zone 4A would be completely replaced by 3A (table  3). The whole Pearl River Delta region belongs to zone 2A in the present day, and 62% of it would shift to 1A by the late-century. According to the code for thermal design of civil building (MOHURD 2016), the requirements, such as for wall insulation, are less stringent in warmer climate zones. Thus, when the region is reassigned to a warmer climate zone, it will influence construction materials required, construction cost, energy demand, and energy efficiency of new buildings (Athalye et al 2016).

Percentage coverage of each climate zone
To examine the model-related uncertainties, the late-century changes of the building climate zones projected by the five individual models are compared in figure 5. The general spatial patterns of the projected changes are similar among the models, all featuring changes from zone 8 to zone 7 over the Tibetan Plateau, 3A to 2A south of the Yangtze River, and 7 to 6A, 6A to 5A, 4A to 3A, and 2A to 1A in eastern China. The spatial extent of areas influenced by such changes differ substantially among the models, and reaches 45%, 46%, 48%, 62% and 72% based on climate changes projected by the RegCM model driven by EC-EARTH, MPI-ESM-MR, NorESM1-M, CSIRO-Mk3-6-0 and HadGEM2-ES, respectively. The CSIRO-Mk3-6-0 and HadGEM2-ES models are the most sensitive due to the greater magnitude of warming in these two models (Wu and Gao 2020).

Conclusions and discussion
In this paper, using the definition described by the U.S. Department of Energy Pacific Northwest National Laboratory (PNNL), we investigate future changes in building climate zones based on five sets of high-resolution regional climate change projections under RCP4.5 scenario over East Asia. The main findings are summarized as follows: Large uncertainties in the spatial pattern of climate zones can originate from the GCMs boundary forcing, the RCP scenarios, and the RCMs. Multimodel ensemble of climate projections is an important and effective way to reduce these uncertainties (Giorgi et al 2009). While considering multiple GCMs, our study makes use of one RCM only. Adding other RCMs may reveal further uncertainties in projecting the future changes. While the bias correction method used in the study has some impact on the uncertainty, differences between the projected changes with and without bias correction indicate that uncertainties related to bias correction is rather low, influencing approximately 4% of the country land by the end of the century.
Compared with the previous climate classification methods that were developed for different application purpose, the climate zone classification used in this study is specifically designed for considerations related to building code and energy efficiency. The spatial distribution of building climate zones over China based on our results is similar to that based on the Köppen climate classification (Shi et al 2012) and the climate types developed by Zheng et al (2010), and thus can be used as an alternative to the existing methods. In this study the climate zone boundaries are entirely determined by climate factors; in practice, the climate zone boundaries should follow those of the municipal administration units to facilitate the implementation of building energy efficiency code by the local government.

Data availability statement
The data that support the findings of this study are available upon reasonable request from the authors.