The Boundary Dataset Based on Two Algorithms of the Hierarchical Solar Radiation Zones in China

Jiang, H. ^1,2*  Yao, L.¹

1. The State Key Laboratory of Resource and Environmental Information Systems, Institute of Geographic Sciences and Natural Resources Research, Chinese Academy of Sciences, Beijing 100101, China;

2. The State Key Laboratory of Remote Sensing Science, School of Geography, Beijing Normal University, Beijing 100875, China

Abstract: Solar radiation zoning is the foundation for guiding solar energy utilization and formulating regional development plans. The dataset provide the boundaries based on two algorithms for hierarchical solar radiation zoning in China, including fine boundaries of 5 primary zones and 10 sub-zones. An automatic zoning algorithm based on Gaussian mixture model is employed to identify solar radiation zones, whose number is adaptively determined by Bayesian inference. We utilize the ground observations of solar radiation at 98 stations from 2007 to 2020 for Gaussian mixture model fitting, and then introduce spatially continuous solar radiation products from remote sensing images into the fitted model to identify the boundaries of adjacent zones. The zoning results are validated using sunshine-based solar radiation products at 716 weather stations. It is revealed that the zoning algorithm can divide stations with different solar radiation characteristics into plausible zones with an accuracy rate of approximately 90%. In addition, most inaccurate stations are located within the zone rather than near the boundaries, which further proves the reliability of the used algorithm and identified boundaries.

Keywords: solar radiation; zoning boundary; Gaussian mixture model; solar energy utilization

DOI: https://doi.org/10.3974/geodp.2023.02.07

CSTR: https://cstr.escience.org.cn/CSTR:20146.14.2023.02.07

Dataset Availability Statement:

The dataset supporting this paper was published and is accessible through the Digital Journal of Global Change Data Repository at: https://doi.org/10.3974/geodb.2023.07.07.V1 or https://cstr.escience.org.cn/CSTR:20146.11.2023.07.07.V1.

1 Introduction

Solar photovoltaic (PV) shows great potential in promoting global carbon neutrality^[1]. Due to the geographical diversity of climate, the power generation efficiency of solar PV always presents significant spatial differences^[2]. Solar radiation zoning is considered to be the most direct and practical method to gain insight into regional differences of solar energy, and is the basis for strategic planning oriented to solar market, site selection of PV power plants and placement of energy storage facilities^[3?C5]. For example, solar radiation zoning is beneficial for government energy departments to design PV development roadmaps^[6], screen suitable regions for centralized PV plants^[7], put solar facilities in public, etc.^[8].

Climate zones have been extensively studied worldwide, and researchers have established a variety of climate zoning systems based on different climate variables and indices^[9]. In China, by taking passive heat energy utilization of buildings as reference^{[10, 11]}, the widely used building climate zones divide the climate into five major categories: severe cold, cold, hot summer & cold winter, hot summer & warm winter, and temperate. The development of bioclimatic map theory has gradually promoted the concept of bioclimatic zones^{[10, 12]}. Lam et al. (2006) analyzed weather data from 18 typical cities in China and classified the country into nine bioclimatic zones. Wan et al. (2010) identified 5 bioclimatic zones and 8 subzones in China based on a long-term investigation of summer and winter heat stress discomfort. In addition to temperature, environmental factors such as humidity and wind speed are also gradually included into the zoning process of climates, resulting in more elaborate climate zones, which however become fragmented and complicated.

There are relatively few zoning schemes aiming at solar energy use. Traditional solar radiation zoning is based on the amount of annually cumulative solar radiation observed at weather stations, searching for natural discontinuities to determine the zoning threshold^{[11, 13]}. For example, Lau et al. (2007) calculated the monthly average clear sky index with solar radiation observations of 123 stations and proposed a scheme containing 5 solar radiation zones through cluster analysis^[14]. However, due to the sparse and uneven distribution of ground observation stations, it is often difficult to determine the zone to which a point far from the observation stations belongs. Liu et al. (2017) first divided 98 solar radiation observation stations into 5 zones through K-Means clustering and then adopted support vector machine to establish an empirical relationship between global solar radiation and meteorological variables within each zone. Subsequently, this relationship was used to predict the zones that more densely distributed weather stations belong to, thus identifying more reliable zone boundaries than that raised by Lau et al. (2017)^[15]. As the regional zoning process still relies on the spatial density of weather stations, the scheme proposed by Liu et al. (2017) suffers from incorrect zoning and inaccurate boundaries in the climate transition zone. Spatially continuous estimates of surface solar radiation are the only method to determine the accurate boundary of adjacent zones. A large number of studies have demonstrated that remote sensing inversion products can capture regional differences and hourly changes of surface solar radiation at the kilometer to meter scale^[16], so that they can be employed to identify the boundary of solar radiation zones at fine scales. In addition, traditional clustering methods require determining the number of solar radiation zones based on prior knowledge. However, to determine the most appropriate number of zones in practice is faced with serious challenges, and inappropriate selection of the number of zones usually leads to undesirable or even misleading results, such as over-fitting^[17].

In view of the diversification of zoning systems, inaccurate identification of zoning boundaries, and difficulty in determining the right number of zones, a new algorithm based on Gaussian mixture model is proposed to identify the hierarchical system of solar radiation zoning and determine zoning boundaries by using spatial continuous remote sensing inversion products^[18]. The algorithm treats the number of zones as a random variable and then automatically determines the most appropriate number from the training data by virtue of Bayesian inference. Bayesian inference also ensures more reliable zoning results by selecting a more appropriate prior distribution and additional physical knowledge^[19]. Through integrating surface solar radiation observations from 2007 to 2020 and remote sensing estimates with spatial and temporal continuity, the algorithm proposed in previous studies is applied for solar radiation zoning and boundary identification in China, thus generating the dataset on the boundaries of hierarchical solar radiation zones in China.

2 Metadata of the Dataset

The metadata of the Boundary dataset based on two algorithms of the hierarchical solar radiation zones in China are summarized in Table 1.

Table 1  Metadata summary of the Boundary dataset based on two algorithms of the hierarchical solar radiation zones in China

3 Methods

3.1 Data Sources

The raw data used in the dataset include ground station observations and remote sensing inversion products of surface solar radiation. The ground measurements are obtained from the monthly value dataset provided by the Meteorological Data Center of the China Meteorological Administration[1], which provides monthly average of the total solar radiation in units of 0.01 MJ/m² at 98 solar radiation stations over the period from 2007 to 2020. Specific procedures for quality control of the raw data include automatic checks of physical thresholds, temporal continuity, and time series consistency, as well as additional manual checks of automatically identified error records. More stricter quality control needs to be implemented during the production of the dataset through comparison to the reconstructed data based on remote sensing inversions^[22], i.e., data records with absolute difference ratios (divide the difference between measured and reconstructed values by measured values) greater than 20% are removed and then filled in by the multi-year average.

Remote sensing inversion products used in this study come from research of Jiang et al.^[2], which developed a deep learning algorithm to retrieve the surface solar radiation from geostationary meteorological satellite images. The algorithm relies on convolutional neural networks to process the spatial proximity effect of solar radiation transmission^[23], which significantly improves the inversion accuracy of surface solar radiation. Dataset during the 2007 to 2020 have been published^[24] and are freely available, providing monthly averages of total surface solar radiation at 0.05?? ?? 0.05?? resolution in China. Validation of data based on ground-based solar radiation stations shows that the root mean square error of the monthly average is about 1.49 MJ/m^2[25], which is better than other widely used products such as CERES-EBAF and GEWEX-SRB^[22].

3.2 Data Processing

Solar radiation zoning mainly depends on cluster analysis, which is the process of classifying a dataset into different classes or clusters, where objects in the same cluster display great similarities while objects between different clusters have great dissimilarities. Solar radiation zones in the dataset are identified through a clustering method based on Gaussian mixture model (GMM), which can smoothly approximate the density distribution of arbitrary shapes. Our previous study^[18] examined the feasibility in solar radiation zoning using finite GMM and infinite GMM. Specifically, the finite GMM generates zoning results under a predefined zone number like the K-Means method while the infinite GMM model can determine the appropriate number of zones adaptively with its Bayesian properties. The results manifest that the finite GMM model is prone to overfitting effects during the zoning process, while the infinite GMM can effectively avoid overfitting and ensure that the zoning results fully reflect the inherent characteristics of solar radiation in different regions^[18].

Therefore, the infinite GMM is adopted as a clustering algorithm in the dataset to implement hierarchical identification of solar radiation zones in China. The implementation process of the whole algorithm (as in Figrue 1) includes the following main steps. First, taking the multi-year (2007?C2020) monthly average of global solar radiation measured by ground solar radiation stations as input, 98 stations in China are clustered based on GMM clustering method. Second, by inputting the spatially continuous multi-year average (2007?C2020) of global solar radiation obtained by remote sensing inversion into the fitted GMM model in step 1, the posterior probability of each pixel belonging to different clusters is predicted, and the cluster label with the maximum probability is assigned to the pixel. In this way, a spatially continuous zone map is generated. Third, since the direct zoning results of the GMM model usually have raster polygons with very small areas (containing only a few pixels or even one pixel), we further use a sieve filter to remove polygons whose number of pixels are smaller than the predefined threshold and assign the classification labels of their neighboring polygons with the largest area to them. Fourth, with the zone boundaries based on the obtained spatially continuous zone map extracted, post-processing operations is performed such as smoothing the boundaries.

3.3 GMM Algorithm

A GMM is a mixture distribution consisting of K Gaussian sub-distributions, which can be expressed as:

                                                                                      (1)

Figure 1  The process of acquiring the boundaries of hierarchical solar radiation zones in China

where x refers to the data vector of N observations (e.g., measurements from 98 stations) and each observation contains D dimensional features (e.g., 12 monthly mean global solar radiation),  ??_k, k=1, ??, K represents the mixture weights and satisfies ??_k=1, and , k=1, ??, K, denotes the probability sub-distribution. Each sub-distribution is a D dimensional Gaussian function of the following form:

                                   (2)

with mean vector  and covariance matrix . The complete GMM is parameterized by mean vectors , covariance matrices  and mixture weights . Herein, we represent these parameters by the notation: . Compared with the classical unimodal Gaussian model or nearest neighbor model, GMM exhibits better modeling capability owing to the combination of multiple Gaussian functions. Expectation-maximization (EM) which is a widely used algorithm to estimate the parameters of GMM aims to figure out the parameters that maximize the likelihood of GMM on the training data. Given the training observations , the likelihood of GMM is expressed as:

                                                                                                     (3)

EM begins with an initial  and estimates a new , such that . The new  then becomes the initial  for the next iteration. Notably, the process is repeated until a convergence threshold is reached (details can refer to the reference^[18]).

4 Data Results and Validation

4.1 Data Composition

The dataset consists of two data files: one is the boundaries of the five primary solar radiation zones; the other is the boundaries of the ten sub-zones.

4.2 Zoning Results

Figure 2 illustrates the 5 primary zones in the hierarchical solar radiation zoning system and their annual solar radiation variations. The base map is the topography in China for illustrating the geographical units involved in each zone. Zone I includes the Qinghai-Tibet Plateau, Yunnan-Guizhou Plateau, and their surrounding areas, and the monthly average global solar radiation in Zone I is always higher than that in other regions, with the maximum value occurring in June at 24.8 MJ/m² and the annual average global solar radiation of about 20.1 MJ/m², which is the result of the combined effect of high altitude, thin aerosols, and low cloudiness^[25]. Zone II includes the Junggar Basin, Tarim Basin, Inner Mongolia, and part of the Loess Plateau. The annual average global solar radiation (16.3 MJ/m²) of Zone II is in the second rank, but the absolute intra-annual variation is the largest, with the highest in June and the lowest in December (a difference of 15.6 MJ/m²). Zone III mainly covers the northern and northeastern China, and the maximum global solar radiation of Zone III appears in June, with a monthly average value of 18.4 MJ/m². The changes of Zone II and Zone III are almost synchronized, but the magnitude is different. Zone IV is located in south of the Qinling-Huaihe line, including the middle and lower reaches of the Yangtze River plain, the southern hills and the eastern part of the Yunnan- Guizhou plateau. Zone IV consists of the Sichuan Basin and the Wuling Mountains. The annual average global solar radiation of Zone IV and Zone V is 12.7 MJ/m² and 11.2 MJ/m², respectively; and their differences are mainly concentrated in autumn and winter, when there is more solar radiation in Zone IV. The highest global solar radiation of Zone IV occurs in July at 18.3 MJ/m², while that of Zone V occurs in August at 16.7 MJ/m². In particular, both Taiwan Island and Hainan

Figure 2  Maps of five primary zones in the hierarchical solar radiation zoning system and their annual variations in solar radiation

Island are involved in 2 zones. The solar radiation characteristics of the central mountain of Taiwan Island and the southwest region of Hainan Island are similar to those of Zone I, and those of other regions of the islands are similar to those of Zone IV.

As the number of zones determined by the GMM gradually increases, the 5 primary zones begin to split and produce subzones. Figure 3a presents the results of the ultimate ten subzones. The splitting process of the 5 primary zones is as follows. First, the Junggar Basin and Xilinguole Steppe are separated from Zone II; then Zone I is divided into 2 parts, i.e., Qinghai-Tibet Plateau and Yunnan-Guizhou Plateau; subsequently, Zone II continues to split and forms a new zone on the Loess Plateau; later, the Daxinganling and its western area are separated from Zone III; and finally, Zone IV is classified into 2 parts with the dividing line being close to the latitude line of approximately 23????3??N. The ten subzones are marked by two-level labels. As can be seen from Figure 3a, the spatial division of the ten subzones is basically within the boundaries of the 5 primary zones, and the boundaries between the major zones remain almost unchanged, which indicates the stability of the five primary division schemes determined by GMM. Hainan Island is still divided into two subzones, and its southwestern region is similar to the Yunnan-Guizhou Plateau in terms of solar radiation characteristics, while the other regions are closer to the southern coastal zone of the mainland. Taiwan Island is redivided into three subzones encompassing IV-A, IV-B, and I-B from north to south, indicating that its solar radiation characteristics have higher similarity to those of the middle and lower reaches of the Yangtze River region, the southern coastal region, and the Yunnan-Guizhou Plateau.

Figure 3b compares the variation of monthly average of global solar radiation in ten subzones. The global solar radiation of subzone I-B is significantly lower than that of subzone I-A, especially in summer. Their difference is primarily caused by altitude difference. Under the influence of the North Indian Ocean monsoon, the global solar radiation of subzone I-B fluctuates greatly. Influenced by summer monsoon, the solar radiation received by zone II-B in May to September is less than that in subzone II-A. The difference between subzone II-A and subzone II-C is concentrated in winter, when the solar altitude angle of subzone II-C is smaller on the whole, resulting in lower global solar radiation than subzone II-A. The east and west parts of subzone II-C are basically within the same latitudes, thus contributing to similar solar radiation characteristics. The monthly average of global solar radiation of subzone III-A is lower than that of subzone III-B in summer but higher in winter. The global solar radiation of subzone IV-B is higher than subzone IV-A, especially in autumn and winter. Zone IV (Sichuan basin and its surroundings) always has the lowest global solar radiation among all subzones due to frequent cloudy and rainy weathers. It is worthy to note that high mountains in the northwestern of the Sichuan basin block the southeastern monsoon that carries water vapor, making the topographic rain prevail in the basin. Meanwhile, the lower elevation of the basin relative to the surrounding areas constitutes an enclosed space where evaporated water vapor collects to form rainfall. Therefore, perennial fog and cloudy weather reduce the amount of solar radiation received in the Sichuan Basin.

4.3 Data Validation

The validation of solar radiation zoning results is a challenging task considering no real zone labels are available for references. Herein, we make use of the solar radiation dataset based on sunshine hours^[26] to manually divide 716 meteorological stations. From the analysis of the results in Section 4.2, it can be tentatively determined that the GMM zoning is mainly based on the total amount of global solar radiation and the difference in seasonal variations.

Figure 3  Maps of the subzones in the hierarchical solar radiation zoning system and their annual variations in solar radiation

Afterwards, we manually classify them based on corresponding rules and then compare the

classification results with the results of GMM. The manual classification process is as follows: 1) Group the 716 stations using the boundaries in Figure 3a and calculate the centroid of each group according to the annual mean and monthly variation of solar radiation, respectively; 2) Calculate the Euclidean distance from each station to the centroid of the zone to which it belongs using the Dynamic Time Warping (DTW) algorithm; 3) Calculate the distance from each station to the centroid of its nearest neighbor employing the DTW algorithm; 4) If the distance of a station to the centroid of its zone is greater than the distance to the centroid of its nearest neighbor zone, the station is reassigned the label of its nearest neighbor zone. The final comparison is depicted in Figure 4, where 642 of the 716 sites (circles) remain in its original zone, and only 74 sites (purple symbols) are reassigned to a new zone. This result indicates that the GMM basically divides the meteorological stations into correct solar radiation zones and that the boundaries determined depending on spatially continuous solar radiation from remote sensing inversions are relatively reliable. We also note that most of the stations whose zones are changed fall inside the spatial extent of the ten subzones rather than at the edges, suggesting that zoning based on spatially continuous solar radiation can effectively avoid the influence of individual unrepresentative stations.

5 Discussion and Conclusion

The dataset provides the results of the hierarchical system with 5?C10 solar radiation zones in China. The experiments based on ground observations or spatially continuous estimates of solar radiation support the establishment of a hierarchical zoning system. The 5 primary zones

Figure 4  Map of manually zoning results of 716 meteorological stations in China based on solar radiation data derived from sunshine hours (Note: The symbol of the station indicates that the point is reclassified to the zone behind the label rather than the zone corresponding to its location.)

mainly reflect the differences in solar radiation caused by diverse climates, while the 10 subzones reveal the subtle differences in seasonal variation of solar radiation due to local microclimate and topographic influences. In China, the basic geographical units with distinct climatic characteristics, such as Sichuan basin, Qinghai-Tibet plateau, Yunnan-Guizhou plateau, and Loess plateau, have formed independent solar radiation zones. It is observed that there are some similarities between the 10 subzones and the building climate zones. For example, the boundaries between III-A and IV-A, IV-A and IV-B, and II-A and II-C-1 are basically consistent with the building climate zone boundaries^{[14, 27, 28]}, indicating that the 10 subzones also reflect the differences in surface heat distribution associated with differences in solar radiation.

The dataset identifies the precise boundaries of adjacent zones using spatially continuous solar radiation estimates, overcoming the shortcomings of the station-based zoning method. For instance, considering that Hainan Island and Taiwan Island have obvious differences in internal solar radiation characteristics, it is crucial to select an appropriate zoning scheme to divide them into different zones (as shown in Figures 2a and 3a). It is likely that Hainan Island and Taiwan Island would be considered a single zone if actual division were solely based on one or a few observation stations within them. Spatially continuous estimates can distinguish the nuances of solar radiation and therefore provide accurate pixel-level zoning results, which are important for some specific applications, such as assessing the potential of rooftop solar PV in mountainous areas to inform energy sector decisions^{[5, 6]}. Overall, it can be concluded that different solar energy utilization strategies should be implemented in the northern and southern of Hainan Island and the western and eastern of Taiwan Island.

Author Contributions

Jiang, H. designed the algorithm that generates the dataset. Yao, L. contributed to the data processing and analysis. Jiang, H. wrote the data paper.

Conflicts of Interest

The authors declare no conflicts of interest.

References

[1]      Ding, Z. L. Carbon neutral framework roadmap study in China [J]. China Industry & Information Technology, 2021(8): 54?C61.

[2]      Jiang, H., Lu, N., Qin, J., et al. A deep learning algorithm to estimate hourly global solar radiation from geostationary satellite data [J]. Renewable and Sustainable Energy Reviews, 2019, 114: 109327.

[3]      Kabir, E., Kumar, P., Kumar, S., et al. Solar energy: potential and future prospects [J]. Renewable and Sustainable Energy Reviews, 2018, 82: 894?C900.

[4]      Zhang, Y., Ren, J., Pu, Y., et al. Solar energy potential assessment: a framework to integrate geographic, technological, and economic indices for a potential analysis [J]. Renewable Energy, 2020, 149: 577?C586.

[5]        B??dis, K., Kougias, I., Jäger-Waldau, A., et al. A high-resolution geospatial assessment of the rooftop solar photovoltaic potential in the European Union [J]. Renewable and Sustainable Energy Reviews, 2019, 114: 109309.

[6]      Sweerts, B., Fenninger, S., Yang, S., et al. Estimation of losses in solar energy production from air pollution in China since 1960 using surface radiation data [J]. Nature Energy, 2019, 4(8): 657?C663.

[7]      Mensour, O. N., Ghazzani, B. E., Hlimi, B., et al. A geographical information system-based multi-criteria method for the evaluation of solar farms locations: a case study in Souss-Massa area, southern Morocco [J]. Energy, 2019, 182: 900?C919.

[8]      Lin, A. L. Research on assessment and planning of available space of urban solar applications [D]. Harbin: Harbin Institute of Technology, 2020.

[9]      Walsh, A., C??stola, D., Labaki, L. C. Review of methods for climatic zoning for building energy efficiency programs [J]. Building and Environment, 2017, 112: 337?C350.

[10]   Wan, K., Li, D., Yang, L., et al. Climate classifications and building energy use implications in China [J]. Energy and Buildings, 2010, 42(9): 1463?C1471.

[11]   Wang, B. Z. Solar energy resource division in China [J]. Acta Energiae Solaris Sinica, 1983(3): 221?C228.

[12]   Al-Azri, N. A., Zurigat, Y. H., Al-Rawahi, N. Z. Development of bioclimatic chart for passive building design [J]. International Journal of Sustainable Energy, 2013, 32(6): 713?C723.

[13]   Zhou, Y., Wu, W., Hu, Y., et al. The temproal-spatial distribution and evaluation of potential solar energy resources in Northwest China [J]. Journal of Natural Resource, 2010, 25(10): 1738?C1749.

[14]   Lau, C. C. S., Lam, J. C., Yang, L. Climate classification and passive solar design implications in China [J]. Energy Conversion and Management, 2007, 48(7): 2006?C2015.

[15]   Liu, Y. F., Yong, Z., Wang, D., et al. Classification of solar radiation zones and general models for estimating the daily global solar radiation on horizontal surfaces in China [J]. Energy Conversion and Management, 2017, 154: 168?C179.

[16]   Jiang, H., Lu, N., Qin, J., et al. Hourly 5-km surface total and diffuse solar radiation in China, 2007?C2018 [J]. Scientific Data, 2020, 7(1): 311.

[17]   Antoniak, C. E., Mixtures of dirichlet processes with applications to Bayesian nonparametric problems [J]. The Annals of Statistics, 1974, 2(6): 1152?C1174.

[18]   Jiang, H., Lu, N., Qin, J., et al. Hierarchical identification of solar radiation zones in China [J]. Renewable and Sustainable Energy Reviews, 2021, 145: 111105.

[19]   Soubdhan, T., Emilion, R., Calif, R. Classification of daily solar radiation distributions using a mixture of Dirichlet distributions [J]. Solar Energy, 2009, 83(7): 1056?C1063.

[20]   Jiang, H. Boundary dataset based on two algorithms of the hierarchical solar radiation zones in China [J/DB/OL]. Digital Journal of Global Change Data Repository, 2023. https://doi.org/10.3974/geodb.

2023.07.07.V1. https://cstr.escience.org.cn/CSTR:20146.11.2023.07.07.V1.

[21]   GCdataPR Editorial Office. GCdataPR data sharing policy [OL]. https://doi.org/10.3974/dp.policy.2014.05 (Updated 2017).

[22]   Zhang, X. T., Liang, S., Wild, M, et al. Analysis of surface incident shortwave radiation from four satellite products [J]. Remote Sensing of Environment, 2015, 165: 186?C202.

[23]   Jiang, H., Lu, N., Huang, G., et al. Spatial scale effects on retrieval accuracy of surface solar radiation using satellite data [J]. Applied Energy, 2020, 270: 115178.

[24]   Jiang, H., Lu, N. High-resolution surface global solar radiation and the diffuse component dataset over China [DB/OL]. Pangaea, 2019. https://doi.pangaea.de/10.1594/PANGAEA.904136.

[25]   Jiang, H., Yang, Y., Wang, H., et al. Surface diffuse solar radiation determined by reanalysis and satellite over East Asia: evaluation and comparison [J]. Remote Sensing, 2020, 12: 1387.

[26]   Tang, W. J. Daily mean solar radiation dataset of 716 weather stations in China (1961?C2010) [DB/OL]. National Tibetan Plateau Data Center, 2015. https://doi.org/10.11888/AtmosphericPhysics.tpe.249399.file.

[27]   Wan, K. K. W., Yang, H. L., Yang, L., et al. An analysis of thermal and solar zone radiation models using an Angstrom?CPrescott equation and artificial neural networks [J]. Energy, 2008, 33(7): 1115?C1127.

[28]   Bai, L., Yang, L., Song, B., et al. A new approach to develop a climate classification for building energy efficiency addressing Chinese climate characteristics [J]. Energy, 2020, 195: 116982.