1. Introduction
Climate change has drawn increasing attention from governments, the public, and the media due to the rapid global warming that has occurred over the past century and its profound influence on human society. The Sixth Assessment Report (AR6) of Working Group I of the Intergovernmental Panel on Climate Change (IPCC) concluded that “Global surface temperature was 1.09 [0.95 to 1.20] °C higher in 2011–2020 than 1850–1900” (IPCC 2021, p. 5). “It is unequivocal that human influence has warmed the atmosphere, ocean, and land” (IPCC 2021, p. 4). Hegerl et al. (2019) reviewed the causes of climate change from the nineteenth century, identifying that greenhouse gases have been the main driver of global warming since the beginning of industrialization. The cooling effects from anthropogenic aerosol forcing are likely to have masked some warming since the post-1950s when sulfate aerosols emissions increased. The combined effects of solar and volcanic forcings also contribute to surface temperature changes, largely within a shorter period (Shiogama et al. 2006; Cole-Dai et al. 2009; Hegerl et al. 2018). The internal variability of climate systems causes multidecadal and/or interannual variations in temperature (Brönnimann 2009; Tung and Zhou 2013; Staten et al. 2018). Gillett et al. (2021) recently separated the relative contribution of greenhouse gases and anthropogenic aerosols to global temperature from the preindustrial period based on models that participated in phase 6 of the Coupled Model Intercomparison Project (CMIP6; Eyring et al. 2016). They demonstrated that anthropogenic global temperature change may already be close to the 1.5°C warming above preindustrial level that has been considered as a key threshold to limit global warming in the Paris Agreement.
At the regional scale, possible causes of temperature changes since the mid-twentieth century have been widely investigated. Since the 1960s, China has experienced rapid warming at a rate almost 2 times that of the global mean change. The annual mean near-surface temperature has increased by 1.44°C over the period 1961–2013, which can be attributed to the greenhouse gases (GHGs), other anthropogenic forcing (dominated by aerosols), and urban heat island effects (Sun et al. 2016). Among these factors, GHGs explained about 1.24°C of observed warming, while the cooling due to other anthropogenic forcings is 0.43°C. Xu et al. (2015) and Zhao et al. (2016) further confirmed the dominant role of anthropogenic forcings in temperature change. At a smaller regional scale in western China, the increase in annual mean temperature may also be attributed to human influence (Wang et al. 2017). On the other hand, anthropogenic forcing also leads to an increase in warm extremes and a decrease in cold extremes, including their intensity, frequency, and duration (Lu et al. 2016; Yin et al. 2017). These previous studies have detected clear anthropogenic signals in the observed mean and extreme temperature changes in China since the 1950s.
Limited data availability has meant that these previous studies mainly focused on climate change after 1950s in China. The recovery of historical datasets and an abundance of climate reconstruction data have enabled the development of several sets of observational products to investigate temperature changes in China since the 1900s. Based on the most recent homogenized dataset, Li et al. (2017) indicated that surface air temperatures increased by approximately 0.13°C decade−1 for 1900–2015 based on CRUTEM4 observational data. The newly released Blue Book on Climate Change in China (CMA Climate Change Centre 2021) shows that mean temperatures in China have increased by 0.15°C decade−1 from 1901 to 2020. Discrepancies between different observational datasets largely occur in the first half of the twentieth century because of differences in data processing (Parker 2010; Li et al. 2010, 2017; Xu et al. 2018); however, most surface air temperature series in China based on homogenized data have revealed similar warming rates since the 1900s. Moreover, CMIP5 model simulations have well reproduced the trend, temporal evolution, and spatial distribution in observational temperature records (Zhang et al. 2016; Li et al. 2017). The influence of external forcings on temperature change before 1950s in China has received increasing attention with greater data availability from earlier historical periods. A few detection studies focusing on annual and seasonal temperature changes since 1900 show that weakening temperature seasonality since the late nineteenth century in some regions (e.g., northeastern Asia and the Tibetan Plateau) may be attributed to anthropogenic forcing (Duan et al. 2019). However, there are large uncertainties for many aspects, such as the quality of instrumental and proxy data, the complicated feedbacks of the climate system when simulating regional climate change, and estimating the aerosol contribution to climate change (Ren 2008; Knutson et al. 2018).
The detection and attribution of climate change at the regional scale over an extended historical period continues to be highly limited. Quantifying the contribution of different forcing factors to temperature change since the preindustrial period is important for simulating reliable future projections and developing appropriate climate change policy under different warming targets. This study analyzes annual, summer, and winter temperature changes based on four observational datasets. Then, changes based on these datasets are compared with models from CMIP5 (Taylor et al. 2012) and CMIP6. When compared with CMIP5 models, the newest generation of CMIP6 models mostly shows improved physical and chemical processes, as well as higher resolution. The Detection and Attribution Model Intercomparison Project (DAMIP; Gillett et al. 2016) of CMIP6 designed a series of historical simulations driven by aerosols-only, stratospheric ozone-only, GHG-only, solar-only, and volcanic-only forcings. These experiments provide a unique opportunity to investigate the effects of individual external forcings on climate change. With these data, we aim to investigate temperature changes in China since the preindustrial period under different forcings, and understand the intermodel difference of temperature response. Considering the data limitations at the century scale, the effects of urbanization heat island are not included in this study, unlike the study by Sun et al. (2016). This may lead to differences in the attribution results from previous studies, which is discussed in the final section. The remainder of this paper is organized into three additional sections. Section 2 describes the datasets and analysis methods, while section 3 presents the main results, and section 4 provides the discussion and conclusions.
2. Data and methods
a. Observations
This study compared four observational temperature datasets: 1) The China-Land Surface Air Temperature (CLSAT v1.3; https://climexp.knmi.nl/select.cgi?id=someone@somewhere&field=clsat_tavg); 2) the Climatic Research Unit temperature (CRUTEM 4.6; https://www.metoffice.gov.uk/hadobs/crutem4/data/download.html); 3) the Goddard Institute for Space Studies surface land temperature (GISTEMP v4; https://data.giss.nasa.gov/gistemp/); and 4) the Global Historical Climatology Network (GHCN v4.0.1; https://climexp.knmi.nl/select.cgi?id=someone@somewhere&field=tempa). The CLSAT v1.3 (CLSAT hereinafter) dataset developed by the China Meteorological Administration is homogenized monthly surface air temperature anomalies (relative to the 1961–90 reference period) across the world since 1900 (Xu et al. 2018); the dataset has a spatial resolution of 5° longitude × 5° latitude. CLSAT contains 12 374 stations and shows improved data coverage when compared with the other global datasets (i.e., CRUTEM4 and GHCNv3) in most Asian countries, particularly China and its neighboring regions. CRUTEM 4.6 (CRU hereinafter) provides global historical near-surface air temperature anomalies over land at a spatial resolution of 5° longitude × 5° latitude. Note that the observational data are subject to uncertainty (e.g., CRU provided an error model range to describe such uncertainty) (Jones et al. 2012). This dataset was collaboratively developed by the Met Office Hadley Centre and the Climatic Research Unit at the University of East Anglia (Jones et al. 2012). The data were based on an archive of monthly mean temperatures provided by more than 5500 weather stations distributed worldwide. The gridded value is based on the average of temperature anomalies relative to the 1961–90 mean within each grid box. GISTEMPv4 (GISTEMP hereinafter) global land surface temperature dataset at a spatial resolution of 2° longitude × 2° latitude is established by the Goddard Institute for Space Studies of the National Aeronautics and Space Administration based on GHCN-M version 4. (Hansen et al. 2010; Lenssen et al. 2019; GISTEMP Team 2022). The GHCN v4.0.1 (GHCN hereinafter) monthly temperature dataset has a spatial resolution of 5° longitude × 5° latitude and is developed by the National Center for Environment Information of the National Oceanic and Atmospheric Administration (Menne et al. 2018). All these observational data were regridded onto grid boxes at the same 5° longitude × 5° latitude resolution for comparison. Linear trends in annual and seasonal temperature series were estimated by ordinary least squares regression, with confidence interval of observed temperature trend estimated as ±1.96σ, where σ represents the standard error of observation data (Wilks 2011).
Figure 1 shows the available number of grid boxes in China for the four observational datasets. For the 1901–50 period, most data were collected in eastern China (east of 105°E), mainly along coastal areas (Figs. 1a–d). In most parts of western China (west of 105°E), the length of available data in the grid boxes was generally less than 20 years, and there were no data for some areas before the 1950s. It was apparent that the CLSAT dataset had more available data than the other datasets; this was clearly indicated by the change in the number of grid boxes (Fig. 1e). The number of grid boxes was less than 30 for the pre-1920s and then gradually increased and stabilized at 55 across China.
Maps representing how many years of data are available in each grid point for each dataset during 1901–50: (a) CLSAT, (b) CRU, (c) GISTEMP, and (d) GHCN. (e) The number of available grid boxes from different datasets over China. The grid boxes for 1910 are marked by a cross (24 grid boxes).
Citation: Journal of Climate 36, 7; 10.1175/JCLI-D-22-0122.1
b. Models
CMIP5 and CMIP6 multimodel simulations were used to compare observations and estimate the temperature response to external forcing and internal variability in the detection analysis. Tables 1 and 2 list details of model simulations used in this study; 111 simulations from 18 CMIP5 models and 125 simulations from 26 CMIP6 models were used under the combined anthropogenic and natural forcings (ALL), respectively. Under GHG forcing, there were 34 simulations from 7 CMIP5 models and 59 simulations from 11 CMIP6 models; under natural-only (NAT) forcing, there were 47 simulations from 10 CMIP5 models, and 57 simulations from 11 CMIP6 models. Separate aerosol forcing experiments (AA) were based on 54 simulations from 11 CMIP6 models. For CMIP5, most simulations ended in 2012, and some historical simulations that ended in 2005 were extended to 2012 using the representative concentration pathway 8.5 (RCP8.5) trajectory. This was done because there was little difference in the early twenty-first century under varying external forcing simulations (IPCC 2013). For CMIP6, all historical simulations were extended with the Shared Socioeconomic Pathways 4.5 (SSP2–4.5) simulations for the 2015–18 period such that data covers the 1901–2018 period. We assumed that responses to anthropogenic and natural forcings combine linearly (Gillett et al. 2004; Marvel et al. 2015) and estimated that anthropogenic forcing (ANT) responses are the difference between the ALL and NAT forcing simulations. Additionally, 190 112-yr chunks of preindustrial control (CTL) simulations from 35 CMIP5 models were used to evaluate internal climate variability during 1901–2012, whereas 366 62-yr chunks of CTL simulations from 35 CMIP5 models were used to evaluate internal climate variability during 1951–2012. Likewise, 109 118-yr chunks of CTL simulations from 20 CMIP6 models were used to evaluate internal climate variability during during 1901–2018, whereas 198 68-yr chunks of CTL simulations from 21 CMIP6 models were used to evaluate internal climate variability during 1951–2018. Because of different resolutions of different climate models, all of the model simulations were gridded onto the same 5° × 5° grid boxes as the observation. All the model simulations were masked by the available observational data. Regional means series over China were calculated based on these masked data.
List of multimodel simulations from CMIP5 used in this study. Numbers represent the ALL, GHG, and NAT simulation ensemble sizes or the number of CTL simulations.
List of multimodel simulations from CMIP6 used in this study. Numbers represent the ALL, GHG, NAT, and AA simulation ensemble sizes or the number of CTL simulations.
c. Detection method
We compared the spatiotemporal evolution of observed temperatures in China with model simulations using an optimal fingerprinting technique (Allen and Stott 2003; Ribes et al. 2013). The method regresses the observations Y onto the signals X scaled by scaling factors β: Y = (X − υ)β + ε. Here, υ indicates noise in model simulation signals X; ε indicates residual of the regression, typically assumed to have the same covariance structure (or internal variability) as those in model simulations. The scaling factors β and their uncertainty were estimated based on the total least squares method (Ribes and Terry 2013). When the 90% range of β was above zero, this suggests that the external forcings signal is detectable; when the 90% range of β is above zero and includes unity, this means that observed changes are consistent with model simulated responses to external forcing.
For both observations and model results, the regional mean time series of mean temperature was calculated over China as a whole using area weighting. The multirun mean temperatures were first calculated for each model and then the multimodel ensemble mean were obtained to estimate the model response to different forcings. The detection and attribution analysis were conducted on the non-overlapping 5-yr mean time series of China regional mean temperature. This means the spatial dimension is reduced to one-region analysis and the temporal dimension is reduced to 14 data values for the period 1951–2018 and 24 data values for the period 1901–2018, with the last point having 3-yr data. To estimate the noise, we used the data from the CTL and within-ensemble differences of simulation. For CMIP5, we used 390 112-yr chunks (190 chunks of CTL and 200 chunks of within-ensemble differences of simulation) during 1901–2012 and 784 62-yr chunks (366 chunks of CTL and 418 chunks of within-ensemble differences of simulation) during 1951–2012. For CMIP6, we used 396 118-yr chunks (108 chunks of CTL and 288 chunks of within-ensemble differences of simulation) during 1901–2018 and 774 68-yr chunks (198 chunks of CTL and 576 chunks of within-ensemble differences of simulation) during 1951–2018. The within-ensemble differences and CTL were split into two independent sets, which were used to optimize analysis to obtain the best estimates and confidence intervals of the scaling factors. A residual consistency test (Ribes et al. 2013) was performed to examine whether internal variability in model simulation is consistent with the regression residual. The residual consistency test implementation used a nonparametric estimation of the null distribution and was conducted at the 10% level. If residual consistency test value is under the 10% threshold, the assumption of consistency of observed and simulated internal variability did not pass.
We conducted single-, two-, and three-signal regression analyses to assess the relative contribution of external forcings to observed changes. In the former, observed changes were regressed onto the multimodel mean responses to individual external forcings or sets of forcings. This approach enables an estimate of whether model simulated responses to an individual or combined forcing (e.g., the ALL signal) may be detected in the observation. In the two-signal analyses, observed changes were simultaneously regressed onto the ANT and NAT responses; the objective was to estimate whether the ANT and NAT signals may be detected separately in observed changes. For the three-signal analyses, observed changes were simultaneously regressed onto the GHG, AA, and NAT responses to determine whether the three external forcing signals may be separated and detected in the analyzed data. It is worth noting that the GHG, AA, and NAT experiments from CMIP6 leave out the forcing factors of tropospheric ozone and land cover changes, both of which have moderate radiative forcing influences.
3. Results
a. Observed temperature changes
We compared the long-term variation and trend of temperature anomalies (relative to the 1961–90 average) using the four observed temperature datasets in China (Fig. 2). The annual, summer, and winter mean temperatures from the four datasets were similar in terms of warming since 1901; the largest warming was observed in winter and the lowest occurred in summer. The linear trends of annual, summer, and winter temperatures in China from the four datasets ranged from 0.12°, 0.07°, and 0.15°C decade−1 to 0.14°, 0.09°, and 0.17°C decade−1 for the 1901–2018 period, respectively. Slight differences among datasets occurred in the early pre-1950s period, mainly due to different coverage of raw data and/or the gridding methods (Li et al. 2017). During the 1951–2018 period, the warming in annual, summer, and winter temperatures was larger than that in the 1901–2018 period. The annual, summer, and winter temperature trend ranged from 0.23°, 0.18°, and 0.29°C decade−1 to 0.25°, 0.18°, and 0.33°C decade−1, respectively. These changes were almost 2 times as large as those in the early half century, indicating rapid warming in the latter half century. These warming rates were consistent with a warming of ∼0.13°C per decade and ∼0.24°C per decade for annual mean temperatures reported in previous studies during the 1900–2015 and 1951–2015 periods (Li et al. 2017), and the warming up to 2020 (CMA Climate Change Centre 2021). CLSAT was selected for comparison with model simulations and for attribution analysis because of its better coverage (Xu et al. 2018) and the high consistency of warming in China from the different observational datasets.
Time series of (a) annual, (b) summer [June–August (JJA)], and (c) winter [December–February (DJF)] temperature anomalies (relative to the 1961–90 average), and (d) linear trends of the temperature anomalies during the 1901–2018 period (°C), averaged over China, from the CLSAT, CRU, GISTEMP, and GHCN observational datasets. The whiskers in (d) show the 90% confidence intervals (black error bars) of linear trends of observed temperatures, which were estimated by standard error of regression coefficient (Wilks 2011). Note that a different scale is used for (c).
Citation: Journal of Climate 36, 7; 10.1175/JCLI-D-22-0122.1
Given the poor data availability in the first half of the twentieth century, only grid boxes with data available since the early period (we chose 1910) could be selected to estimate the national mean temperature change for 1901–2018. This means that only data (24 grid boxes for 1910; Fig. 1a) mainly from eastern China were used to estimate warming in China for the 100-yr time series. To investigate the impact of data availability on temperature change, two tests were conducted: 1) the CMIP6 model-simulated historical temperature change were compared based on all grid boxes in China and grid boxes available in 1910 (Fig. 3); and 2) for the CLSAT and CRU, observed temperature changes were compared based on all the available data across China for 1901–2018 and available data at grid boxes across China in 1910 (24 grid boxes mostly located in eastern China). For test 1, the two time series corresponded well for annual and summer temperatures, whereas they slightly differed for winter temperatures. For test 2, overall, the two data series appeared to show similar changes; linear trends for annual temperature from CLSAT and CRU data over China (the CLSAT_whole and CRU_whole datasets shown in Fig. 3) and over grid boxes in 1910 (the CLSAT_1910 and CRU_1910 datasets shown in Fig. 3) were 0.13° and 0.12°C per decade, and 0.13° and 0.13°C per decade, respectively. Although there were some differences in the early period and for winter temperatures, this influence on the trend estimate was relatively small. These comparisons show that although we cannot obtain real temperature series for the past century because of sparse observational data and model limitations, the grid boxes in 1910 demonstrate relatively good representativeness of data time series. In the following analyses, we used time series based on 24 grid boxes (18 grid boxes in eastern China) for the analysis over the 1901–2018 period; for the analysis over the 1951–2018 period, the time series was based on 55 grid boxes across China. All model results were masked by available observational data for corresponding periods. The comparison between the models and observations is discussed in the next section.
Time series of annual, summer (JJA), and winter (DJF) mean temperature anomalies (relative to the 1961–90 average) in China from CLSAT and CRU observational data and the CMIP6 multimodel mean simulation under ALL forcing. CLSAT_whole, CRU_whole, and ALL_whole represent observation and ALL forcing simulations across all of China. CLSAT_1910, CRU_1910, and ALL_1910 represent observation and ALL forcing simulations in available grids for 1910.
Citation: Journal of Climate 36, 7; 10.1175/JCLI-D-22-0122.1
b. Comparison between observation and models
Figures 4 and 5 display the spatial distributions of the observed and model-simulated trends in annual, summer, and winter temperatures during 1901–2018 and 1951–2018 based on CLSAT and CMIP6 models, respectively, under ALL, GHG, AA, and NAT forcings. The temperature change from CMIP5 were similar to those from CMIP6 (data not shown). During 1901–2018, increasing temperature trends in annual, summer, and winter from CLSAT datasets occurred almost everywhere, with the exception of slight cooling in parts of central China. The largest warming trends were observed in northern China, which is located in the mid- to high latitudes, and the Tibetan Plateau; this is consistent with previous findings (Zhou et al. 2016; Yin et al. 2019). The warming in winter was the strongest with a rate of approximately 0.17°C per decade, which was faster than warming in the annual and summer mean temperatures. After 1950s, more apparent and pronounced warming was observed across China. The increasing trend in northern China was larger than that in the early half century, suggesting a strengthening of warming with increasing greenhouse gas emissions. The cooling zone around southwestern China disappeared and was replaced by a small warming area.
Geographic distribution of (top) annual, (middle) summer (JJA), and (bottom) winter (DJF) temperature change trends for the 1901–2018 period (°C decade−1) from (a) CLSAT dataset (OBS) and the CMIP6 multimodel response to the (b) ALL, (c) GHG, (d) AA, and (e) NAT forcings. Only grid boxes with more than 25 years of all months of data available for the 1901–50 period are shown.
Citation: Journal of Climate 36, 7; 10.1175/JCLI-D-22-0122.1
Similar to Fig. 4, but for the 1951–2018 period.
Citation: Journal of Climate 36, 7; 10.1175/JCLI-D-22-0122.1
These spatiotemporal changes were generally captured by the CMIP6 model experiments under ALL forcing. The models were able to reproduce the heterogeneous spatiotemporal features of warming, showing stronger warming in the mid- to high latitudes and high mountain areas, and in the second half of the twentieth century. However, the ALL forcing simulations slightly underestimated the magnitudes of observed temperature trends in some areas, particularly in winter. The GHG forcing simulation on average exhibited stronger warming trends than the ALL simulations. The largest warming in the mid- to high latitudes and seasonal differences between warming in winter and summer were all captured by the GHG results; this indicates the dominant role of GHG forcing on temperature change in the past century. The AA results showed clear cooling effects, offsetting part of the warming trend induced by GHG, with a rate of −0.10°C per decade in most areas. In the AA experiments, the largest cooling areas were observed in southeastern China, where large amounts of aerosols were emitted (Zheng et al. 2012). For the NAT experiments, the results in both periods exhibited minor cooling trends, indicating a small influence from solar forcing and volcanic eruptions.
Figure 6 shows the temporal change in annual, summer, and winter temperature averaged in China for the 1901–2018 period; there was a clear upward increase in all the mean temperatures, particularly after the late 1950s, and rapid warming was observed after the 1980s. The linearly increasing trends in annual and summer temperatures (0.13° and 0.08°C decade−1) were lower than those in winter (0.17°C decade−1), with a larger variation in winter than other seasons. The results from the CMIP5 and CMIP6 models under ALL forcing were consistent with the observed summer temperature, capturing long-term changes with similar amplitudes. However, for annual and winter temperatures, both generations of models under ALL forcing underestimated the magnitude of change. This underestimate of warming by CMIP5 models in China was also reported in a previous study (Sun et al. 2016), where urbanization effects were considered to be an important factor contributing to this difference. In addition, the internal variability of climate system, model parametric and structural uncertainties may lead to the differences between observation and model simulation (Lee et al. 2021). The GHG forcing simulations showed a large increase, similar to observations for annual and winter temperatures, while the AA experiments displayed a cooling effect. It was apparent that the GHG-induced warming was larger than the AA-induced cooling; for example, for annual mean temperature, there was a linear trend of 0.16°C decade−1 in the GHG experiments as opposed to −0.10°C decade−1 in the AA experiments. The NAT simulations did not exhibit clear linear trends when compared with the temperature response to other forcings, suggesting the lesser role of NAT forcings on long-term temperature change in China.
(left) Time series of (a) annual mean, (b) summer (JJA), and (c) winter (DJF) temperature anomalies (relative to the 1901–30 average) and (right) linear trends averaged over China (°C) from CLSAT observational dataset (OBS; black) and the CMIP5 and CMIP6 multimodel response to the ALL (purple and red), GHG (green), AA (orange), and NAT (blue) forcings. The pink and blue shadings in the time series show the 5%–95% range of the CMIP6 ALL and NAT simulations, respectively. Vertical black dashed lines in the left panels indicate the timing of major volcanic events. Linear trends in the right panel are averaged over China for 1901–2018 and are based on CLSAT observation and CMIP6 models, with the 90% confidence intervals shown by gray error bars.
Citation: Journal of Climate 36, 7; 10.1175/JCLI-D-22-0122.1
Moreover, large volcanic eruptions with a volcanic explosivity index (VEI; Newhall and Self 1982) > 4 (Table 3) affected short-term temperature variation. Large amounts of sulfur were emitted into the stratosphere after significant volcanic eruptions, forming sulfate aerosols that shield incoming solar radiation and may cause short-term cooling of the surface temperature (Robock 2000). The annual mean temperature in China dropped by 0.89° and 0.71°C in the year after the El Chichón and Pinatubo volcanic eruptions in 1982 and 1991, respectively. A greater temperature decline occurred in the year following the volcanic eruption than during the eruption year, which is consistent with previous studies (Zhang and Zhang 1985; Liang et al. 2009; Yin et al. 2021).
Information on major volcanic eruptions since 1901. VEI is the volcanic explosivity index (Newhall and Self 1982), information on the volcanic events is from Robock (2000), and Lag−1 T indicates the annual temperature change with a 1-year lag relative to the volcanic eruption year.
c. Detection results
To detect the temperature response (fingerprint) to different external forcings, we conducted single- and two-signal detection for the 1901–2018 and 1951–2018 periods based on observed and model simulated non-overlapping 5-yr time series. The detection results based on CMIP5 and CMIP6 were similar; as such, only the CMIP6 results were presented in this paper. For single-signal analyses (Fig. 7), the ALL forcing signal was strongly detected in annual, summer, and winter temperature changes in the two analyzed periods commencing from 1901 and 1951. For annual and winter temperatures, the 90% confidence interval of scaling factors exceeded unity, indicating a detectable human influence, although the model underestimated the observed warming. For summer temperatures, the 90% confidence intervals of scaling factors in both analyzed periods were approximately unity, indicating good consistency between observed and simulated temperatures. The residual consistency tests were passed for annual and summer temperature detection in 1951–2018 but not for the winter temperature; winter experienced a smaller variability in the models than the observation. This also indicates the reduced reliability of winter temperature detection relative to annual and summer temperatures, which was partly due to the large variability in observed winter temperature. Basically, these detection results are consistent with the qualified relations between the observation and models as indicated in Fig. 6.
Best estimates of scaling factors and their 5%–95% confidence intervals from single-signal analyses for ALL forcing and two-signal analyses for ANT and NAT forcings in annual, summer (JJA), and winter (DJF) temperatures for the (a) 1901–2018 and (b) 1951–2018 periods, based on CMIP6 simulations for China. The triangles indicate a failure of the residual consistency test as a result of the residual not having the expected variability as assumed from models, where upward triangles indicate that model simulations underestimate the observed variability according to the residual consistency test.
Citation: Journal of Climate 36, 7; 10.1175/JCLI-D-22-0122.1
For the two-signal analyses (Fig. 7), the ANT signal was detected in annual, summer, and winter temperatures for the two analyzed periods, whereas the NAT signal could not be detected for most cases, except for summer temperature during 1901–2018. Similar to the ALL forcing results, the best estimates of scaling factors for ANT forcing were greater than unity for annual and winter temperatures and close to unity for summer temperature. The NAT results reflect the influence of solar and volcanic signals on summer temperatures and were detected at a longer time scale but showing larger uncertainty range. Generally, the models passed the residual consistency tests for annual, summer, and winter temperature for 1951–2018, but failed for annual and winter in 1901–2018. The smaller variability (upward triangles in Fig. 7) in the models than in observations was evident for those results without residual consistency passed. These results are consistent with single-signal detection as shown in Fig. 6, confirming the anthropogenic influence on temperature change over 1901–2018.
Figure 8 shows the results of the three-signal analyses for the 1901–2018 and 1951–2018 periods. The GHG signal was detected for the annual, summer, and winter temperatures for the two periods commencing from 1901 to 1951, indicating the dominant role of GHG forcing. Generally, the confidence intervals of scaling factors for the second half of the twentieth century were larger than those for the 1901–2018 period. The AA signals were detectable for observed temperature changes in annual, summer, and winter temperatures during 1901–2018 and for summer during 1951–2018, but not for winter and annual temperature during 1951–2018. This suggests low detectability of the AA signal within a short period against natural variability. This low AA signal detectability for the 1951–2018 period may also be linked to the strong colinearity between the GHG and AA results (Jones et al. 2013; DelSole et al. 2019); this may lead to difficulty in AA signal detection. In general, the residual consistency tests were passed for annual temperature during 1951–2018 and for summer during the two periods’ detection analyses, while an underestimation of model-simulated variability occurred in the winter and annual simulations, as per the single- and two-signal analyses. The NAT signal was not detectable for most cases, except for summer temperature during the period 1901–2018, which are consistent with the two-signal results shown in Fig. 7.
Best estimates of scaling factors and their 5%–95% confidence intervals from three-signal analyses based on the CMIP6 simulation (GHG, AA, and NAT) in annual, summer (JJA), and winter (DJF) from (a) 1901 to 2018 and (b) 1951 to 2018. The upward triangles indicate that model simulations underestimate the observed variability according to the residual consistency test.
Citation: Journal of Climate 36, 7; 10.1175/JCLI-D-22-0122.1
We further conducted single- and three-signal detections based on individual models (Fig. 9). In single-signal analyses, the ALL signal was detectable in 8, 10, and 10 out of 11 models for annual, summer, and winter temperature change since 1901, respectively. This means that most models demonstrated good capability in reproducing mean temperature change. The residual consistency test showed failure for 5, 2, and 8 out of 11 model simulations under ALL forcing for annual, summer, and winter temperature. These indicate that it becomes more difficult to fit the regression model once model signals are noisier. We compared the trend between ALL forcing simulation for individual models and observation and found that the models in which the ALL signal was undetectable showed a clear underestimation of observational trends (not shown). In the three-signal analyses, GHG signals were detected for annual, summer, and winter temperature change for the 1901–2018 period across all models. This highlights the robustness of the GHG influence and the dominant role of GHG forcing in temperature change. AA signals were only detected in 5, 10, and 1 out of 11 models for annual, summer, and winter temperature change, respectively; this indicates the large model difference in AA detection and the considerable uncertainty of AA influence. NAT signals were detected in annual and summer temperatures for a few models, although they were not detected in winter temperature changes. Briefly, these results indicate the dominant role of GHG emissions in observed temperature changes since 1901, while the AA influence is less important and varies from model to model. Gillett et al. (2021) quantified the contributions of individual forcings to observed trends in global mean temperature for the 1850–2019 period. In their study, the Australian Community Climate and Earth System Simulator (ACCESS-ESM1.5) model exhibited an unrealistic global mean annual temperature in the ALL simulation. Moreover, the present study showed that not only the ACCESS-ESM1.5, but also the Meteorological Research Institute Earth System Model (MRI-ESM2-0), and the Norwegian Earth System Model (NorESM2-LM) models clearly underestimate observed annual temperature change for China during the time period (not shown), which may be one reason leading to undetectable ALL signal in these models.
The 5%–95% confidence intervals of scaling factors based on individual CMIP6 model simulation response to ALL forcing under single-signal analyses, and GHG, AA, and NAT forcings under three-signal analyses for the 1901–2018 period in terms of (a) annual, (b) summer (JJA), and (c) winter (DJF) temperatures. The upward and downward triangles indicate that model simulations underestimate and overestimate, respectively, the observed variability according to the residual consistency test.
Citation: Journal of Climate 36, 7; 10.1175/JCLI-D-22-0122.1
To estimate the attributable contributions from the ALL, GHG, AA, and NAT signals to observed temperature trends, the linear trends of temperature response to different external forcings were multiplied by corresponding scaling factors estimated in the single- (for ALL) and three-signal analyses (for GHG, AA, and NAT) (Fig. 10). For the 1901–2018 period, the observed annual, summer, and winter temperatures increased by approximately 1.54°C (90% confidence interval 1.32°–1.77°C), 0.99°C (0.77°–1.20°C), and 1.95°C (1.45°–2.45°C), in which the ALL contributed to 1.51°C (1.29°–1.73°C), 0.88°C (0.68°–1.07°C), and 1.83°C (1.42°–2.23°C) of warming, respectively. The GHG contributed to 2.06°C (1.58°–2.54°C), 1.70°C (1.33°–2.08°C), and 2.88°C (2.00°–3.79°C) of warming for annual, summer, and winter temperatures, while the AA contributed to 0.45°C (0.17°–0.73°C), 0.61°C (0.41°–0.82°C), and 0.76°C (0.09°–1.46°C) of cooling, respectively. NAT was only detectable for summer temperatures, although its contribution was negligible, accounting for only ∼0.001°C of temperature change. The cooling effects of AA partially offset approximately one-quarter of GHG-induced warming for annual and winter temperatures and one-third of GHG-induced warming for summer temperatures. During 1951–2018, the observed annual, summer, and winter temperatures increased by approximately 1.70°C (1.43°–1.97°C), 1.22°C (0.94°–1.50°C), and 2.22°C (1.58°–2.85°C), respectively. ALL contributed to 1.56°C (1.27°–1.86°C), 1.26°C (0.98°C–1.54°C), and 2.03°C (1.44°–2.64°C) of warming for annual, summer, and winter temperatures, in which the GHG contributed to 1.86°C (1.26°–2.45°C), 1.75°C (1.34°–2.17°C), and 2.54°C (1.06°–4.24°C), respectively. For the CMIP6 models, AA was not detected for annual and winter temperatures, although it exhibited clear cooling effects [−0.42°C (from −0.13°C to −0.72°C)] and offset approximately one-quarter of GHG warming for summer temperatures. The NAT contribution was not detected for annual, summer, and winter temperatures.
Attributable contribution from ALL, GHG, AA, and NAT forcings to observational trend (OBS) in annual, summer (JJA), and winter (DJF) temperature for the (a) 1901–2018 and (b) 1951–2018 periods, and their 5%–95% confidence interval (error bars). The attributable contribution were estimated based on the single- and three-signal analyses. The confidence interval of the linear trends for observed temperature change was estimated by the standard errors of regression coefficients of temperature trend.
Citation: Journal of Climate 36, 7; 10.1175/JCLI-D-22-0122.1
Sensitivity tests were also conducted to evaluate the robustness of detection results. First, because we applied all the available CMIP6 models (Table 2) that have different ALL and NAT experiments, we used the same model subsets that simultaneously have both the ALL and NAT experiments to repeat all the detection analyses as shown above. The results were relatively similar based on all available model runs and the model subsets that have same ALL and NAT experiments. Second, the robustness of detection results for different observational datasets were evaluated; detection results based on the four observational temperature datasets showed roughly consistent results. Third, three-signal detections using GHG, NAT, and other anthropogenic forcings (OANT) (i.e., OANT = ALL − GHG − NAT) were jointly considered based on the CMIP5 and CMIP6 models. The main reason to do this test is because the previous studies (e.g., Sun et al. 2016) have shown a detectable OANT signal using the CMIP5 models. We found that the OANT signal was detectable for annual, summer, and winter mean temperature during 1901–2012 for CMIP5 and CMIP6, and during 1951–2012 for CMIP5 only. This means that OANT can be detected in annual mean temperature during 1951–2012 for CMIP5 but not for CMIP6, and also the AA signal cannot be detected during 1951–2018 for CMIP6. To explore the possible reasons, we compared the OANT changes for both CMIP5 and CMIP6 and found a lower estimate of OANT results for CMIP6. The key difference may be due to a warmer ALL response in CMIP6 and different NAT forcing used in both generations of models. This indicates that the difference in forcing between the two generations of climate models may cause discrepancies in signal detectability.
4. Discussion and conclusions
This study analyzed the annual, summer, and winter temperature change in China over the period 1901–2018 using four observational datasets; these were then compared with the CMIP5 and CMIP6 models. Clear evidence of warming was observed in annual, summer, and winter temperatures; linear trends of 0.13°, 0.08°, and 0.17°C per decade were estimated for the 1901–2018 period, respectively. Slight differences were observed in various pre-1950s datasets; however, generally the warming trends were similar. The most significant warming was observed in the mid- to high latitudes and mountainous areas, particularly in winter.
The CMIP5 and CMIP6 models reproduced the aforementioned warming features, showing good consistency with observations in summer temperatures, while generally underestimating observed changes in annual and winter temperatures. Based on the optimal fingerprinting method, the ALL and ANT signals were clearly detected in the annual, summer, and winter temperature changes in the two analyzed periods from 1901 to 1951. The effects of GHG and AA were separable and simultaneously detectable in summer temperatures from 1901 to 1951; however, based on the CMIP6 models, this was not the case for annual and winter temperatures when the analysis period commenced in 1951. Further detection analyses based on individual models show that the GHG forcing signal may be detected in all models for annual, summer and winter temperature change, whereas the AA signal was less detected; this highlights the dominant role of GHG forcing in temperature change. The ALL and ANT forcing signals were also detected in most individual models, with the exception of a few that showed large differences from observed annual or seasonal temperature changes.
The estimated annual mean temperature increase of 1.54°C during 1901–2018 over China was approximately one-third larger than the global warming that occurred during 2010–19 relative to the 1850–1900 period (Gillett et al. 2021). For annual, summer and winter warming, the GHG is the dominant contributor while the AA offsets about one-third to one-quarter of GHG-induced warming. The ratio at which AA offsets GHG-induced annual temperature warming in China is similar to that at the global scale. During 1951–2018, most of the warming evident in annual, summer, and winter temperatures was attributable to GHG. Based on CMIP6 models, the AA contribution was not detectable for annual and winter temperatures. In summer, the AA offset approximately one-third to one-quarter of GHG-induced warming, similar to the global annual mean (IPCC 2013). When compared with that in the study by Sun et al. (2016), the estimated contribution from GHG was larger in this study because urbanization effects were not considered. The underestimated annual warming in the models were attributed to GHG-induced warming during the fitting of the optimal fingerprinting method in this study. However, the estimated AA cooling effects in summer were similar to that in the study by Sun et al. (2016) in terms of the influence of other anthropogenic forcings on annual temperature. This suggests that the use of different attribution methods may result in discrepancies in attribution results, which should be explained in the appropriate context. The dominant roles of ALL and GHG were very evident, confirmed by the multimodel ensemble mean and in most individual models. However, there is still uncertainty around the relative contribution of GHG and AA to the observed surface warming in China. Improved observational data and models, larger sample sizes, and improved analysis methods could all contribute toward further investigations.
Acknowledgments.
This study is supported by the National Key R&D Program of China (Grant 2018YFA0605604) and the National Natural Science Foundation of China (Grants 42077427 and 42025503). We acknowledge the Program for Climate Model Diagnosis and Intercomparison and the World Climate Research Programmer’s Working Group on Coupled Modeling (WCRP) for their roles in making the WCRP CMIP multimodel datasets available. The authors declare no conflict of interest.
Data availability statement.
Observed data in this study are openly available at https://climexp.knmi.nl. The CMIP5 and CMIP6 model simulation data are available at https://esgf-node.llnl.gov/search/cmip5/ and https://esgf-node.llnl.gov/search/cmip6/, respectively.
REFERENCES
Allen, M. R., and P. A. Stott, 2003: Estimating signal amplitudes in optimal fingerprinting. Part I: Theory. Climate Dyn., 21, 477–491, https://doi.org/10.1007/s00382-003-0313-9.
Brönnimann, S., 2009: Early twentieth-century warming. Nat. Geosci., 2, 735–736, https://doi.org/10.1038/ngeo670.
CMA Climate Change Centre, 2021: Blue Book on Climate Change in China. Science Press, 119 pp.
Cole-Dai, J., D. Ferris, A. Lanciki, J. Savarino, M. Baroni, and M. H. Thiemens, 2009: Cold decade (AD 1810–1819) caused by Tambora (1815) and another (1809) stratospheric volcanic eruption. Geophys. Res. Lett., 36, L22703, https://doi.org/10.1029/2009GL040882.
DelSole, T., L. Trenary, X. Yan, and M. K. Tippett, 2019: Confidence intervals in optimal fingerprinting. Climate Dyn., 52, 4111–4126, https://doi.org/10.1007/s00382-018-4356-3.
Duan, J., and Coauthors, 2019: Detection of human influences on temperature seasonality from the nineteenth century. Nat. Sustainability, 2, 484–490, https://doi.org/10.1038/s41893-019-0276-4.
Eyring, V., S. Bony, G. A. Meehl, C. A. Senior, B. Stevens, R. J. Stouffer, and K. E. Taylor, 2016: Overview of the Coupled Model Intercomparison Project phase 6 (CMIP6) experimental design and organization. Geosci. Model Dev., 9, 1937–1958, https://doi.org/10.5194/gmd-9-1937-2016.
Gillett, N. P., M. F. Wehner, S. F. B. Tett, and A. J. Weaver, 2004: Testing the linearity of the response to combined greenhouse gas and sulfate aerosol forcing. Geophys. Res. Lett., 31, L14201, https://doi.org/10.1029/2004GL020111.
Gillett, N. P., and Coauthors, 2016: The Detection and Attribution Model Intercomparison Project (DAMIP v1.0) contribution to CMIP6. Geosci. Model Dev., 9, 3685–3697, https://doi.org/10.5194/gmd-9-3685-2016.
Gillett, N. P., and Coauthors, 2021: Constraining human contributions to observed warming since the pre-industrial period. Nat. Climate Change, 11, 207–212, https://doi.org/10.1038/s41558-020-00965-9.
GISTEMP Team, 2022: GISS Surface Temperature Analysis (GISTEMP), version 4. NASA Goddard Institute for Space Studies, accessed 21 September 2022, https://data.giss.nasa.gov/gistemp/.
Hansen, J., R. Ruedy, M. Sato, and K. Lo, 2010: Global surface temperature change. Rev. Geophys., 48, RG4004, https://doi.org/10.1029/2010RG000345.
Hegerl, G. C., S. Bronnimann, A. Schurer, and T. Cowan, 2018: The early 20th century warming: Anomalies, causes, and consequences. Wiley Interdiscip. Rev.: Climate Change, 9, e522, https://doi.org/10.1002/wcc.522.
Hegerl, G. C., and Coauthors, 2019: Causes of climate change over the historical record. Environ. Res. Lett., 14, 123006, https://doi.org/10.1088/1748-9326/ab4557.
IPCC, 2013: Climate Change 2013: The Physical Science Basis. Cambridge University Press, 1535 pp., https://doi.org/10.1017/CBO9781107415324.
IPCC, 2021: Summary for policymakers. Climate Change 2021: The Physical Science Basis, V. Masson Delmotte et al., Eds., Cambridge University Press, 31 pp., https://www.ipcc.ch/report/ar6/wg1/downloads/report/IPCC_AR6_WGI_SPM.pdf.
Jones, G. S., P. A. Stott, and N. Christidis, 2013: Attribution of observed historical near-surface temperature variations to anthropogenic and natural causes using CMIP5 simulations. J. Geophys. Res. Atmos., 118, 4001–4024, https://doi.org/10.1002/jgrd.50239.
Jones, P. D., D. H. Lister, T. J. Osborn, C. Harpham, M. Salmon, and C. P. Morice, 2012: Hemispheric and large-scale land surface air temperature variations: An extensive revision and an update to 2010. J. Geophys. Res., 117, D05127, https://doi.org/10.1029/2011JD017139.
Knutson, T. R., J. Kam, F. Zeng, and A. T. Wittenberg, 2018: CMIP5 model based assessment of anthropogenic influence on record global warmth during 2016. Bull. Amer. Meteor. Soc., 99 (1), S11–S15, https://doi.org/10.1175/BAMS-D-17-0104.1.
Lee, J. Y., and Coauthors, 2021: Future global climate: Scenario-based projections and near-term information. Climate Change 2021: The Physical Science Basis, V. Masson-Delmotte et al., Eds., Cambridge University Press, 553–672.
Lenssen, N. J. L., G. Schmidt, J. Hansen, M. Menne, A. Persin, R. Ruedy, and D. Zyss, 2019: Improvements in the GISTEMP uncertainty model. J. Geophys. Res. Atmos., 124, 6307–6326, https://doi.org/10.1029/2018JD029522.
Li, Q., W. Li, P. Si, X. Gao, W. Dong, P. Jones, and L. Cao, 2010: Assessment of surface air warming in northeast China, with emphasis on the impacts of urbanization. Theor. Appl. Climatol., 99 469, https://doi.org/10.1007/s00704-009-0155-4.
Li, Q., L. Zhang, W. Xu, T. Zhou, J. Wang, P. Zhai, and P. Jones, 2017: Comparisons of time series of annual mean surface air temperature for China since the 1900s. Bull. Amer. Meteor. Soc., 98, 699–711, https://doi.org/10.1175/BAMS-D-16-0092.1.
Liang, E. Y., X. M. Shao, and Y. Xu, 2009: Tree-ring evidence of recent abnormal warming on the southeast Tibetan Plateau. Theor. Appl. Climatol., 98, 9–18, https://doi.org/10.1007/s00704-008-0085-6.
Lu, C., Y. Sun, H. Wan, X. Zhang, and H. Yin, 2016: Anthropogenic influence on the frequency of extreme temperatures in China. Geophys. Res. Lett., 43, 6511–6518, https://doi.org/10.1002/2016GL069296.
Marvel, K., G. A. Schmidt, D. Shindell, C. Bonfils, A. N. LeGrande, L. Nazarenko, and K. Tsigaridis, 2015: Do responses to different anthropogenic forcings add linearly in climate models? Environ. Res. Lett., 10, 104010, https://doi.org/10.1088/1748-9326/10/10/104010.
Menne, M. J., C. N. Williams, B. E. Gleason, J. J. Rennie, and J. H. Lawrimore, 2018: The Global Historical Climatology Network monthly temperature dataset, version 4. J. Climate, 31, 9835–9854, https://doi.org/10.1175/JCLI-D-18-0094.1.
Newhall, C. G., and S. Self, 1982: The volcanic explosivity index (VEI): An estimate of explosive magnitude for historical volcanism. J. Geophys. Res., 87, 1231–1238, https://doi.org/10.1029/JC087iC02p01231.
Parker, D. E., 2010: Uncertainties in early Central England temperatures. Int. J. Climatol., 30, 1105–1113, https://doi.org/10.1002/joc.1967.
Ren, G. Y., 2008: History, current state and uncertainty of studies of climate change attribution (in Chinese). Adv. Earth Sci., 23, 1084–1091., https://doi.org/10.11867/j.issn.1001-8166.2008.10.1084.
Ribes, A., and L. Terray, 2013: Application of regularised optimal fingerprinting to attribution. Part II. Application to global near-surface temperature. Climate Dyn., 41, 2837–2853, https://doi.org/10.1007/s00382-013-1736-6.
Ribes, A., S. Planton, and L. Terray, 2013: Application of regularized optimal fingerprinting to attribution. Part I: Method, properties and idealised analysis. Climate Dyn., 41, 2817–2836, https://doi.org/10.1007/s00382-013-1735-7.
Robock, A., 2000: Volcanic eruptions and climate. Rev. Geophys., 38, 191–219, https://doi.org/10.1029/1998RG000054.
Shiogama, H., T. Nagashima, T. Yokohata, S. A. Crooks, and N. Toru, 2006: Influence of volcanic activity and changes in solar irradiance on surface air temperatures in the early twentieth century. Geophys. Res. Lett., 33, L09702, https://doi.org/10.1029/2005GL025622.
Staten, P. W., J. Lu, K. M. Grise, S. M. Davis, and T. Birner, 2018: Re-examining tropical expansion. Nat. Climate Change, 8, 768–775, https://doi.org/10.1038/s41558-018-0246-2.
Sun, Y., X. Zhang, G. Ren, F. W. Zwiers, and T. Hu, 2016: Contribution of urbanization to warming in China. Nat. Climate Change, 6, 706–709, https://doi.org/10.1038/nclimate2956.
Taylor, K. E., R. J. Stouffer, and G. A. Meehl, 2012: An overview of CMIP5 and the experiment design. Bull. Amer. Meteor. Soc., 93, 485–498, https://doi.org/10.1175/BAMS-D-11-00094.1.
Tung, K.-K., and J. Zhou, 2013: Using data to attribute episodes of warming and cooling in instrumental records. Proc. Natl. Acad. Sci. USA, 110, 2058–2063, https://doi.org/10.1073/pnas.1212471110.
Wang, Y. J., Y. Sun, T. Hu, D. Qin, and L. Song, 2017: Attribution of temperature changes in western China. Int. J. Climatol., 38, 742–750, https://doi.org/10.1002/joc.5206.
Wilks, D. S., 2011: Statistical Methods in the Atmospheric Sciences. 3rd ed. Elsevier, 676 pp.
Xu, W., and Coauthors, 2018: A new integrated and homogenized global monthly land surface air temperature dataset for the period since 1900. Climate Dyn., 50, 2513–2536, https://doi.org/10.1007/s00382-017-3755-1.
Xu, Y., X. J. Gao, Y. Shi, and B. T. Zhou, 2015: Detection and attribution analysis of annual mean temperature changes in China. Climate Res., 63, 61–71, https://doi.org/10.3354/cr01283.
Yin, H., Y. Sun, H. Wan, X. Zhang, and C. Lu, 2017: Detection of anthropogenic influences on the intensity of extreme temperatures in China. Int. J. Climatol., 37, 1229–1237, https://doi.org/10.1002/joc.4771.
Yin, H., Y. Sun, and M. G. Donat, 2019: Changes in temperature extremes in the Tibetan Plateau and their attribution. Environ. Res. Lett., 14, 124015, https://doi.org/10.1088/1748-9326/ab503c.
Yin, H., M.-Y. Li, and L. Huang, 2021: Summer mean temperature reconstruction based on tree-ring density over the past 440 years on the eastern Tibetan Plateau. Quat. Int., 571, 81–88, https://doi.org/10.1016/j.quaint.2020.09.018.
Zhang, X. G., and F. G. Zhang, 1985: The relationship between large volcanic eruptions and the dryness/wetness and cold/warm in China (in Chinese). Acta Meteor. Sin., 43, 196–207.
Zhang, Y. W., L. Zhang, and Y. Xu, 2016: Simulations and projections of the surface air temperature in China by CMIP5 models. Climate Change Res., 1, 10–19., https://doi.org/10.12006/j.issn.1673-1719.2015.113.
Zhao, T. B., C. X. Li, and Z. Y. Zuo, 2016: Contributions of anthropogenic and external natural forcings to climate changes over China based on CMIP5 model simulations. Sci. China Earth Sci., 59, 503–517, https://doi.org/10.1007/s11430-015-5207-2.
Zheng, X. B., Y. X. Luo, T. L. Zhao, J. Chen, and W. M. Kang, 2012: Geographical and climatological characterization of aerosol distribution in China. Sci. Geogr. Sin., 32, 265–272, https://doi.org/10.13249/j.cnki.sgs.2012.03.265.
Zhou, B., Y. Xu, J. Wu, S. Dong, and Y. Shi, 2016: Changes in temperature and precipitation extreme indices over China: Analysis of a high-resolution grid dataset. Int. J. Climatol., 36, 1051–1066, https://doi.org/10.1002/joc.4400.
