3.1 Spatial coverage
Figure 1 shows the annual coverage comparison among CMA-GMST, HadCRUT5 analysis, NOAAGlobalTemp, GISTEMP (1200 km smoothing, so is thereafter), BEST and CMST (the average of the Imax and Imin datasets, so is thereafter) from 1850 to 2022. The latest version up to 2023 of each published dataset is applied here, such as V5.1 of NOAAGlobalTemp. It could be found that the spatial coverages differ among these datasets, especially in the periods with sparse observations. For example, NOAAGlobalTemp is the only one with complete coverage of all land and ocean areas for the entire period of record, while the data coverages of HadCRUT5 analysis and BEST have increased over time and have three obvious drops in the 1860s, 1910s and 1940s. By comparison, time-varying characteristics of CMA-GMST coverage are primarily consistent with those of HadCRUT5 analysis and BEST, and its coverage magnitude is between the two published datasets and is somewhat higher than that of HadCRUT5 analysis. This is likely related to: (1) the land component of CMA-GMST is generated based on the stations with data length ≥ 20 year, and by local statistical interpolation with a 1000 km radius which is close to that of HadCRUT5 analysis (1300 km). These parameters are stricter than those adopted in the other published datasets, such as BEST (Rohde and Hausfather 2020; Morice et al. 2021); (2) although the ocean component of CMA-CMST is interpolated by the same technique as that adopted in NOAAGlobalTemp, a geographic masking is postprocessed to prevent global averages from depending heavily upon the highly smoothed extrapolation estimates (Vose et al. 2012), which however is omitted in NOAAGlobalTempV5.1 in order to get the full coverage (Vose et al. 2021).
Specifically, the areal coverage in the CMA-GMST grids reaches 90% in the middle 1950s and exceeds 99% from the late 1970s, similarly to HadCRUT5 analysis. Moreover, consistent with HadCRUT5 analysis and BEST, an obvious higher data coverage of the Northern Hemisphere (NH) than the Southern Hemisphere (SH) is detected in CMA-GMST especially before the 1950s, and two drops in the 1910s and 1940s are found in the SH associated with the drops of SST observations while the two world wars (Morice et al. 2021). Quantitatively, both the NH coverages of CMA-GMST and HadCRUT5 analysis exceed 95% in the early 1920s and reach 100% in the mid-1950s. However, both the SH coverages of CMA-GMST and HadCRUT5 analysis exceed 95% in the early 1970s. In addition, the NH coverage of CMA-GMST is persistently higher than that of HadCRUT5 analysis, while a somewhat lower SH coverage of CMA-GMST is found around the period 1957–1972. Nevertheless, the corresponding SH coverage of CMA-GMST is obviously higher than that of CMST.
3.2 Correlation with the published products
Figure 2 shows the annual average ST anomaly spatial correlations between CMA-GMST and HadCRUT5 analysis, NOAAGlobalTemp, GISTEMP, BEST and CMST across the globe. CMA-GMST is resampled to match the spatial resolution of each reference product. The spatial correlations between CMA-GMST and each of the existing products are calculated for the monthly data first and then averaged over the 12 months of the year to get the annual values, to assess the spatial structure similarity of temperatures between CMA-GMST and the existing datasets. It could be found that the spatial correlation of CMA-GMST with HadCRUT5 analysis is highest, it increases from about 0.65 in 1850 to around 0.8 from the early 1880s, and keeps 0.8–0.9 to present. The persistent higher correlation of CMA-GMST with HadCRUT5 analysis than those with the other published products is reasonable due to their close parameters applied to screen the land stations with enough record length and control the data extrapolation in an appropriate distance range, which have been pointed out in the coverage comparison part. Besides, the revised SST bias estimations by historical HadSST4 data might also be a potential contributor to the higher correlation with HadCRUT5 analysis. By comparison, the spatial correlations with NOAAGlobalTemp, GISTEMP, BEST and GMST are smaller by about 0.1–0.2 relative to that with HadCRUT5 analysis before the 1900s, and keep ≥ 0.7 after the 1950s, except for a retreat of correlation to about 0.6–0.7 from the mid-1950s to the early 2000s with NOAAGlobalTemp.
Figure 3 shows the spatial distribution of ST anomaly correlations between CMA-GMST and HadCRUT5 analysis, BEST and CMST over time from 1880 to 2022. Only HadCRUT5 analysis, BEST and CMST are applied to compare in this part, because they provide the spatial field of anomalies relative to a same 1961–1990 baseline period as CMA-GMST, or provide additional climate fields for conversion such as BEST. Correlations have been calculated where paired data are present for at least 70% of all months during the statistical period. It could be found that the correlations in terrestrial area are obviously higher in regions including the North America, Europe excluding Greenland and the middle-to-high latitudes of Asia, where the correlations of CMA-GMST with the published products universally concentrate in > 0.9. Similarly, the coherently higher correlations > 0.9 in oceanic area occur in parts of the North Atlantic. Beyond that, the correlations of CMA-GMST with the published datasets are mainly > 0.8, except for regions with limited observations, such as Africa, South America, the Arctic ice covered area and the high latitudes of SH (Morice et al. 2021).
Albeit the consistency in most regions, some regional differences are also found, for example, the correlations for Australia and the middle-to-low latitudes of Asia are overall higher with HadCRUT5 analysis and BEST, while the correlations for the Equatorial East Pacific region with CMST are mainly > 0.9 and higher than those with HadCRUT5 analysis and BEST at about 0.8–0.9. Reconstruction technique might be a potential contributor to these differences. Specifically, CMA-GMST, HadCRUT5 analysis and BEST conduct reconstruction for terrestrial area only referring to the observations from spatial neighbors, while a low- and high-frequency components reconstruction by referring the observations not only from the spatial but also the temporal neighbors is applied by CMST and CMA-GMST for oceanic reconstruction (Morice et al. 2021; Sun et al. 2022).
3.3 Annual to decadal climate monitoring
Figure 4 shows the comparisons of the annual and decadal average global ST anomalies (relative to a 1961–1990 baseline period) among CMA-GMST, HadCRUT5 analysis, NOAAGlobalTemp, GISTEMP, BEST and CMST from 1850 to 2022. The ST anomalies for each year are calculated as the mean of the area-weighted average for the NH and SH. The anomalies for each decade are calculated as the mean of each 10 yearly values. Overall, the annual average global ST anomalies and their interannual variations of CMA-GMST are comparable to those of the published datasets. Fluctuations with no obvious changing trend of the global STs before the 1900s, and thereafter a temperature decline in the 1900s are detected by all the datasets including CMA-GMST. From the 1920s, the average global STs in CMA-GMST and all the published products undergo two obvious warming periods, including the one from the 1920s to the middle 1940s and the other from the mid-1960s to present. Between these two periods, there is a slight retreat of the increasing temperatures. Quantitatively, the fitted linear trends of the global temperature anomalies are approximately 0.84, 1.69 and 1.87°C/100 year for the periods 1880–2022, 1960–2022 and 1980–2022 respectively, and they agree well with the range of trends computed from other published ST analyses (Table 1).
Meanwhile, appreciable differences among the six datasets in the period 1850–1900 are found. The differences are reasonable due to the larger uncertainties of the ST analyses associated with the sparse observations in the earlier periods (Morice et al. 2021), however, they will pose confusion while the assessment of the global warming above pre-industrial levels owing to the mean temperature over the period 1850–1900 is used as the pre-industrial baseline (Schurer et al. 2017). Quantitative comparison shows that the mean value of the annual records over the period 1850–1900 is − 0.377°C for CMA-GMST, it is close to HadCRUT5 analysis with − 0.357°C and BEST with − 0.370°C, due to the closer coverages and higher anomaly correlations of CMA-GMST with HadCRUT5 analysis and BEST. Comparatively, the mean anomalies over the period 1850–1900 for NOAAGlobaltemp and CMST are higher as − 0.299 and − 0.254°C.
Figure 5 shows the comparisons of the annual area-weighted average ST anomalies (relative to a 1961–1990 baseline period) for land, ocean, NH and SH from 1850 to 2022. The same as the assessment result on global scale, the annual average ST anomalies and their inter-annual variations of CMA-GMST are consistent with the published datasets. Quantitatively, the fitted linear trends of land/ocean are 1.08/0.77, 2.43/1.54 and 2.90/1.82°C/100 year, and the trends of NH/SH are 0.95/0.74, 2.15/1.22 and 2.76/0.98°C/100 year for the periods 1880–2022, 1960–2022 and 1980–2022 respectively. Comparison assessments indicate that the regional observed trends of CMA-GMST agree well with the range of trends computed from other published ST analyses, no matter for land, ocean, or hemispheres. In addition, higher warming rates of land than ocean and higher warming rates of NH than SH for the various periods, and the higher warming rates of recent period for the globe, land, ocean and NH are detected in all the ST datasets including CMA-GMST. Moreover, the warming rate differences between land and ocean, and those between NH and SH are also detected to increase in the recent period by all the ST datasets including CMA-GMST (Table 1).
Table 1
Observed trends (°C/100 year) in CMA-GMST, HadCRUT5 analysis, NOAAGlobalTemp, GISTEMP, BEST and CMST for periods 1880–2022, 1960–2022 and 1980–2022 over the globe, land, ocean and hemispheres. Trend values are calculated based on the annual average anomalies with ordinary least squares
Region | Time period | CMA-GMST | HadCRUT5 analysis | NOAAGlobalTemp | GISTEMP | BEST | CMST |
Globe | 1880–2022 | 0.84 | 0.83 | 0.78 | 0.78 | 0.86 | 0.74 |
1960–2022 | 1.69 | 1.77 | 1.67 | 1.74 | 1.85 | 1.60 |
1980–2022 | 1.87 | 1.90 | 1.82 | 1.92 | 1.95 | 1.72 |
Land | 1880–2022 | 1.08 | 1.04 | 1.15 | 1.09 | 1.10 | 1.06 |
1960–2022 | 2.43 | 2.47 | 2.53 | 2.61 | 2.50 | 2.40 |
1980–2022 | 2.90 | 2.88 | 2.89 | 3.00 | 2.83 | 2.76 |
Ocean | 1880–2022 | 0.77 | 0.77 | 0.65 | 0.70 | 0.79 | 0.65 |
1960–2022 | 1.54 | 1.65 | 1.50 | 1.57 | 1.76 | 1.42 |
1980–2022 | 1.82 | 1.84 | 1.79 | 1.87 | 1.97 | 1.68 |
NH | 1880–2022 | 0.95 | 0.96 | 0.93 | 0.94 | 0.97 | 0.90 |
1960–2022 | 2.15 | 2.29 | 2.22 | 2.25 | 2.35 | 2.14 |
1980–2022 | 2.76 | 2.78 | 2.77 | 2.82 | 2.87 | 2.70 |
SH | 1880–2022 | 0.74 | 0.70 | 0.62 | 0.63 | 0.76 | 0.59 |
1960–2022 | 1.22 | 1.25 | 1.12 | 1.23 | 1.34 | 1.05 |
1980–2022 | 0.98 | 1.03 | 0.88 | 1.01 | 1.03 | 0.75 |
Figure 6 shows the annual average ST anomalies for each 30° latitude bands. It shows that the differences between the six datasets are pronounced in 90S–60ºS, which is related to the large uncertainty mainly due to the sparse observations in this region (Morice et al. 2021). Quantitatively, the annual average anomalies of NOAAGlobalTemp and CMST are relatively smooth with standard deviations (SDs) of 0.191 and 0.213 over the period 1850–2022 in 90S–60ºS, while BEST has the largest variability with SD of 0.583, and the corresponding SDs of CMA-GMST and HadCRUT5 analysis are between them by 0.257 and 0.384. In addition, it is obviously cooler in HadCRUT5 analysis and BEST than in NOAAGlobalTemp and CMST before the 1880s in 60S–30ºS, and by comparison, CMA-GMST is closer to HadCRUT5 analysis and BEST. Apart from the appreciable differences that are mainly located in regions and periods with less observations, the observed trends of the annual average anomalies over the different latitudinal zones basically agree well with each other between the six ST datasets.
Figure 7 shows the spatially resolved trends (°C/10 year) of CMA-GMST (a), HadCRUT5 analysis (b), NOAAGlobalTemp (c), GISTEMP (d), BEST (e) and CMST (f) over 1900–1980. Trends have been calculated where data are present in both the first and last decade and for at least 70% of all years within the period, referenced to IPCC AR6 report (IPCC 2021). All the datasets including CMA-GMST show that the temperature trends focus on − 0.15–0.15°C/10 year over 1900–1980 and are dominated by the positive trends. Whatever, some regional differences of the trends among the six ST datasets are found, for example, cooling trends in the central region of South America are detected by HadCRUT5 analysis, while corresponding warming trends are detected by BEST, CMST and CMA-GMST, although the overall higher correlation of CMA-GMST with HadCRUT5 analysis has been mentioned.
When it goes to the period 1981–2022 (Fig. 8), all the datasets including CMA-GMST show significant increases of warming rates for most regions of the world. The strongest warming rates over 1981–2022 occur in the Arctic ice covered area, where the warming trends are detected ≥ 0.6°C/10 year by all the six datasets. Followed are the warming trends in Eurasia, where large areas are detected with warming rates ≥ 0.3°C/10 year by the datasets except for NOAAGlobalTemp with cooling trends over Greenland. Comparatively, the warming trends over ocean are lower, they are donated by 0–0.3°C/10 year, and moreover all the datasets including CMA-GMST show cooling trends in the high-latitudes of the SH and the near-equatorial region of the South Pacific. Except for the overall consistency, some regional differences are also detected, for example, some small regions with warming trends ≥ 0.45°C/10 year over 1981–2022 are detected in the eastern Asia by CMA-GMST, while the local higher warming is relatively inconspicuous in the published datasets. The impacts of the regional differences detected in Fig. 7 and Fig. 8 on the assessment of global temperature and its trends might be negligible, however, it is important for regional assessment and emphasizes the necessity of conducting multiple products comparison while assessing the temperature changes in small regions.
3.4 Monthly to seasonal climate monitoring
The statements have been issued by several institutions that 2023 was the warmest year on record since meteorological records began. Our quantitative evaluation indicates that the average global temperatures of 2023 above the 1991–2020 monthly climatology reached 0.54, 0.56, 0.55, 0.55, 0.55 and 0.51°C for CMA–GMST, HadCRUT5, NOAAGlobalTemp, GISTEMP, BEST and CMST respectively. Moreover, the ranking of the average global temperature for each calendar month also shows that each month from July 2023 to December 2023 ranked as the globe's hottest month in recorded history (Table 2).
Table 2
The year with historical highest temperature for each calendar month and annual average up to 2023
Date | CMA-GMST | HadCRUT5 analysis | NOAAGlobalTemp | GISTEMP | BEST | CMST |
Jan | 2016 | 2016 | 2016 | 2016 | 2016 | 2016 |
Feb | 2016 | 2016 | 2016 | 2016 | 2016 | 2016 |
Mar | 2016 | 2016 | 2016 | 2016 | 2016 | 2016 |
Apr | 2016 | 2020 | 2016 | 2020 | 2020 | 2016 |
May | 2020 | 2020 | 2016 | 2020 | 2016 | 2023 |
Jun | 2023 | 2023 | 2023 | 2023 | 2023 | 2023 |
Jul | 2023 | 2023 | 2023 | 2023 | 2023 | 2023 |
Aug | 2023 | 2023 | 2023 | 2023 | 2023 | 2023 |
Sep | 2023 | 2023 | 2023 | 2023 | 2023 | 2023 |
Oct | 2023 | 2023 | 2023 | 2023 | 2023 | 2023 |
Nov | 2023 | 2023 | 2023 | 2023 | 2023 | 2023 |
Dec | 2023 | 2023 | 2023 | 2023 | 2023 | 2023 |
Yearly | 2023 | 2023 | 2023 | 2023 | 2023 | 2023 |
Figure 9 shows the average ST anomalies (relative to a 1991–2020 monthly climatology) for the JJA (June, July and August) season over the globe and hemispheres from 1850 to 2023 in CMA-GMST, HadCRUT5 analysis, NOAAGlobalTemp, GISTEMP, BEST and CMST. It is shown that the JJA season 2023 ranked as the warmest June–August period in the 174-year record over the globe. The average global STs of the JJA season 2023 were about 0.61, 0.55, 0.57, 0.61 and 0.55°C above the 1991–2020 monthly climatology in HadCRUT5 analysis, NOAAGlobalTemp, GISTEMP, BEST and CMST, and it was comparable as 0.60°C in CMA-GMST. Moreover, all the datasets including CMA-GMST indicate that the JJA season 2023 was the NH’s hottest meteorological summer on record, at about 0.71, 0.74, 0.73, 0.69 and 0.69°C above the 1991–2020 monthly climatology in HadCRUT5 analysis, NOAAGlobalTemp, GISTEMP, BEST and CMST, and about 0.74°C in CMA-GMST. The JJA season 2023, which also marked the SH’s winter, was the SH’s warmest winter, it was on record at 0.51, 0.36, 0.42, 0.53 and 0.41°C above the 1991–2020 monthly climatology in HadCRUT5 analysis, NOAAGlobalTemp, GISTEMP, BEST and CMST and was comparable as 0.45°C in CMA-GMST.
Figure 10 is the same as Fig. 9, but for the SON (September, October and November) season. It is also shown that the SON season 2023 ranked as the warmest September–November period in the 174-year record over the globe. The average global STs of the SON season 2023 above the 1991–2020 monthly climatology were about 0.79, 0.78, 0.79, 0.80, 0.79 and 0.72°C in CMA-GMST, HadCRUT5 analysis, NOAAGlobalTemp, GISTEMP, BEST and CMST respectively and higher than those of the JJA season. A significantly higher average ST of the SON season 2023 above the 1991–2020 monthly climatology than that of the JJA season was also found in all the six datasets including CMA-GMST in the NH, the SON season 2023 was the NH’s hottest meteorological autumn on record, at about 1.08, 1.04, 1.15, 1.11, 1.07 and 1.03°C above the 1991–2020 monthly climatology in CMA-GMST, HadCRUT5 analysis, NOAAGlobalTemp, GISTEMP, BEST and CMST. Comparatively, the SON season 2023, which also marked the SH’s spring, was the SH’s warmest spring, it was on record at 0.49, 0.52, 0.43, 0.49, 0.52 and 0.40°C above the 1991–2020 monthly climatology in CMA-GMST, HadCRUT5 analysis, NOAAGlobalTemp, GISTEMP, BEST and CMST, and were comparable as those of the JJA season.