Download as pdf or txt
Download as pdf or txt
You are on page 1of 21

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/358138643

A Comparison of Global Surface Temperature Variability, Extremes and


Warming Trend using Reanalysis Data Sets and CMST-Interim

Article  in  International Journal of Climatology · January 2022


DOI: 10.1002/joc.7551

CITATIONS READS

8 273

5 authors, including:

Yang Yang Qingxiang Li


Sun Yat-Sen University Sun Yat-Sen University
3 PUBLICATIONS   11 CITATIONS    138 PUBLICATIONS   4,548 CITATIONS   

SEE PROFILE SEE PROFILE

Zhaoyang Song Wenbin Sun


GEOMAR Helmholtz Centre for Ocean Research Kiel Sun Yat-Sen University
15 PUBLICATIONS   102 CITATIONS    19 PUBLICATIONS   81 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Climate change Observations, Physics and Uncertainties View project

PalMod View project

All content following this page was uploaded by Qingxiang Li on 29 March 2022.

The user has requested enhancement of the downloaded file.


Received: 16 May 2021 Revised: 22 January 2022 Accepted: 25 January 2022
DOI: 10.1002/joc.7551

RESEARCH ARTICLE

A comparison of global surface temperature variability,


extremes and warming trend using reanalysis datasets
and CMST-Interim

Yang Yang1 | Qingxiang Li1,2 | Zhaoyang Song1 | Wenbin Sun1 |


1,2
Wenjie Dong

1
School of Atmospheric Sciences, Sun Yat-
sen University, and Key Laboratory of
Abstract
Tropical Atmosphere-Ocean System, Reanalysis data are widely used to investigate long-term surface temperature
Ministry of Education, Zhuhai, China changes due to insufficient spatial coverage of observational data. However,
2
Southern Laboratory of Ocean Science
because of the limitations of data assimilation and model performance in the
and Engineering (Guangdong Zhuhai),
Zhuhai, China reanalysis datasets, it is essential to evaluate the quality of the reanalysis
datasets. Based on the newly released version of China global Merged Surface
Correspondence
Qingxiang Li, School of Atmospheric
Temperature dataset, Interim version (CMST-Interim), the performance of five
Sciences, Sun Yat-sen University, Zhuhai, reanalysis datasets (ERA5, NCEP/NCAR R1, JRA-55, CERA-20C, and
China. 20CRv3), covering more than 50 years, is compared in terms of long-term vari-
Email: liqingx5@mail.sysu.edu.cn
ation bias, warming trend, and the consistency of the extreme temperature
Funding information years from the period 1958–2010. The results reflect that the above reanalysis
National Natural Science Foundation of
datasets have reasonable representativeness of global temperature change.
China, Grant/Award Number: 41975105;
National Key R&D Program of China, ERA5 and 20CRv3 perform better than the other reanalysis datasets, with rela-
Grant/Award Numbers: 2017YFC1502301, tively small deviation, their warming trends are closer to the observation
2018YFC1507705
results, the appearance of extreme temperature years are also more consistent
with the observed, and CERA-20C is the next. JRA-55 is slightly better in
detecting the extremely cold years, while NCEP/NCAR R1 is slightly worse in
biases but performs well in high-temperature years reproduction.

KEYWORDS
bias, CMST-Interim, reanalysis datasets, surface temperature anomaly, trend

1 | INTRODUCTION to 2012 (Cowtan and Way, 2014; Karl et al., 2015; Huang
et al., 2017; Simmons et al., 2017; Kadow et al., 2020; Li
Surface temperature is a crucial indicator of global et al., 2021). Additionally, reanalysis datasets can be used
climate change and one of the most important essential to improve our understanding in terms of the global
climate variables. Nevertheless, there are still large warming trend, surface temperature variability, and
uncertainties within the widely used global surface tem- extremes (Ma et al., 2008; Mao et al., 2010; Li et al., 2017;
perature observational datasets, mainly due to data 2021; Simmons et al., 2017). Because reanalysis datasets
sparseness at high latitudes (IPCC, 2001; 2007; 2013). provide complete coverage data in spatial and temporal
Using global surface temperature datasets, numerous resolution relative to the observations, several reanalysis
studies estimate the global warming trend, especially for datasets have been applied to climate research since the
the so-called “warming hiatus” period, that is, from 1998 1990s (IPCC, 2013). Simmons et al. (2017) and Li

Int J Climatol. 2022;1–20. wileyonlinelibrary.com/journal/joc © 2022 Royal Meteorological Society 1


2 YANG ET AL.

et al. (2021) estimated the trend of global temperature (Ma et al., 2008; Vose et al., 2012; Hu et al., 2014; Luo
during the “warming hiatus” period using European Cen- et al., 2019; Retamales-Muñoz et al., 2019; Przybylak and
tre for Medium-Range Weather Forecasts (ECMWF) Wyszy nski, 2020; Balmaceda-Huarte et al., 2021; Engdaw
reanalysis (ERA-Interim) and the fifth generation et al., 2022; Zhu et al., 2021). On the one hand, it is help-
ECMWF atmospheric reanalysis (ERA5) datasets, respec- ful to quantitatively identify the contributions to global
tively. Furthermore, these two datasets have also warming from various regions (Simmons and Poli, 2015).
achieved good results in regional climate research On the other hand, it could improve the applicability of
(Goddard and Tett, 2019; Chao et al., 2020). reanalysis data in the models, like the arid and semi-arid
Current reanalysis datasets covering the period around region and permafrost region (Qin et al., 2020; Huai
1979 fall into two main categories. One is to make use of et al., 2021). These analyses focus on the regional perfor-
all assimilated data as much as possible. Such as the earli- mance of surface temperature reanalyses, but there have
est National Centers for Environmental Prediction/ been fewer global evaluations for the latest versions of
National Center for Atmospheric Research (NCAR/NCEP) reanalyses. The assessments of low-frequency and trends
Reanalysis (R1) (Kalnay et al., 1996), NCEP/U.S. Depart- of global temperature from ERA-40, ERA-Interim, ERA5,
ment of Energy (DOE) Reanalysis 2 (NCEP/DOE R2) and JRA-55 show similar estimates of global-mean sur-
(Kanamitsu et al., 2002), ERA-Interim (Dee et al., 2011) face warming and interannual variability compared with
and Climate Forecast System Reanalysis of NCEP (CFSR) several conventional analyses (Kobayshi et al., 2015;
(Saha et al., 2010), the satellite data are assimilated since Simmons et al., 2017; 2021; Hersbach et al., 2020). The
1979 in these datasets. However, ECMWF Reanalysis recent release of the preliminary version of the ERA5
40-year (ERA-40) (Uppala et al., 2005), Japan Meteorologi- reanalysis back extension from 1950 to 1978 from
cal Agency (JMA) 55-year Reanalysis (JRA-55) (Kobayshi ECWMF (Bell et al., 2021) provides a basis for being able
et al., 2015), and ERA5 (Hersbach et al., 2020) assimilate to assess the performance of state-of-the-art reanalysis
VTPR satellite data since 1972/1973. The other is that sat- datasets over a relatively long period, yet the evaluations
ellite data are not assimilated, that is, only surface pressure are focused on the period of 1979 afterward.
and marine winds are assimilated in all periods, including This paper aims to better quantify and detailed evalu-
two kinds of reanalysis datasets, namely Twentieth Cen- ate the global surface temperature from reanalyses. Five
tury Reanalysis version 3 (20CRv3) (Slivinski et al., 2019) widely used state-of-the-art reanalysis datasets, namely
and ECMWF coupled twentieth-century reanalysis ERA5, NCAR/NCEP R1, CERA-20C, 20CRv3, and JRA-
(CERA-20C) (Laloyaux et al., 2018). Each of the two types 55, covering data from both the periods of pre-1979 and
of data has its distinctive characteristics. The first type post 1979, are compared in this paper. The surface tem-
may have higher precision because they assimilate more perature biases, extreme values, and the warming trends
data, but there may be discontinuities between the two will be evaluated against the newly released global sur-
periods before and after assimilating satellite data face temperature dataset CMST-Interim to provide a ref-
(Sturaro, 2003). The second type is less sensitive to this erence for the applicability of the reanalysis datasets in
homogeneity problem, but no matter before or after 1970, climate change trend research. The paper is organized as
because the amount of assimilated data is significantly follows. In section 2, we describe the benchmark dataset,
lower than that of the first type, the accuracy of the latter the reanalysis datasets, and methodology used in this
reanalysis data may be significantly lower than that of the study. In section 3, we present a comprehensive compari-
first type. However, with the improvement of the model son on the representation of surface temperature anom-
and assimilation technology, the accuracy of the latter type aly biases, trends, and extremes among the reanalysis
of reanalysis data is also improving (Poli et al., 2016; datasets. The results are summarized in section 4.
Slivinski et al., 2019). Therefore, it is vital to assess various
reanalysis datasets in terms of their similarities and differ-
ences in the representation of global and regional surface 2 | DATA AND M ETHODOLOGY
temperature change.
Based on the in situ observations or satellite data, 2.1 | Data
there have been many efforts made to evaluate the
reanalysis products for surface air temperature in previ- 2.1.1 | China global Merged Surface
ous studies. For example, some studies have been Temperature, Interim version
assessed the accuracy of the variability and trend of sur-
face temperature for reanalysis data in many areas, such As described by Jones and Harpham (2013), average tem-
as the Qinghai-Tibet Plateau, China, the United States, peratures for the hemispheres and the globe are generally
central Asia, Africa, the Polar Regions, and so on expressed as anomalies from a base period, and
YANG ET AL. 3

anomalies are perfectly adequate for use in climate moni- “warming hiatus” period from 1998 to 2012, the warming
toring. In general, observation datasets such as trend estimated from the CMST agrees well with that cal-
HadCRUT (5 × 5 ), NOAAGlobalTemp (5 × 5 ), BEST culated from datasets in which northern high latitudes
(Berkeley Earth Surface Temperature, 1 × 1 ), and are filled with satellite data, buoy observations, and
GISSTEM (2 × 2 ) are widely used as benchmarks on reanalysis data (Li et al., 2021). Recently, combined with
the validation/comparison of the other climatic datasets observation constraints, 756 ensemble C-LSAT 2.0 mem-
(including satellite, reanalysis, models output and other bers are reconstructed using high-frequency and low-
merged datasets) (Cowtan et al., 2015; Li et al., 2017; frequency components from 1854 to 2018. The ensemble
Simmons et al., 2017, 2021; Beusch et al., 2020; Cook members are then merged with the ERSSTv5 to obtain
et al., 2020). CMST is a newly developed global surface the CMST-Interim dataset. CMST-Interim significantly
temperature (ST) anomaly dataset that only uses in-situ improves the coverage of global surface temperature data
observations (no satellite observations included), the ST compared to the original CMST. The reconstruction
trends derived from it are broadly similar to those from increases the data coverage from 78–81% to 81–89% prior
the above benchmark datasets (Li et al., 2020; 2021). to 1950. The total coverage reaches about 93% from 1953
Recently, the newly released IPCC AR6 has cited it as onwards. In addition, the description of multiscale
one of the benchmark datasets to calculate global/ changes of global and regional land surface temperature is
regional warming trends (Gulev et al., 2021). In addition, more accurate (Sun et al., 2021). However, it is worth noting
comparisons between CMST-Interim and the latest ver- that due to the lack of information on the variation of sea
sion of other two classic benchmark datasets HadCRUT5 ice cover over time (interannual and seasonal), the CMST-
(Morice et al., 2021) and NOAAGlobalTemp5 (Huang et Interim still adopts the fixed values for sea surface tempera-
al., 2020) are shown in Supporting Information. As ture (SST; the missing value is set to −1.8 C) like ERSSTv5
shown in Figures S1 and S2, the long-term and inter- in the Arctic sea ice covered area, rather than the simple
annual variabilities in both global and regional scales land air temperature extrapolation method utilized in some
obtained from these three datasets are generally consis- recent studies (Osborn et al., 2021; Vose et al., 2021). As for
tent with each other, as well as the spatial distributions. the Antarctic region, due to the sparse observations, there
CMST is merged by China global Land Surface Air still are certain problems in the reconstructed CMST-
Temperature (C-LSAT) and Extended Reconstructed Sea Interim series (Sun et al., 2021), so observation constraints
Surface Temperature (ERSST) (Yun et al., 2019; Li are made on the reanalysis datasets.
et al., 2020; Sun et al., 2021). The original version of C-
LSAT 1.0 is a homogenized monthly global land surface
temperature dataset developed by 14 data sources, includ- 2.1.2 | Reanalysis datasets
ing 3 global (CRUTEM4 from the Climatic Research Unit
(CRU) of the University of East Anglia, Global Historical ERA5 (Hersbach et al., 2020) is the fifth generation
Climatology Network (GHCN) dataset from National ECMWF atmospheric reanalysis of the global climate,
Oceanic and Atmospheric Administration's National spanning from January 1950 to the present, splitting into
Center for Environment Information, and Berkeley Earth back extension from 1950 to 1978 (preliminary version)
Surface Temperature (BEST) dataset), 3 regions, and and from 1979 onwards. ERA5 uses the Integrated Fore-
8 national sources (see Table S1), spanning from 1900– casting System (IFS) Cy41r2 and significantly enhances
2014 (see also Xu et al., 2018, table 1). To further evaluate the temporal, horizontal and vertical resolutions com-
global warming since industrialization, C-LSAT 2.0 pared to its predecessor ERA-Interim. The horizontal res-
extends the temporal coverage to 1850, and increases the olution is 31 km, with 137 vertical levels in the
spatial coverage and station density. The most updated atmosphere from the surface up to 0.01 hPa. ERA5 pro-
version of C-LSAT used to develop CMST-Interim is C- vides hourly estimates for a large number of atmospheric,
LSAT 2.0 ensemble (Li et al., 2020; 2021; Sun et al., 2021). land and oceanic climate variables, 3-hourly uncertainty
CMST-Interim is the latest version of CMST, which is rel- estimates, and a timely real-time product within 5 days.
ative to the climatic average over 1961–1990, a period is NCEP/NCAR R1 (Kalnay et al., 1996; Kistler et al., 2001)
chosen to satisfy the requirements of station data is the first modern reanalysis dataset produced by NCEP
processing (Xu et al., 2018), with a spatial resolution of and NCAR forecast model revised in 1995, spanning
5 × 5 spanning from 1854 to 2018 (Sun et al., 2021). from January 1948 onwards. NCEP/NCAR reanalysis
Comparative analysis shows that the consistency of the data assimilation system consists of the NCEP Medium
global warming trend derived from CMST has been Range Forecasting (MRF) spectral model and a three-
strengthened with other similar datasets since 1880 dimensional variational (3D-var) analysis scheme. The
(Li et al., 2020). In particular, during the debated horizontal resolution of NCEP/NCAR R1 is T62 Gaussian
4 YANG ET AL.

grid (around 210 km), with 28 vertical levels from the incremental 4D-Var, which is improved compared to
surface up to 3 hPa. It provides 6-hourly analysis and JRA-25, and with an increased model resolution of TL319
pressure fields on a 2.5 latitude-longitude grid. ECMWF (about 55 km) in horizontal and 60 vertical levels
coupled 20th-century reanalysis (CERA-20C) is the between the surface to 0.1 hPa. In addition, JRA-55 adds
10-member ensemble of coupled climate reanalyses of a new bias correction procedure for satellite data and
the 20th century, produced from the European other observations. It offers atmospheric analysis, screen-
Reanalysis of Global Climate Observations Project (ERA- level analysis, and land surface analysis with a 6-hr
CLIM2), which provides global atmospheric data for the resolution.
period 1901–2010. CERA-20C uses IFS Cy41r2 modelling
and the CERA data assimilation system and assimilates
surface pressure and marine winds observations as well 2.2 | Methodology
as ocean temperature and salinity profiles with an incre-
ment four-dimensional variational (4D-Var) analysis. The To perform the differences and agreements between the
horizontal resolution is 125 km horizontal grid with five global reanalyses and CMST-Interim, the results are
91 vertical levels between the surface and 0.01 hPa compared spanning the common period 1958–2010. Due
(Laloyaux et al., 2018). 20CRv3 (Slivinski et al., 2019; to the five reanalyses being in different resolutions
2021) is generated by the National Oceanic and Atmo- (Table 1), all reanalyses have been re-interpolated
spheric Administration's Cooperative Institute for and reduced to consistent with the 5 × 5 grid of
Research in Environmental Sciences (NOAA/CIRES), CMST-Interim's grid using a bilinear interpolation
and the U.S. Department of Energy (DOE). The latest ver- method. Then, surface temperature anomalies from all
sion NOAA-CIRES-DOE 20CR version 3 spans from 1836 reanalysis datasets have been also calculated from the
to 2015. 20CRv3 uses an 80-member ensemble Kalman corresponding reference period 1961–1990. The seasonal
filter to assimilate a larger set of surface pressure observa- analysis is carried out for March–April–May (MAM),
tions into an updated NCEP Global Forecast System June–July–August (JJA), September–October–November
(GFS) model, version 14.0.1, relative to its predecessor (SON), and December–January–February (DJF). Because
20CRv2, with a spectral horizontal resolution of T254 of the scarcity of observation data in the Antarctic (high-
(effectively 60 km at the equator) and a vertical atmo- latitude area in the Southern Hemisphere, taking the
spheric resolution of 64 levels up to about 0.3 hPa. It pro- north of 60 S here), this region is not included in the
vides a 3-hr global atmospheric estimate by assimilating comparison. Based on the processed datasets, the root-
observed surface pressure and specifying sea surface tem- mean-square error (RMSE), mean bias error (MBE), Pear-
perature, sea ice concentration, and radiation forcing. son linear correlation coefficient (CC), probability density
JRA-55 (Kobayshi et al., 2015) is the second Japanese function (PDF) skill score, and the ranking score in
global atmospheric reanalysis generated by the JMA and extreme years are used to quantify the accuracy or differ-
is also called the Japanese 55-year Reanalysis. It covers ences between the observation and estimated STA from
the period of 1958 to the present and uses the TL319 ver- each reanalysis dataset. Moreover, the time series and its
sion of JMA's global spectral model system with bias over the global, regional and zonal STA are

TABLE 1 Description for CMST-Interim and five reanalysis datasets

Spatial
Abbreviation Full name resolution Time period Reference
 
CMST- China Merged Surface Temperature-Interim 5 ×5 Jan 1854–Dec Sun et al. (2021)
Interim 2018
ERA5 Fifth generation ECMWF reanalysis 0.25 × 0.25 Jan 1950–present Hersbach et
al. (2020)
NCEP1 NCEP/NCAR Reanalysis 2.5 × 2.5 Jan 1948–present Kalnay et al. (1996)
 
CERA-20C ECMWF's Coupled Reanalysis of the Twentieth 1.0 × 1.0 Jan 1901–Dec Laloyaux
Century 2010 et al. (2018)
20CRv3 NOAA-CIRES-DOE Twentieth Century 1.0 × 1.0 Jan 1836–Dec Slivinski
Reanalysis (V3) 2015 et al. (2019)
JRA-55 Japanese 55-year Reanalysis 1.25 × 1.25 Jan 1958–present Kobayshi
et al. (2015)
YANG ET AL. 5

compared spanning the common period 1958–2010. Since 2.2.3 | Other metrics for consistency of
the time scales of CREA-20C and 20CRv3 are more than surface temperature anomaly
a century, the time series of their annual variations for
1901–2018 are also compared with CMST-Interim. Trend and uncertainty
The long-term trend is calculated by the method of
restricted maximum likelihood regression to assess the
2.2.1 | Error analysis change in STA over time, and its 95% confidence level
(~1.96 sigma) is taken as the uncertainty range. The
Root-mean-square error REML method is the basic method used to calculate cli-
Root-mean-square error (RMSE) is a measure for the mate change trend since IPCC Third Assessment Report
mean error derived from the sum of squared errors, (Li et al., 2021). We also test whether there are significant
which is defined as follows: differences between the trends obtained from the
observed and reanalysed data by the overlap of uncer-
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
1X n tainty ranges, that is, when the uncertainty ranges of the
RMSE = ðxi−yiÞ2 , ð1Þ two trends do not overlap, it is considered that these two
n i=1
trends are significantly different.
where n represents the number of samples, xi is the tem-
perature anomaly of the reanalysis data, and yi is the Ranking score in extreme years
temperature anomaly of the CMST-Interim. RMSE indi- To measure the consistency of yearly extreme high and
cates the random difference between reanalysis datasets low temperature between the observations and reanalysis
and the CMST-Interim. datasets, we propose a ranking score to assess the ability
of reanalysis datasets to capture extreme temperatures,
Mean bias error

Mean bias error (MBE) is an average of bias errors, absðOi −RiÞ , absðOi −RiÞ ≤ 10
Scorei= , ð3Þ
expressed as follows: 10 , absðOi− RiÞ > 10

1X n
X
n
MBE= ðxi −yiÞ: ð2Þ Score = 100 − Scorei, ð4Þ
n i=1
i=1

where i is the order, n is the total number of ranking


MBE is intended to measure average bias from the (10 is taken). Oi denotes the order of the ranked year for
reanalysis data to the CMST-Interim. CMST-Interim (i = 1, 2, …, n), and Ri represents the order
of the ith year derived from CMST-Interim in ranked
reanalysis data. For example, if 2007 is located at the first
2.2.2 | PDF skill scores in the observation ranking and 5th in the reanalysis rank-
ing, then Oi is 1, Ri is 5. The original score for each
Perkins et al. (2007) proposed the use of probability den- reanalysis dataset is set to 100. If the ranking (top
sity functions (PDFs) to evaluate the ability of climate 10 extreme records) obtained from the reanalysis datasets
models to simulate the entire distributions of a simulated is consistent with observations, the score will not be
variable. PDFs calculated from the model data are com- changed. Otherwise, the difference will be subtracted
pared against those obtained from the observational data. from the score. A maximum of 10 for the difference can
A large overlapping area between the simulated and be subtracted from the original score.
observed PDFs refers to better performance of the models
and vice versa. Relative to mean or standard deviation,
PDF provides a more robust and straightforward measure 3 | RESULTS
for the similarity between the observed and modelled
PDFs. In this paper, we compare the relative similarity 3.1 | Systematic and random biases
between CMST-Interim and reanalysis datasets. The bin
size of the temperature anomaly is 0.05 C. According to 3.1.1 | Spatial distribution
Perkins et al. (2007), a skill score of 1.0 suggests that the
two PDFs are completely coincident, while a skill score Mean bias error
of 0.0 suggests that the PDFs are independent of each Figure 1 shows the spatial distributions of MBE for the
other. annual STA between the reanalysis datasets and the
6 YANG ET AL.

90°N 90°N 90°N


60°N 60°N 60°N
30°N 30°N 30°N
0 0 0
30°S 30°S 30°S
60°S 60°S 60°S
180 120°W 60°W 0 60°E 120°E 180 180 120°W 60°W 0 60°E 120°E 180 180 120°W 60°W 0 60°E 120°E 180

90°N 90°N
60°N 60°N
30°N 30°N
0 0
30°S 30°S
60°S 60°S
180 120°W 60°W 0 60°E 120°E 180 180 120°W 60°W 0 60°E 120°E 180

F I G U R E 1 Mean bias errors (MBEs) for annual surface temperature anomalies in (a) ERA5, (b) NCEP1, (c) CERA-20C, (d) 20CRv3 and
(e) JRA-55 over 1958–2010 relative to CMST-Interim (unit:  C). All anomalies are calculated relative to the 1961–1990 reference period
[Colour figure can be viewed at wileyonlinelibrary.com]

CMST-Interim (hereafter referred to as observation) from followed by ERA5, which amounts to 0.28 C, and CERA-
1958 to 2010. Overall, the global averaged MBEs are 20C and JRA-55 are both 0.30 C, NCEP/NCAR R1 shows
small: −0.01 C for 20CRv3, 0 C for ERA5, and 0.01 C for the maximum RMSE of 0.34 C. Some regions of East
CERA-20C, followed by −0.03 C for JRA-55 and −0.04 C Asia, Africa, South America, and Australia show larger
for NCEP/NCAR R1. Except for NCEP/NCAR R1 and RMSEs derived from NCEP/NCAR R1, CERA-20C, and
JRA-55, other reanalysis datasets agree better with obser- JRA-55 than that from the other two datasets in those
vation at mid and low latitudes than at high latitudes, the places, the RMSEs are able to reach 0.6 C or larger of
MBEs are within about ±0.2 C in mid and low latitudes. above three reanalyses. Regarding the seasonal variations
The MBEs of NCEP/NCAR R1 in Sahara, South America, (see Figure S4), the maximum of global mean RMSEs
Australia, and the Tibet Plateau are about −0.2 C or less, both occur in DJF, with values ranging from 0.45 to
and the MBEs reach ±0.2 C in some regions of Australia, 0.54 C for all five reanalysis datasets.
Iranian Plateau, and central Africa computed from JRA-
55. While the MBEs derived from all these data reach
over 0.3 C near the Arctic. The spatial patterns of MBEs 3.1.2 | Temporal changes
for seasonal STA in 5 reanalysis datasets (see Figure S3)
are similar to their annual distributions and the global Temporal changes of MBEs and RMSEs for annual STA
averaged errors slightly increase. The global averaged in five reanalysis data are investigated over the globe
MBEs in NCEP/NCAR R1 and JRA-55 are negative in all (north of 60 S) from 1958 to 2010 (Figure 3a,b), MBEs
seasons, and all reanalysis data exhibit negative global and RMSEs derived from ERA5 and 20CRv3 are much
averaged MBEs in JJA. Especially, the absolute values of smaller than those from NCEP/NCAR R1, CERA-20C,
global averaged MBE are larger in JJA than in the other and JRA-55, suggesting that global surface temperature
seasons, while CERA-20C exhibits the maximum value variation in ERA5 and 20CRv3 agree well with observa-
in SON. tion. Among them, MBE and RMSE for NCEP/NCAR R1
still indicate larger errors than the other datasets, espe-
Root-mean-square error cially since 1990. Additionally, CERA-20C shows evi-
As in Figure 2, spatial distributions of RMSEs in dently larger MBE than the others after 2004. It should
reanalysis datasets show large errors at high latitude be pointed out that the MBEs and RMSEs increased
while small at low and mid-latitude. This could be caused slightly after 1990 for all reanalysis datasets. This is
by the observational data being underrepresented and because the comparisons of STA are all relative to the ref-
having coverage bias at high latitudes or at the border of erence period 1961–1990, so the MBEs and RMSEs are
oceans and continents (Cowtan and Way, 2014; Cowtan smaller in this period than those in the other times.
et al., 2015; Kobayshi et al., 2015; Morice et al., 2021; MBEs and RMSEs for monthly STA are shown in
Osborn et al., 2021). Overall, the minimum of global Figure 3c,d. MBEs from five reanalysis datasets show
mean RMSEs is amounting to 0.26 C in 20CRv3, negative values from May through August, with the
YANG ET AL. 7

maximum positive value in October. The maximum June. As for RMSEs, ERA5, and 20CRv3 exhibit smaller
values of MBEs are from CERA-20C in all positive RMSEs for all calendar months, while monthly errors for
months, while the minimum is from NCEP/NCAR R1 in NCEP/NCAR R1 and JRA-55 are relatively larger. The

90°N 90°N 90°N


60°N 60°N 60°N
30°N 30°N 30°N
0 0 0
30°S 30°S 30°S
60°S 60°S 60°S
180 120°W 60°W 0 60°E 120°E 180 180 120°W 60°W 0 60°E 120°E 180 180 120°W 60°W 0 60°E 120°E 180

90°N 90°N
60°N 60°N
30°N 30°N
0 0
30°S 30°S
60°S 60°S
180 120°W 60°W 0 60°E 120°E 180 180 120°W 60°W 0 60°E 120°E 180

FIGURE 2 Same as Figure 1, but for root-mean-square errors (RMSEs) (unit:  C) [Colour figure can be viewed at wileyonlinelibrary.com]

F I G U R E 3 MBEs and RMSEs for (a, b) annual and (c, d) monthly surface temperature anomalies from 1958 to 2010. (unit:  C) [Colour
figure can be viewed at wileyonlinelibrary.com]
8 YANG ET AL.

maximum values of reanalyses occur in January (except smaller in ERA5, CERA-20C, and 20CRv3 than in NCEP/
for NCEP/NCAR R1 is in October). A similar conclusion NCAR R1 and JRA-55. It shows that the surface tempera-
can be seen in the map of the temporal evolution of the ture anomalies are more concentrated from the above
spatial distribution (see Figures S3 and S4). three reanalysis datasets relative to CMST-Interim.

3.2 | Comparisons of annual 3.2.2 | Pearson correlation coefficients


temperature anomaly
To assess the interannual variability of reanalysis temper-
3.2.1 | Distributions of scattered points ature datasets (Bell et al., 2021), we examined the Pear-
son correlation coefficient (CCs) of surface temperature
Averaged temperature anomalies over 1958–2010 at each anomaly between reanalysis datasets and observation
5 × 5 grid (north of 60 S) from five reanalysis datasets (Figure 5). It is evident that the reanalysis data have high
are compared against CMST-Interim in Figure 4. ERA5 CCs with CMST-Interim, and the global average (north
and 20CRv3 (coefficient of determination R2 are 0.52) are of 60 S) CCs reach about 0.8. Only CC between NCEP/
more comparable to observation than the other reanalysis NCAR R1 and observation is slightly lower but also
(R2 are within 0.5). The scatter distributions of ERA5 and reaches 0.76, and the rest are JRA-55 (0.80), CERA-20C
20CRv3 are closer to the diagonal line. In contrast, (0.82), ERA5(0.82), and 20CRv3 (0.85). As for spatial dis-
because of many high- and low-temperature anomalies tributions of CC, the higher CC is mainly concentrated in
in NCEP/NCAR R1, CERA-20C, and JRA-55, the scatter the middle and high latitudes of the Northern Hemi-
points of these three reanalysis data deviate significantly sphere, except for central and northern Africa, South
from the diagonal line in Figure 4b,c,e. It implies that the America, and the Tibet Plateau have lower CCs. In addi-
temperature data in NCEP/NCAR R1, CERA-20C, and tion, the CC in the mid-latitudes ocean of the Southern
JRA-55 are less consistent with observation than ERA5 Hemisphere is slightly lower. This also agrees well with
and 20CRv3. Moreover, the “spotlight” regions, that is, the results obtained by Simmons et al. (2021) for ERA-5
the areas with the largest scattering point density, are and JRA-55.

(a) 4 (b) 4 (c) 4 1


R 2 = 0.52 R 2 = 0.46 R 2 = 0.47
0.8

2 2 2

Density
0.6

0.4
0 0 0
0.2

–2 –2 –2 0
–0.5 0 0.5 1 –0.5 0 0.5 1 –0.5 0 0.5 1

(d) 4 2
(e) 4 2
1
R = 0.52 R = 0.49
0.8

2 2
Density

0.6

0.4
0 0
0.2

–2 –2 0
–0.5 0 0.5 1 –0.5 0 0.5 1

F I G U R E 4 Density scatter plot of mean surface temperature anomalies on the global grid (north of 60 S) for 1958–2010 from each of
five reanalysis datasets, (a) ERA5, (b) NCEP1, (c) CERA-20C, (d) 20CRv3, (e) JRA-55, and observation. The solid line shows the linear least
squares relationship between the reanalysis and the observed, the dash line indicates 1–1. R2 is the coefficient of determination [Colour
figure can be viewed at wileyonlinelibrary.com]
YANG ET AL. 9

90°N 90°N 90°N


60°N 60°N 60°N
30°N 30°N 30°N
0 0 0
30°S 30°S 30°S
60°S 60°S 60°S
180 120°W 60°W 0 60°E 120°E 180 180 120°W 60°W 0 60°E 120°E 180 180 120°W 60°W 0 60°E 120°E 180

90°N 90°N
60°N 60°N
30°N 30°N
0 0
30°S 30°S
60°S 60°S
180 120°W 60°W 0 60°E 120°E 180 180 120°W 60°W 0 60°E 120°E 180

F I G U R E 5 Correlations of annual surface temperature anomalies between (a) ERA5, (b) NCEP1, (c) CERA-20C, (d) 20CRv3, (e) JRA-55
and CMST-Interim for 1958–2010. The results are significant at the 95% confidence level using a Student's t test [Colour figure can be viewed
at wileyonlinelibrary.com]

F I G U R E 6 (a) Annual, (b) March–April–May (MAM), (c) June–July–August (JJA), (d) September–October–November (SON), and
(e) December–January–February (DJF) PDFs skill scores of surface temperature anomalies in five reanalysis datasets for 1958–2010 over the
globe (north of 60 S) [Colour figure can be viewed at wileyonlinelibrary.com]

3.2.3 | PDF skill scores datasets. The annual probabilities are more consistent
than the seasonal (Figure S5). The PDFs of ERA5, CERA-
PDFs for annual and seasonal temperature anomalies at 20C, and 20CRv3 are more consistent with the observed
each grid (north of 60 S) are computed from 1958 to PDF. NCEP/NCAR R1 has a slightly worse seasonal per-
2010. Observed seasonal temperature anomalies exhibit a formance. It is accordant with the results of skill scores
slightly higher probability around 0 C than reanalysis for PDFs (Figure 6). On the whole, reanalysis datasets
10 YANG ET AL.

provide good representations of the global annual and calculated over the common period 1958–2010
seasonal temperature distributions. Skill scores for PDFs (Figure 7c). Differences between all reanalysis data and
are generally larger than 0.9. The skill scores are larger in the observation are shown in Figure 7b,d. As can be seen
ERA5, CERA-20C, and 20CRv3, which reach about 0.95 from the annual temperature anomalies, all the
or above for the PDFs of annual and seasonal tempera- reanalysis datasets show an obviously warming trend,
ture anomalies. and the interannual variability also agrees well with
CMST-Interim (Figure 7a,b). CERA-20C and 20CRv3
have noticeably larger amplitudes before the 1940s but
3.2.4 | Comparison of the time series smaller amplitudes around 1940, compared with CMST-
between reanalysis datasets and CMST-Interim Interim. The overall deviations of ERA5 and 20CRv3 are
less than that of others. There is a slightly large ampli-
Global mean surface temperature anomalies tude in CERA-20C after the 1990s, while NCEP/NCAR
To display the annual STA variations of reanalysis R1 and JRA-55 show smaller amplitudes than the obser-
datasets for as long as possible, we evaluate the global vation (Figure 7b). ERA5, CERA-20C, and 20CRv3
mean (north of 60 S) annual STA for long time records describe better than NCEP/NCAR R1 and JRA-55 in
over a long common period of 1901–2018, which not only monthly temperature changes, with slightly small ampli-
includes the variations of CERA-20C, and 20CRv3 on the tude in spring and summer and large amplitude in
century time scales, but also exhibits the performance of autumn and winter. NCEP/NCAR R1 and JRA-55 show
NCEP/NCAR R1, ERA5, and JRA-55 in recent decades. small amplitudes in all months (Figure 7c,d).
For these relatively short records, they are shown in their
respective starting years (Figure 7a). The common period Regional surface temperature anomalies
1958–2010 is labelled with vertical lines in Figures 7–9 to Following the regional division by Xu et al. (2018),
highlight the performance of these datasets in the same we compared annual land surface air temperature varia-
period. Again, monthly temperature anomalies are tions in different regions (Figure 8). Compared with

F I G U R E 7 (a) The annual mean of 1901–2018 and (c) monthly mean of 1958–2010 for surface temperature anomalies from five
reanalysis datasets and CMST-Interim over the globe (north of 60 S). Differences in the (b) annual mean and (d) monthly mean data
between reanalysis datasets and CMST-Interim. The vertical lines indicate the common period 1958–2010 [Colour figure can be viewed at
wileyonlinelibrary.com]
YANG ET AL. 11

F I G U R E 8 (a–g) Annual surface temperature anomalies from five reanalysis datasets and CMST-Interim for 1901–2018 in different
regions. Differences in the (a1–g1) annual STA between reanalysis datasets and CMST-Interim for 1901–2018. The vertical lines indicate the
common period 1958–2010 [Colour figure can be viewed at wileyonlinelibrary.com]

CMST-Interim in Asia (5 –60 N, 60 –180 E), reanalysis reanalyses and the observation from 1958 onwards,
datasets overestimate surface temperatures before the which is similar to the results of those obtained by Bell
1940s, afterward slightly underestimate temperatures in et al. (2021) for comparison between ERA-5, JRA-55 and
the past decade. Especially, NCEP/NCAR R1 shows an observational climatological datasets for the 1950–2020
obvious warm period from 1948 to 1955 (Figure 8a). period, but the defined range for Europe is slightly differ-
Averaged differences over Asia are relatively smaller in ent. There are large discrepancies between observations
ERA5, CERA-20C, and JRA-55 (Figure 8a1). Over Africa and reanalysis datasets conspicuously prior to the 1940s.
(40 S–35 N, 20 W–45 E), CERA-20C slightly underesti- CERA-20C and 20CRv3 exhibit large positive differences
mates temperature anomalies during 1901–1940. In relative to CMST-Interim, by up to 1 C in 1920. Similarly,
ERA5, temperature anomalies are underestimated over NCEP/NCAR R1 also shows too warm anomalies from
1950–1960, afterward overestimated in the past decade. 1948 to 1958. Differences in temperature anomalies for
Compared to the observations, NCEP/NCAR R1 and all reanalysis datasets slightly decrease from 1960
CERA-20C consistently underestimate the anomalies in onwards, except that ERA5 slightly overestimates warm
the past three decades (Figure 8b, b1). Reanalysis datasets anomalies in the past two decades (Figure 8d, d1).
do not agree on the temperature evolution over South Regarding North America (15 –60 N, 140 –50 W)
America (55 S–15 N, 30 –80 E). For instance, CERA-20C (Figure 8e, e1), CERA-20C and 20CRv3 demonstrate
and 20CRv3 overestimate the anomalies in 1901–1960. In warm differences by up to 1 C prior to 1950. From 1960
NCEP/NCAR R1, temperature anomalies are not much onwards, differences between reanalysis data and obser-
different from CMST-Interim prior to 1990, then under- vations profoundly decrease. The overall temperature dif-
estimated from 1990 onwards. CERA-20C shows a large ferences in ERA5 and JRA-55 are smaller compared to
amplitude in the mid-1980s (Figure 8c, c1). In Europe other datasets. Reanalysis datasets exhibit larger differ-
(35 –60 N, 15 W–60 E), the good agreements between ences from CMST-Interim in Australia (15 –50 S, 110 –
12 YANG ET AL.

F I G U R E 9 (a–e) Annual surface temperature anomalies from five reanalysis datasets and CMST-Interim for 1901–2018 in different
latitudes. Differences in (a1–e1) annual STA between reanalysis datasets and CMST-Interim for 1901–2018. The vertical lines indicate the
common period 1958–2010 [Colour figure can be viewed at wileyonlinelibrary.com]

150 W) compared to other regions, particularly over increased significantly (Figure 9e, e1), especially before
recent years (Figure 8f, f1). For ERA5, there is a large 1950 for CERA-20C and 20CRv3. The deviation in the
positive difference in the earliest period of its coverage, past 20 years has shown an increasing trend. Although
which is partly due to insufficient initialization of deep the above STA deviations are relatively small from 1960
soil moisture (Bell et al., 2021). However, 20CRv3 shows to 1990, it is related to the selection of this period as the
smaller differences than other datasets in the common climatological reference period.
period. Over the land in the northern high latitudes (60 –
90 N, land) (Figure 8g, g1), CERA-20C and 20CRv3 show
a warm deviation in 1901–1930. From 1960 onwards, 3.3 | Trends analysis
except for 20CRv3, the deviations of the rest reanalysis
data are relatively small. Spatial distribution
Figure 9 shows the annual temperature variations The temporal trend and its spatial distribution for the
and deviations in different latitudes from 1901 to 2018. common periods 1958–2010 are examined. Figure 10
At low latitudes (Figure 9a, a1 and b, b1), the annual var- shows the spatial distribution of the trends in STA from
iations of CMST-Interim and reanalysis data are more CMST-Interim and reanalysis datasets. As can be seen in
consistent, especially after 1960. The deviation between Figure 10a, the global warming trend is obvious, particu-
reanalysis datasets and observation is within ±0.2 C. larly in Eurasian and northern North America, which is
Before 1960, CERA-20C and 20CRv3 show large ampli- consistent with the results obtained by Simmons et al.
tudes. The deviation at mid-latitudes in the Northern (2004) and Vose et al. (2005), although their study period
Hemisphere (Figure 9c, c1) is similar to that at low lati- is earlier, and it is also similar to Simmons et al. (2021)
tudes. NCEP/NCAR R1 has a significantly large ampli- for the recent 40-year period from 1979 to 2018 for the
tude from 1948 to 1958. CERA-20C and 20CRv3 show ERA5 and JRA-55. The trends from the reanalysis data
large amplitudes before 1940, and the deviations are sig- are generally consistent with the observation, and only in
nificantly reduced after 1960. At southern mid-latitudes, sporadic areas are significantly different (Figure 10b–f).
CERA-20C and 20CRv3 show small amplitudes before Among all datasets, ERA5 captures most of these
1950, and then the deviations of all reanalysis are rela- warming trends, only the trends of sporadic places differ
tively small (Figure 9d, d1). Compared to low and middle significantly from observation, such as very small areas
latitudes, the deviation at northern high latitudes has in the Arabian Peninsula and northern South America,
YANG ET AL. 13

and in the ocean. The warming trend of 20CRv3 data is CERA-20C (0.143 ± 0.024 Cdecade−1), and 20CRv3
also similar to the observation, in which there are obvi- (0.136 ± 0.020 Cdecade−1) agree better with that from
ous differences in the trends of the small regions of the CMST-Interim (0.145 ± 0.019 Cdecade−1), while NCEP/
Barents Sea, Greenland Sea, southernmost South Amer- NCAR R1 and JRA-55 (0.118 ± 0.023 and 0.124 ±
ica, and mid-latitude in the Southern Hemisphere. In 0.018 Cdecade−1, respectively) are slightly underestimated.
addition, there are a few areas with significant differences The seasonal trends further indicate that trends obtained
in CERA-20C trends, which generally appear in the from ERA5, CERA-20C, and 20CRv3 are closer to the
ocean. In contrast, NCEP/NCAR R1 data misses the observed results than the other reanalysis datasets, while
warming in the small regions of Sahara, northern South these are slightly underestimated in NCEP/NCAR R1 and
America, southeast Australia, the Tibet Plateau, and the JRA-55. The autumn and winter trends of CERA-20C are
ocean near the Arctic and mid-latitudes of the Southern slightly overestimated.
Hemisphere. JRA-55 data fails to capture the observed Figure 11 shows the heat map of the trends of
warming trend in northern South America, southern annual STA obtained from CMST-Interim and re-
Australia, some areas on the ocean (Figure 10b–f). The analysis datasets in different regions and latitudes from
surface temperature trends in the reanalysis datasets do 1958 to 2010. Considering the confidence interval, the
not show much seasonality, neither do their differences warming trends in the reanalysis data are broadly con-
from the observations (see Figure S6). Among them, sistent with the observation. Only CERA-20C and
there are significant differences in the warming trend of 20CRv3, like CMST-Interim, have significant warming
western Africa and southern mid-latitude in JJA obtained trends over Australia. The trend over South America of
by CERA-20C (see Figure S6, d2). NCEP/NCAR R1 is also insignificant (as shown in
Table S2). From the perspective of the trend magnitude,
Global and regional change NCEP/NCAR R1 and JRA-55 slightly underestimate the
Temporal trends of global (north 60 S) annual and seasonal trends at the latitude zones except for the northern high
surface temperature anomaly during 1958–2010 and their latitude. It is noteworthy that the trends of reanalysis
uncertainties are listed in Table 2. In general, the trends of datasets at northern high latitudes are stronger than the
annual and seasonal STA obtained from the reanalysis observed warming trend, as reported by Przybylak and
datasets are not significantly different from the observation. Wyszyn  ski (2020) that reanalyses overestimate Arctic
From the aspect of the trend magnitude, the warming warming. However, ERA5 (0.361 ± 0.074 Cdecade−1)
trends estimated from ERA5 (0.138 ± 0.022 Cdecade−1), and 20CRv3 (0.318 ± 0.075 Cdecade−1) still approach

90°N 90°N 90°N


60°N 60°N 60°N
30°N 30°N 30°N
0 0 0
30°S 30°S 30°S
60°S 60°S 60°S
180 120°W 60°W 0 60°E 120°E 180 180 120°W 60°W 0 60°E 120°E 180 180 120°W 60°W 0 60°E 120°E 180

90°N 90°N 90°N


60°N 60°N 60°N
30°N 30°N 30°N
0 0 0
30°S 30°S 30°S
60°S 60°S 60°S
180 120°W 60°W 0 60°E 120°E 180 180 120°W 60°W 0 60°E 120°E 180 180 120°W 60°W 0 60°E 120°E 180

(Trend: °C·decade−1)

F I G U R E 1 0 The trend of surface temperature anomalies from (a) CMST-Interim, (b) ERA5, (c) NCEP1, (d) CERA-20C, (e) 20CRv3, and
(f) JRA-55 for 1958–2010 (unit:  Cdecade−1). The shade indicates the spatial distribution of significant differences in trends between
reanalysis datasets and CMST-Interim [Colour figure can be viewed at wileyonlinelibrary.com]
14 YANG ET AL.

T A B L E 2 The trends of annual and seasonal global surface temperature anomalies (north of 60 S) and their uncertainties (at the 95%
confidence interval) from different datasets for 1958–2010 (unit:  Cdecade−1)

Annual MAM JJA SON DJF


CMST-Interim 0.145 ± 0.019 0.151 ± 0.021 0.140 ± 0.019 0.142 ± 0.022 0.143 ± 0.029
ERA5 0.138 ± 0.022 0.143 ± 0.024 0.120 ± 0.022 0.139 ± 0.026 0.142 ± 0.031
NCEP1 0.118 ± 0.023 0.131 ± 0.025 0.115 ± 0.023 0.113 ± 0.028 0.100 ± 0.033
CERA-20C 0.143 ± 0.024 0.137 ± 0.024 0.120 ± 0.025 0.149 ± 0.027 0.154 ± 0.031
20CRv3 0.136 ± 0.020 0.132 ± 0.023 0.118 ± 0.020 0.148 ± 0.023 0.137 ± 0.029
JRA-55 0.124 ± 0.019 0.131 ± 0.021 0.106 ± 0.019 0.120 ± 0.022 0.126 ± 0.029

0.5 F I G U R E 1 1 Heat map for


JRA-55 the trends of surface temperature
0.45
anomalies from 1958 to 2010 in
0.4 seven regions and at different
20CRv3
0.35 latitudes in five reanalysis

(Trend: °C·decade−1)
datasets and CMST-Interim
0.3
CERA-20C (unit:  Cdecade−1) [Colour figure
0.25 can be viewed at
NCEP1 wileyonlinelibrary.com]
0.2

0.15
ERA5
0.1

0.05
CMST-Interim
0

sia a
ric eric
a pe rica ralia d) H H H H H
A Af u ro e s t ( l an de N de N de N de S de S
A m E A m u t i c t u t u t u i t u i t u
A c ti ti ti t t
uth rth Ar La La La La La
So No o w
M id- igh- ow Mid-
L H L

the observed (0.290 ± 0.051 Cdecade−1), while CERA- a benchmark for ranking comparisons. As far as the first
20C (0.383 ± 0.067 Cdecade−1), JRA-55 (0.393 ± 0.077 C 10 of the warmest and coldest years are concerned, these
decade−1), and NCEP/NCAR R1 (0.400 ± 0.083 C extreme years are well represented in each reanalysis
decade−1) are stronger relative to the observation. As the dataset. At least 8 years are consistent with observation,
surface temperatures over sea ice at northern high latitudes although the sequences are not completely matched.
are set to a fixed value of −1.8 C in CMST-Interim, which ERA5 is able to capture all coldest years revealed in the
may account for an underestimation of the warming trend observational average, while 20CRv3 could reproduce
in the Arctic. However, from the global average (Table 2), 10 warmest years. In terms of the first 3 years, CERA-20C
this underestimation is not significant and requires further successfully reproduces the first three warmest years as
analysis. in averaged observations, but in a different order from
that detected by the observational average. ERA5,
NCEP/NCAR R1, and 20CRv3 can also capture the first
3.4 | The consistency of the years of two warmest years, and JRA-55 only captures the
extreme temperature warmest year.
In the warmest year ranking, the same years as the
Table 3 shows the rankings of temperature anomaly in observed ranking from reanalysis datasets are 2 (JRA-55),
the 10 warmest and 10 coldest years in 5 reanalysis 4 (NCEP/NCAR R1), 4 (20CRv3), 5 (ERA5), and
datasets from 1958 to 2010. To make ranking results 5 (CERA-20C). The total number of the recreated warm-
more reliable, the average of annual surface temperature est years is 8, 9, 10, 9, and 9. Furthermore, the ranking
anomalies from three observation data is calculated from scores of five reanalysis data are 87 (JRA-55), 91 (NCEP/
CMST-Interim, HadCRUT5, and NOAAGlobalTemp5, as NCAR R1), 92 (CERA-20C), 92 (20CRv3), and 94 (ERA5).
YANG ET AL.

TABLE 3 The ranking of the warmest and coldest year in the observation and five reanalysis datasets for the period of 1958–2010

Ten warmest years Ten coldest years

Average ERA5 NCEP1 CERA-20C 20CRv3 JRA-55 Average ERA5 NCEP1 CERA-20C 20CRv3 JRA-55
1 2010 2010** 2010** 2010** 2010** 2010** 1964 1976 1964** 1976 1964** 1964**
2 2005 2005** 2005** 2005** 2005** 1998 1971 1964* 1976 1974* 1965 1976
3 2006 1998 2007 2006** 2007 2005* 1974 1971* 1965* 1964 1976 1974**
4 2009 2006* 2006* 2007 2006* 2006* 1965 1974* 1971 1971 1974* 1971
5 1998 2009* 2009* 1998** 2009* 2009* 1976 1975 1974 1975 1971 1965*
6 2007 2007** 1998* 2009 1998* 2007** 1968 1965 1975 1968** 1966* 1968**
7 2003 2003** 2003** 2002* 2002* 2002* 1966 1968* 1966** 1965 1967* 1967*
8 2002 2002** 2002** 2003* 2003* 2003* 1967 1972 1963 1967** 1960 1966*
9 2001 2004* 2008 2008 2001** 1995 1975 1966 1968 1972* 1968 1975**
10 2004 2008* 2001* 2004** 2004** 1997 1972 1967 1962 1970* 1975* 1962
Total number 9 9 9 10 8 10 8 9 9 9
Score 94 91 92 92 87 80 73 82 82 87

Note: Average indicates the ranking of the average temperature from three observations (CMST-Interim, HadCRUT5, and NOAAGlobalTemp5); total number stands for the number of reanalysis ranking years
consistent with the first 10 extreme ranking years from the average; score represents the ranking score of reanalysis datasets. (**) indicates that the order of ranked year from reanalysis datasets is consistent with that
from the average, (*) represents that the order of the ranked year from reanalysis datasets is one place away from that of the average.
15
16 YANG ET AL.

It can be seen that the order of the year of extremely high the observations as possible. As another climate
temperature by ERA5 is more consistent with observa- reanalysis data to span more than 100 years, although
tion. As discussed by Simmons et al. (2017) that the the seasonal biases (errors) are slightly larger than that of
reanalyses are quite different from the observational cli- the previous two, CERA-20C also performs well, espe-
matological datasets in the magnitude estimates of indi- cially in long-term variability and the warming trend. It
vidual warm and cold spells, it is related to very small indicates that the performance of 4D-Var and EnKF tech-
but nonignorable differences from air temperature over nologies seems to be comparable when a reanalysis
sea and SST, and it also affects the ranking of the set of assimilates only “surface-input”, as stated by Whitaker
warmest years. Although the comparison here is based et al. (2009). Although NCEP/NCAR R1, released in
on the ranking scores about the consistency of years, 1995, is produced using a modified 1995 version of the
rather than the value of STA itself, the very similar scores NCEP forecast model with 3D-FGAT (first guess at the
indicate that the reanalyses are good representative in appropriate time) as data assimilation, the results of the
warmest calendar years. While for the coldest years, representative of the warmest years are still relatively
except ERA5 and CERA-20C, the other three reanalysis consistent with the observations. The warmest years
data can reproduce the first year. Only JRA-55 agrees occurred in the late 20th or early 21st century, and the
with the third coldest year of the observational average. data assimilation is also large in this period. Although
Additionally, ERA5 can reproduce the same 10 years as the anomalies of NCEP/NCAR R1 are lower than obser-
the observed result but in a different order, CERA-20C, vations since 1990, the rank of the warmest years is con-
20CRv3, and JRA-55 reproduce 9 years, and NCEP/ sistent well. Moreover, JRA-55 could reproduce most of
NCAR R1 is 8 years. The ranking scores are 73 (NCEP/ the coldest years that match or differ one place from
NCAR R1), 80 (ERA5), 82 (CERA-20C), 82 (20CRv3), and observed ranking results but perform poorly in the warm-
87 (JRA-55), respectively, which indicates that JRA-55 is est years. The results show that JRA-55 has good agree-
slightly better than other reanalysis datasets in the repro- ment with observation in the coldest years that appear
duction of year of extremely low temperature. around the 1970s, during which there are relatively fewer
assimilation data. On the contrary, JRA-55 also has sys-
tematic negative biases similar to that of NCEP/NCAR
4 | DISCUSSION AND R1 after 1990 when more data are assimilated. It is likely
C O N C L U S IO N due to the STA calculated based on the period 1961–1990
during which there are fewer assimilated observations.
The agreement of reanalysis air temperatures from Global trends and interannual variation in surface
ERA5, NCEP/NCAR R1, CERA-20C, 20CRv3, and JRA- temperature anomalies from different reanalyses are gen-
55 with newly released global surface temperature anom- erally consistent with the CMST-Interim since 1958, and
aly dataset CMST-Interim is evaluated from many the same is true for 20CRv3 and CREA-20C from 1901
aspects, including spatiotemporal variations of biases, onwards. However, there are still significant regional dif-
error estimation, long-term trend analysis, and the repre- ferences between reanalyses and the observation. It partly
sentation of the occurrence of extreme annual tempera- lies in the lack of regional observations at some periods,
ture. Results indicate that all five reanalysis data are in which may lead to regional blemishes in the reanalyses.
good agreement with CMST-Interim in the above aspects, For example, as described by Simmons et al. (2021) that
where ERA5 and 20CRv3 are generally more consistent observations over Asia, South America, and Africa are
than the other reanalysis data. It is worth noting that insufficient in ERA5's input database from 1950 to 1957;
these two reanalysis datasets represent two different observations on islands and the coast are not used in the
schemes of data assimilation, the “full-input” (surface screen-level analysis of JRA-55, which may cause the dif-
and upper-air conventional and satellite data) 4D-Var ference in those areas where are affected by the observa-
and the “surface-input” (surface data only) ensemble tions in coastal waters (Kobayshi et al., 2015). Another
Kalman filter (EnKF) approach, respectively. From the source of differences between observed data and
comparison to the observations (represented by CMST- reanalyses is the difference in SST and marine air tem-
Interim), it seems to indicate that there may be no signifi- perature (Simmons et al., 2017, 2021; Bell et al., 2021).
cant difference in the representative of the reanalysis The observations provide a combination of land surface
from the perspective of long-term surface temperature air temperature and SST, while the reanalyses present air
change, with or without assimilation of satellite informa- temperature everywhere. Bell et al. (2021) pointed that
tion. While ERA5 could better reproduce the observed ERA5 has by a small margin the highest temperatures
ordering of global surface temperature changes, which relative to 1981–2010 for the latest years partly due to
may be related to the fact that it assimilates as much of this, and the trend would be smaller if SST is used in
YANG ET AL. 17

reanalyses (Simmons et al., 2017). In addition, as been Writing – review and editing. Wenbin Sun: Data
discussed in previous studies, the coverage of observa- curation. Wenjie Dong: Writing – review and editing.
tions in the Antarctic and Arctic regions is incomplete, so
the sources of surface temperature uncertainties in high-
latitude areas are mainly from coverage bias or the bias ORCID
from the treatment of temperature at the sea ice bound- Yang Yang https://orcid.org/0000-0002-8143-3865
ary (Cowtan and Way, 2014; Cowtan et al., 2015; Morice
et al., 2021; Osborn et al., 2021), this may be the cause of RE FER EN CES
the large deviation at high latitudes between the observa- Balmaceda-Huarte, R., Olmo, M.E., Bettolli, M.L. and Poggi, M.M.
(2021) Evaluation of multiple reanalyses in reproducing the
tion and reanalyses. It should be pointed out that there
spatio-temporal variability of temperature and precipitation indi-
are also some problems about uncertainties at high lati- ces over southern South America. International Journal of Clima-
tudes in the CMST-Interim. For example, the tempera- tology, 41(12), 5572–5595. https://doi.org/10.1002/joc.7142.
ture in the Arctic sea ice covered area is set to −1.8 C Bell, B., Hersbach, H., Simmons, A., Berrisford, P., Dahlgren, P.,
(regarded as the default value in the calculation), and it Horanyi, A., Muñoz-Sabater, J., Nicolas, J., Radu, R.,
can be seen that the biases (errors) are large in the north- Schepers, D., Soci, C., Villaume, S., Bidlot, J.-R., Haimberger, L.,
ern high latitudes. And due to the scarcity of observation Woollen, J., Buontempo, C. and Thépaut, J.-N. (2021) The ERA5
global reanalysis: preliminary extension to 1950. Quarterly Jour-
data, the interannual variability of reconstructed data
nal of the Royal Meteorological Society, 147(741), 4186–4227.
near Antarctica is significantly reduced, so it is not
https://doi.org/10.1002/qj.4174.
involved in the previous comparisons and needs further Beusch, L., Gudmundsson, L. and Seneviratne, S.I. (2020) Cross-
refinement in the future. The reconstruction of CMST- breeding CMIP6 Earth System models with an emulator for
Interim on the sea ice cover and unobserved areas near regionally optimized land temperature projections. Geophysical
the poles needs to be further improved. Therefore, these Research Letters, 47, e2019GL086812. https://doi.org/10.1029/
are only preliminary comparisons of surface air tempera- 2019GL086812.
ture reanalysis datasets, and it is hoped that the results of Chao, L.Y., Huang, B.Y., Yang, Y.J., Jones, P., Cheng, J., Yang, Y.
and Li, Q. (2020) A new evaluation of the role of urbanization
this comparison could provide a better understanding of
to warming at various spatial scales: evidence from the
the differences among these data and a reference for the
Guangdong-Hong Kong-Macau region, China. Geophysical
appliance of surface air temperature reanalysis datasets. Research Letters, 47, e2020GL089152. https://doi.org/10.1029/
2020GL089152.
ACK NO WLE DGE MEN TS Cook, B.I., Mankin, J.S., Marvel, K., Williams, A.P., Smerdon, J.E.
This study was supported by the National Key R&D Pro- and Anchukaitis, K.J. (2020) Twenty-first century drought pro-
gram of China (Grant Nos. 2017YFC1502301 and jections in the CMIP6 forcing scenarios. Earth's Futures, 8,
2018YFC1507705) and the National Natural Science e2019EF001461. https://doi.org/10.1029/2019EF001461.
Cowtan, K., Hausfather, Z., Hawkins, E., Jacobs, P., Mann, M.E.,
Foundation of China (Grant No. 41975105). The authors
Miller, S.K., Steinman, B.A., Stolpe, M.B. and Way, R.G. (2015)
are thankful to ECMWF, JMA, NCEP, and NCAR for Robust comparison of climate models with observations using
providing ERA5, CERA-20C, JRA-55, NCEP/NCAR R1, blended land air and ocean sea surface temperatures. Geophysi-
and 20th Century Reanalysis V3 datasets for this study. cal Research Letters, 42(15), 6526–6534. https://doi.org/10.1002/
ERA5 and CERA-20C data can be obtained from the 2015GL064888.
link https://cds.climate.copernicus.eu/ and https://apps. Cowtan, K. and Way, R.G. (2014) Coverage bias in the HadCRUT4
ecmwf.int/datasets/, respectively. NCEP/NCAR R1 temperature series and its impact on recent temperature trends.
Quarterly Journal of the Royal Meteorological Society, 140, 1935–
Reanalysis data and NOAA-CIRES-DOE 20th Century
1944. https://doi.org/10.1002/qj.2297.
Reanalysis V3 data can be obtained from https://psl.
Dee, D.P., Uppala, S.M., Simmons, A.J., Berrisford, P., Poli, P.,
noaa.gov/data/gridded/. The JRA-55 data are obtained Kobayashi, S., Andrae, U., Balmaseda, M.A., Balsamo, G.,
from the link https://rda.ucar.edu/datasets/ds628.1/. Bauer, P., Bechtold, P., Beljaars, A.C.M., van de Berg, L.,
CMST-Interim dataset is available from http://www. Bidlot, J., Bormann, N., Delsol, C., Dragani, R., Fuentes, M.,
gwpu.net/h-col-103.html. Except that CERA-20C is used Geer, A.J., Haimberger, L., Healy, S.B., Hersbach, H., H olm, E.
the version available in September 2021, other datasets V., Isaksen, L., Kållberg, P., Köhler, M., Matricardi, M.,
are used the version available in November 2020. McNally, A.P., Monge-Sanz, B.M., Morcrette, J.-J., Park, B.-K.,
Peubey, C., de Rosnay, P., Tavolato, C., Thépaut, J.-N. and
Vitart, F. (2011) The ERA-Interim reanalysis: confifiguration
A U T H O R C ON T R I B U T I O NS
and performance of the data assimilation system. Quarterly
Yang Yang: Data curation; formal analysis; software; Journal of the Royal Meteorological Society, 137, 553–597.
writing – original draft; writing – review and editing. https://doi.org/10.1002/qj.828.
Qingxiang Li: Conceptualization; writing – original Engdaw, M.M., Ballinger, A.P., Hegerl, G.C. and Steiner, A.K.
draft; writing – review and editing. Zhaoyang Song: (2022) Changes in temperature and heat waves over Africa
18 YANG ET AL.

using observational and reanalysis data sets. International Jour- IPCC. (2013) In: Stocker, T.F., Qin, D., Plattner, G.-K., Tignor, M.,
nal of Climatology, 42(2), 1165–1180. https://doi.org/10.1002/ Allen, S.K., Boschung, J., Nauels, A., Xia, Y., Bex, V. and
joc.7295. Midgley, P.M. (Eds.) Climate Change 2013: The Physical Science
Goddard, I.L.M. and Tett, S.F.B. (2019) How much has urbanisation Basis. Contribution of Working Group I to the Fifth Assessment
affected United Kingdom temperatures? Atmospheric Science Report of the Intergovernmental Panel on Climate Change. Cam-
Letters, 20, e896. https://doi.org/10.1002/asl.896. bridge and New York, NY: Cambridge University Press,
Gulev, S.K., Thorne, P.W., Ahn, J., Dentener, F.J., Domingues, C. 1535 pp.
M., Gerland, S., Gong, D., Kaufman, D.S., Nnamchi, H.C., Jones, P.D. and Harpham, C. (2013) Estimation of the absolute sur-
Quaas, J., Rivera, J.A., Sathyendranath, S., Smith, S.L., face air temperature of the Earth. Journal of Geophysical
Trewin, B., von Shuckmann, K. and Vose, R.S. (2021) Changing Research, 118, 3213–3217. https://doi.org/10.1002/jgrd.50359.
state of the climate system. In: Masson Delmotte, V., Zhai, P., Kadow, C., Hall, D.M. and Ulbrich, U. (2020) Artificial intelligence
Pirani, A., Connors, S.L., Péan, C., Berger, S., Caud, N., reconstructs missing climate information. Nature Geoscience,
Chen, Y., Goldfarb, L., Gomis, M.I., Huang, M., Leitzell, K., 13, 408–413. https://doi.org/10.1038/s41561-020-0582-5.
Lonnoy, E., Matthews, J.B.R., Maycock, T.K., Waterfield, T., Kalnay, E., Kanamitsu, M., Kistler, R., Collins, W., Deaven, D.,
Yelekç, O., Yu, R. and Zhou, B. (Eds.) Climate Change 2021: Gandin, L., Iredell, M., Saha, S., White, G., Woollen, J., Zhu, Y.,
The Physical Science Basis. Contribution of Working Group I to Chelliah, M., Ebisuzaki, W., Higgins, W., Janowiak, J., Mo, K.
the Sixth Assessment Report of the Intergovernmental Panel on C., Ropelewski, C., Wang, J., Leetmaa, A., Reynolds, R.,
Climate Change. Cambridge: Cambridge University Press. Jenne, R. and Joseph, D. (1996) The NCEP/NCAR 40-year
(in press). reanalysis project. Bulletin of the American Meteorological Soci-
Hersbach, H., Bell, B., Berrisford, P., Hirahara, S., Horanyi, A., ety, 77(3), 437–471. https://doi.org/10.1175/1520-0477(1996)
Muñoz, S.J., Nicolas, J., Peubey, C., Radu, R., Schepers, D., 077<0437:TNYRP>2.0.CO;2.
Simmons, A., Soci, C., Abdalla, S., Abellan, X., Balsamo, G., Kanamitsu, M., Ebisuzaki, W., Woollen, J., Yang, S.-K., Hnilo, J.J.,
Bechtold, P., Biavati, G., Bidlot, J., Bonavita, M., Chiara, G., Fiorino, M. and Potter, G.L. (2002) NCEP-DOE AMIP-II
Dahlgren, P., Dee, D., Diamantakis, M., Dragani, R., reanalysis (R-2). Bulletin of the American Meteorological Society,
Flemming, J., Forbes, R., Fuentes, M., Geer, A., 83, 1631–1643. https://doi.org/10.1175/BAMS-83-11-1631.
Haimberger, L., Healy, S., Hogan, R.J., Holm, E., Janiskova, M., Karl, T.R., Arguez, A., Huang, B., Lawrimore, J.H., McMahon, J.R.,
Keeley, S., Laloyaux, P., Lopez, P., Lupu, C., Radnoti, G., Menne, M.J., Peterson, T.C., Vose, R.S. and Zhang, H.M. (2015)
Rosnay, P., Rozum, I., Vamborg, F., Villaume, S. and Possible artifacts of data biases in the recent global surface
Thépaut, J.N. (2020) The ERA5 global reanalysis. Quarterly warming hiatus. Science, 348(6242), 1469–1472. https://doi.org/
Journal of the Royal Meteorological Society, 146(730), 1999– 10.1126/science.aaa5632.
2049. https://doi.org/10.1002/qj.3803. Kistler, R., Kalnay, E., Collins, W., Saha, S., White, G., Woollen, J.,
Hu, Z., Zhang, C., Hu, Q. and Tian, H. (2014) Temperature changes Chelliah, M., Ebisuzaki, W., Kanamitsu, M., Kousky, V.,
in central Asia from 1979 to 2011 based on multiple datasets. Huug, V.D.D., Jenne, R. and Fiorino, M. (2001) The NCEP-
Journal of Climate, 27(3), 1143–1167. https://doi.org/10.1175/ NCAR 50-year reanalysis: monthly means CD-ROM and docu-
JCLI-D-13-00064.1. mentation. Bulletin of the American Meteorological Society,
Huai, B., Wang, J., Sun, W., Wang, Y. and Zhang, W. (2021) Evalua- 82(2), 247–268. https://doi.org/10.1175/1520-0477(2001)
tion of the near-surface climate of the recent global atmo- 082<0247:TNNYRM>2.3.CO;2.
spheric reanalysis for Qilian Mountains, Qinghai-Tibet Plateau. Kobayshi, S., Ota, Y., Harada, Y., Ebita, A., Moriya, M., Onoda, H.,
Atmospheric Research, 250, 105401. https://doi.org/10.1016/j. Onogi, K., Kamahori, H., Kobayshi, C., Endo, H., Miyaoka, K.
atmosres.2020.105401. and Takahashi, K. (2015) The JRA-55 reanalysis: general speci-
Huang, B., Menne, M.J., Boyer, T., Freeman, E., Gleason, B.E., fications and basic characteristics. Journal of the Meteorological
Lawrimore, J.H., Liu, C., Rennie, J.J., Schreck, C.J., III, Sun, F., Society of Japan Series II, 93(1), 5–48. https://doi.org/10.2151/
Vose, R., Williams, C.N., Yin, X. and Zhang, H. (2020) Uncer- jmsj.2015-001.
tainty estimates for sea surface temperature and land surface air Laloyaux, P., de Boisseson, E., Balmaseda, M., Bidlot, J.-R.,
temperature in NOAAGlobalTemp version 5. Journal of Climate, Broennimann, S., Buizza, R., Dalhgren, P., Dee, D.,
33(4), 1351–1379. https://doi.org/10.1175/JCLI-D-19-0395.1. Haimberger, L., Hersbach, H., Kosaka, Y., Martin, M., Poli, P.,
Huang, J., Zhang, X., Zhang, Q., Lin, Y., Hao, M., Luo, Y., Zhao, Z., Rayner, N., Rustemeier, E. and Schepers, D. (2018) CERA-20C:
Yao, Y., Chen, X., Wang, L., Nie, S., Yin, Y., Xu, Y. and a coupled reanalysis of the twentieth century. Journal of
Zhang, J. (2017) Recently amplified arctic warming has contrib- Advances in Modeling Earth Systems, 10, 1172–1195. https://doi.
uted to a continual global warming trend. Nature Climate org/10.1029/2018MS001273.
Change, 7, 875–879. https://doi.org/10.1038/s41558-017-0009-5. Li, Q., Sun, W., Yun, X., Huang, B., Dong, W., Wang, X., Zhai, P.
IPCC. (2001) Climate Change 2001: Synthesis Report. A Contribution and Jones, P. (2021) An updated evaluation of the global mean
of Working Groups I, II, and III to the Third Assessment Report land surface air temperature and surface temperature trends
of the Integovernmental Panel on Climate Change. Cambridge based on CLSAT and CMST. Climate Dynamics, 56, 635–650.
and New York, NY: Cambridge University Press, 398 pp. https://doi.org/10.1007/s00382-020-05502-0.
IPCC. (2007) Climate Change 2007: Synthesis Report. Contribution of Li, Q., Zhang, L., Xu, W., Zhou, T., Wang, J., Zhai, P. and Jones, P.
Working Groups I, II and III to the Fourth Assessment Report of (2017) Comparisons of time series of annual mean surface air
the Intergovernmental Panel on Climate Change. Geneva: IPCC, temperature for China since the 1900s: observation, model sim-
104 pp. ulation and extended reanalysis. Bulletin of the American
YANG ET AL. 19

Meteorological Society, 98, 699–711. https://doi.org/10.1175/ Saha, S., Moorthi, S., Pan, H.-L., Wu, X., Wang, J., Nadiga, S.,
BAMS-D-16-0092.1. Tripp, P., Kistler, R., Woollen, J., Behringer, D., Liu, H.,
Li, Q.X., Sun, W.B., Huang, B.Y., Dong, W.J., Wang, X.L., Zhai, P. Stokes, D., Grumbine, R., Gayno, G., Wang, J., Hou, Y.-T.,
M. and Jones, P. (2020) Consistency of global warming trends Chuang, H.-Y., Juang, H.-M.H., Sela, J., Iredell, M.,
strengthened since 1880s. Science Bulletin, 65(20), 1709–1712. Treadon, R., Kleist, D., van Delst, P., Keyser, D., Derber, J.,
https://doi.org/10.1016/j.scib.2020.06.009. Ek, M., Meng, J., Wei, H., Yang, R., Lord, S., van den Dool, H.,
Luo, H., Ge, F., Yang, K., Zhu, S., Peng, T., Cai, W., Liu, X. and Kumar, A., Wang, W., Long, C., Chelliah, M., Xue, Y.,
Tang, W. (2019) Assessment of ECMWF reanalysis data in com- Huang, B., Schemm, J.-K., Ebisuzaki, W., Lin, R., Xie, P.,
plex terrain: Can the CERA-20C and ERA-Interim data sets Chen, M., Zhou, S., Higgins, W., Zou, C.-Z., Liu, Q., Chen, Y.,
replicate the variation in surface air temperatures over Sichuan, Han, Y., Cucurull, L., Reynolds, R.W., Rutledge, G. and
China? International Journal of Climatology, 39(15), 5619–5634. Goldberg, M. (2010) The NCEP climate forecast system
https://doi.org/10.1002/joc.6175. reanalysis. Bulletin of the American Meteorological Society,
Ma, L., Zhang, T., Li, Q., Frauenfiel, O.W. and Qin, D. (2008) Evalu- 91(8), 1015–1057. https://doi.org/10.1175/2010BAMS3001.1.
ation of ERA-40, NCEP/NCAR R1, and NCEP-2 reanalysis air Simmons, A., Hersbach, H., Munoz-Sabater, J., Nicolas, J.,
temperatures with ground-based measurements in China. Jour- Vamborg, F., Berrisford, P., de Rosnay, P., Willett, K. and
nal of Geophysical Research, 113, D15115. https://doi.org/10. Woollen, J. (2021) Low frequency variability and trends in sur-
1029/2007JD009549. face air temperature and humidity from ERA5 and other
Mao, J., Shi, X., Ma, L., Kaiser, D., Li, Q. and Thornton, P. (2010) datasets. Reading: ECMWF, ECMWF technical memoran-
Assessment of reanalysis daily extreme temperatures with dum 881.
China's homogenized historical dataset during 1979–2001 using Simmons, A.J., Berrisford, P., Dee, D.P., Hersbach, H., Hirahara, S.
probability density functions. Journal of Climate, 23, 6605– and Thépaut, J.N. (2017) A reassessment of temperature varia-
6623. https://doi.org/10.1175/2010JCLI3581.1. tions and trends from global reanalyses and monthly surface
Morice, C.P., Kennedy, J.J., Rayner, N.A., Winn, J.P., Hogan, E., climatological datasets. Quarterly Journal of the Royal Meteoro-
Killick, R.E., Dunn, R.J.H., Osborn, T., Jones, P. and logical Society, 143, 101–119. https://doi.org/10.1002/qj.2949.
Simpson, I.R. (2021) An updated assessment of near-surface Simmons, A.J., Jones, P.D., da Costa Bechtold, V., Beljaars, A.C.M.,
temperature change from 1850: the HadCRUT5 dataset. Jour- Kållberg, P.W., Saarinen, S., Uppala, S.M., Viterbo, P. and
nal of Geophysical Research, 126(3), e2019JD032361. https://doi. Wedi, N. (2004) Comparison of trends and low-frequency vari-
org/10.1029/2019JD032361. ability in CRU, ERA-40, and NCEP/NCAR analyses of surface
Osborn, T.J., Jones, P.D., Lister, D.H., Morice, C.P., Simpson, I.R., air temperature. Journal of Geophysical Research, 109, D24115.
Winn, J.P., Hogan, E. and Harris, I.C. (2021) Land surface air https://doi.org/10.1029/2004JD005306
temperature variations across the globe updated to 2019: the Simmons, A.J. and Poli, P. (2015) Arctic warming in ERA-Interim
CRUTEM5 dataset. Journal of Geophysical Research: and other analyses. Quarterly Journal of the Royal Meteorologi-
Atmospheres, 126, e2019JD032352. https://doi.org/10.1029/ cal Society, 141, 1147–1162. https://doi.org/10.1002/qj.2422.
2019JD032352. Slivinski, L.C., Compo, G.P., Sardeshmukh, P.D., Whitaker, J.S.,
Perkins, S.E., Pitman, A.J., Holbrook, N.J. and McAneney, J. (2007) McColl, C., Allan, R.J., Brohan, P., Yin, X., Smith, C.A.,
Evaluation of the AR4 climate models' simulated daily maxi- Spencer, L.J., Vose, R.S., Rohrer, M., Conroy, R.P.,
mum temperature, minimum temperature, and precipitation Schuster, D.C., Kennedy, J.J., Ashcroft, L., Brönnimann, S.,
over Australia using probability density functions. Journal of Brunet, M., Camuffo, D., Cornes, R., Cram, T.A., Domínguez-
Climate, 20(17), 4356–4376. https://doi.org/10.1175/JCLI4253.1. Castro, F., Freeman, J.E., Gergis, J., Hawkins, E., Jones, P.D.,
Poli, P., Hersbach, H., Dee, D.P., Berrisford, P., Simmons, A.J., Kubota, H., Lee, T.C., Lorrey, A.M., Luterbacher, J., Mock, C.J.,
Vitart, F., Laloyaux, P., Tan, D.G.H., Peubey, C., Thépaut, J., Przybylak, R.K., Pudmenzky, C., Slonosky, V.C., Tinz, B.,
Trémolet, Y., H olm, E.V., Bonavita, M., Isaksen, L. and Trewin, B., Wang, X.L., Wilkinson, C., Wood, K. and
Fisher, M. (2016) ERA-20C: An Atmospheric Reanalysis of the Wyszy nski, P. (2021) An evaluation of the performance of the
Twentieth Century. Journal of Climate, 29(11), 4083–4097. Twentieth Century Reanalysis version 3. Journal of Climate,
https://doi.org/10.1175/JCLI-D-15-0556.1 34(4), 1417–1438. https://doi.org/10.1175/JCLI-D-20-0505.1.
Przybylak, R. and Wyszy nski, P. (2020) Air temperature changes in Slivinski, L.C., Compo, G.P., Whitaker, J.S., Sardeshmukh, P.D.,
the Arctic in the period 1951–2015 in the light of observational Giese, B.S., McColl, C., Allan, R., Yin, X., Vose, R., Titchner, H.,
and reanalysis data. Theoretical and Applied Climatology, 139, Kennedy, J., Spencer, L.J., Ashcroft, L., Brönnimann, S.,
75–94. https://doi.org/10.1007/s00704-019-02952-3. Brunet, M., Camuffo, D., Cornes, R., Cram, T.A.,
Qin, Y., Zhang, P., Liu, W., Guo, Z. and Xue, S. (2020) The applica- Crouthamel, R., Domínguez, C.F., Freeman, J.E., Gergis, J.,
tion of elevation corrected MERRA2 reanalysis ground surface Hawkins, E., Jones, P.D., Jourdain, S., Kaplan, A., Kubota, H.,
temperature in a permafrost model on the Qinghai-Tibet pla- Blancq, F.L., Lee, T.C., Lorrey, A., Luterbacher, J., Maugeri, M.,
teau. Cold Regions Science and Technology, 175, 103067. https:// Mock, C.J., Moore, G.W.K., Przybylak, R., Pudmenzky, C.,
doi.org/10.1016/j.coldregions.2020.103067. Reason, C., Slonosky, V.C., Smith, C.A., Tinz, B., Trewin, B.,
Retamales-Muñoz, G., Duran-Alarcon, C. and Mattar, C. (2019) Valente, M.A., Wang, X.L., Wilkinson, C., Wood, K. and
Recent land surface temperature patterns in Antarctica using Wyszy nski, P. (2019) Towards a more reliable historical
satellite and reanalysis data. Journal of South American Earth reanalysis: improvements for version 3 of the Twentieth Century
Sciences, 95, 102304. https://doi.org/10.1016/j.jsames.2019. Reanalysis system. Quarterly Journal of the Royal Meteorological
102304. Society, 145(724), 2876–2908. https://doi.org/10.1002/qj.3598.
20 YANG ET AL.

Sturaro, G. (2003) A closer look at the climatological discontinuities the global, hemispheric, and grid-box scale. Geophysical Research
present in the NCEP/NCAR reanalysis temperature due to the Letters, 32, L18718. https://doi.org/10.1029/2005GL023502.
introduction of satellite data. Climate Dynamics, 21, 309–316. Whitaker, J.S., Compo, G.P. and Thépaut, J.-N. (2009) A comparison
https://doi.org/10.1007/s00382-003-0334-4. of variational and ensemble-based data assimilation systems for
Sun, W., Li, Q., Huang, B., Cheng, J., Song, Z., Li, H., Dong, W., reanalysis of sparse observations. Monthly Weather Review, 137(6),
Zhai, P. and Jones, P. (2021) The assessment of global surface 1991–1999. https://doi.org/10.1175/2008MWR2781.1.
temperature change from 1850s: the C-LSAT2.0 ensemble and Xu, W., Li, Q., Jones, P., Wang, X.L., Trewin, B., Yang, S., Zhu, C.,
the CMST-Interim datasets. Advances in Atmospheric Sciences, Zhai, P., Wang, J., Vincent, L., Dai, A., Gao, Y. and Ding, Y.
38(5), 875–888. https://doi.org/10.1007/s00376-021-1012-3. (2018) A new integrated and homogenized global monthly land
Uppala, S.M., Kallberg, P.W., Simmons, A.J., Andrae, U., surface air temperature dataset for the period since 1900. Cli-
Bechtold, V.D.C., Fiorino, M., Gibson, J.K., Haseler, J., mate Dynamics, 50(7–8), 2513–2536. https://doi.org/10.1007/
Hernandez, A., Kelly, G.A., Li, X., Onogi, K., Saarinen, S., s00382-017-3755-1.
Sokka, N., Allan, R.P., Andersson, E., Arpe, K., Balmaseda, M. Yun, X., Huang, B., Cheng, J., Xu, W., Qiao, S. and Li, Q. (2019) A
A., Beljaars, A.C.M., Berg, L.V.D., Bidlot, J., Bormann, N., new merge of global surface temperature datasets since the
Caires, S., Chevallier, F., Dethof, A., Dragosavac, M., start of the 20th century. Earth System Science Data, 11(4),
Fisher, M., Fuentes, M., Hagemann, S., Holm, E., Hoskins, B.J., 1629–1643. https://doi.org/10.5194/essd-11-1629-2019.
Isaksen, L., Janssen, P.A.E.M., Jenne, R., Mcnally, A.P., Zhu, J., Xie, A., Qin, X., Wang, Y., Xu, B. and Wang, Y. (2021) An assess-
Mahfouf, J.-F., Morcrette, J.-J., Rayner, N.A., Saunders, R.W., ment of ERA5 reanalysis for Antarctic near-surface air temperature.
Simon, P., Sterl, A., Trenberth, K.E., Untch, A., Vasiljevic, D., Atmosphere, 12(2), 217. https://doi.org/10.3390/atmos12020217.
Viterbo, P. and Woollen, J. (2005) The ERA-40 re-analysis.
Quarterly Journal of the Royal Meteorological Society, 131, 2961–
3012. https://doi.org/10.1256/qj.04.176. SU PP O R TI N G I N F O RMA TI O N
Vose, R., Huang, B., Yin, X., Arndt, D., Easterling, D.R., Additional supporting information may be found in the
Lawrimore, J.H., Menne, M.J., Sanchez-Lugo, A. and Zhang, H. online version of the article at the publisher's website.
M. (2021) Implementing full spatial coverage in NOAA's global
temperature analysis. Geophysical Research Letters, 48,
e2020GL090873. https://doi.org/10.1029/2020GL090873. How to cite this article: Yang, Y., Li, Q., Song,
Vose, R.S., Applequist, S., Menne, M.J., Williams, C.N. and Z., Sun, W., & Dong, W. (2022). A comparison of
Thorne, P. (2012) An intercomparison of temperature trends in global surface temperature variability, extremes
the U.S. Historical Climatology Network and recent atmo- and warming trend using reanalysis datasets and
spheric reanalyses. Geophysical Research Letters, 39, L10703.
CMST-Interim. International Journal of
https://doi.org/10.1029/2012GL051387.
Vose, R.S., Wuertz, D., Peterson, T.C. and Jones, P.D. (2005) An
Climatology, 1–20. https://doi.org/10.1002/joc.7551
intercomparison of trends in surface air temperature analyses at

View publication stats

You might also like