East Asian Summer Monsoon Representation in Re-Analysis Datasets

Eight current re-analyses—NCEP/NCAR Re-analysis (NCEPI), NCEP/DOE Re-analysis (NCEPII), NCEP Climate Forecast System Re-analysis (CFSR), ECMWF Interim Re-analysis (ERA-Interim), Japanese 55-year Re-analysis (JRA-55), NASA Modern-Era Retrospective Analysis for Research and Applications (MERRA), NOAA Twentieth Century Re-analysis (20CR), and ECMWF’s first atmospheric re-analysis of the 20th century (ERA-20C)—are assessed to clarify their quality in capturing the East Asian summer monsoon (EASM) rainfall structure and its associated general circulation. They are found to present similar rainfall structures in East Asia, whereas they illustrate some differences in rainfall intensity, especially at lower latitudes. The third generation of re-analysis shows a better estimate of rainfall structure than that in the first and extended generation of re-analysis. Given the fact that the rainfall is ingested by the data assimilation system, the re-analysis cannot improve its production of rainfall quality. The mean sea level pressure is generated by re-analysis, showing a significant uncertainty over the Tibetan Plateau and its surrounding area. In that region, the JRA-55 and MERRA have a negative bias (BIAS), while the other six re-analyses present a positive BIAS to the observed mean sea level pressure. The 20CR and the ERA-20C are ancillary datasets to analyse the EASM due to the fact that they only apply limit observations into the data assimilation system. These two re-analyses demonstrate a prominent difference from the observed winds in the upper-air. Although the upper level winds exhibit difference, the EASM index is consistent in the eight re-analyses, which are based upon the zonal wind over 850 hPa.


Introduction
The East Asian summer monsoon (EASM) is an important part of the global climate system and plays a vital role in the Asian climate.It influences the livelihood and the socioeconomic status of over a billion residents who live in the EASM dominated region.The EASM is composed of Australian High, cross-equatorial jet, southwest monsoon, monsoon trough (Inter Tropical Convergence Zone), trade wind, west pacific subtropical high, Meiyu/Baiu/Changma and mid-latitudinal turbulence [1].
In the past decades, extensive research has been conducted to increase our knowledge of monsoon variability and predictability, and to improve projections of the impact of human activities on monsoonal systems over East Asia [2][3][4].This extensive work strongly depends on observations.A global observation network was built up in 20th century [5].The observation network collects data from in situ stations, ships, buoys, satellite and aircraft etc.It provides the best estimate of the state of the atmosphere, land and ocean.However, the data is inhomogeneous, due to the fact that there are gaps in spatial and temporal coverage.Global atmospheric assimilated datasets, called re-analysis, have they are the most reliable variable in the first-, second-and third-generation re-analyses.These data are categorised as "A" variables.However, the extended-generation re-analysis (20CR and ERA-20C) only takes surface data (e.g., surface pressures and surface winds) as part of data assimilation.Therefore, pressure level variables (e.g., wind fields and geopotential height) in extended generation reanalysis is classified as "C" class.These variables should be compared to observational data when they are used for scientific research, especially for studying the EASM.
In this paper, we evaluate eight widely used re-analysis datasets representing EASM rainfall and inter-compare them to capture EASM general circulation.The goal is to clarify their quality and efficiency in describing the EASM.The observational and re-analysis datasets, as well as the analysis method, is described in Section 2. Section 3 presents the inter-comparison results, including rainfall structure, general circulation, and monsoon strength in the re-analyses.The conclusion and discussion are summarised in Section 4.

Data and Method
In this study, eight re-analyses are evaluated and inter-compared.A brief overview of the eight re-analyses is presented in Table 1.Because of the spatial (EASM region) and temporal  coverage, monthly data of the eight re-analysis datasets are selected for further analysis.They are the first generation of re-analysis (NCEPI and NCEPII), the third generation of re-analysis (CFSR, ERA-Interim, MERRA, and JRA-55), and the extended-generation re-analysis (20CR and ERA-20C).A brief summary of each re-analysis focusing on its main strengths and limitations is presented in Table 2.A detailed description of each re-analysis dataset can be found at the Climate Data Guide website (https://climatedataguide.ucar.edu/climate-data).
The Global Precipitation Climatology Project (GPCP; [30]) is employed as the precipitation observational data.We choose the extended Hadley Centre's monthly historical mean sea level pressure dataset (HadSLP2r; [31]) as the reference data for mean sea level pressure.The Integrated Global Radiosonde Archive (IGRA; [32]) is applied to evaluate the spread of multi-pressure level wind fields in re-analysis datasets.Over 1500 globally distributed stations contribute to the IGRA, which consists of radiosonde and pilot balloon observations.More than 150 stations are located in our study area, with the longest record since 1958.The station data is interpolated to grid dataset by iterative improvement objective analysis.To eliminate the uncertainty associated with different data resolutions, the validation data and the re-analysis datasets are remapped onto a common gird of 2.5 • × 2.5 • by bi-linear interpolation.Four variables are used in the further analysis, i.e., precipitation (pr), mean sea level pressure (SLP), zonal wind (u850) and meridional wind (v850) over 850 hPa.
The quality of re-analysis data is measured by pattern correlation coefficient (PCC), anomaly correlation coefficient (ACC), root-mean-square error (RMSE) and bias (BIAS).The PCC statistical method is the un-centred statistical measure, which is without removal of the regional mean.

Mean Annual Cycle of the EASM Rainfall
The EASM features a prominent seasonal transition of prevailing winds and an abrupt change from dry to wet climate.In climatology, the onset of EASM has four stages: (1) it starts in the central Indo-China peninsula in late April or early May; (2) it propagates eastward to the South China Sea in mid-May; (3) in early June, the EASM arrives at the Yangtze River Basin (Meiyu) and the south of Japan (Baiu); (4) the EASM reaches northward to Northern China, the Korean Peninsula (namely, the Changma rainy season) and even central Japan in early July [33,34].
Figure 1 shows the seasonal transition of the EASM rainfall through a time-latitude Hovmoller diagram in GPCP and re-analysis datasets.This corresponds to a northward-southward movement of the Inter Tropical Convergence Zone, which is characterised by successive active and break phases of the convective activity.This meridional cross-section analysis provides a good framework to assess multi-dataset skill in representing mean annual cycle and intra-seasonal variations of the EASM and associated monsoon onset and withdrawal process [35].

Mean Annual Cycle of the EASM Rainfall
The EASM features a prominent seasonal transition of prevailing winds and an abrupt change from dry to wet climate.In climatology, the onset of EASM has four stages: (1) it starts in the central Indo-China peninsula in late April or early May; (2) it propagates eastward to the South China Sea in mid-May; (3) in early June, the EASM arrives at the Yangtze River Basin (Meiyu) and the south of Japan (Baiu); (4) the EASM reaches northward to Northern China, the Korean Peninsula (namely, the Changma rainy season) and even central Japan in early July [33,34].
Figure 1 shows the seasonal transition of the EASM rainfall through a time-latitude Hovmoller diagram in GPCP and re-analysis datasets.This corresponds to a northward-southward movement of the Inter Tropical Convergence Zone, which is characterised by successive active and break phases of the convective activity.This meridional cross-section analysis provides a good framework to assess multi-dataset skill in representing mean annual cycle and intra-seasonal variations of the EASM and associated monsoon onset and withdrawal process [35].The first-generation re-analyses exhibit the worst performance in capturing the latitude-time structure of EASM, with a lower PCC of 0.71 and 0.73 for NCEPI and NCEPII, respectively.There is no doubt that the third-generation reanalyses have better skill at representing the seasonal progress of EASM.Its PCC range is from 0.81 to 0.84, while the RMSE range is from 0.93 to 1.44.We can see that 20CR and ERA-20C exhibit a reasonable performance in describing the EASM rainfall structure.It is worth mentioning that JRA-55 is the best re-analysis dataset in capturing the pattern distribution of monsoon rainfall structure, but it demonstrates the worst performance in representing the amplitude of rainfall, due to its having the largest RMSE (1.44).

EASM Inter-Annual Variability
Taylor diagrams are a valuable tool for evaluating model data performance regarding the matching of temporal variability using temporal correlation and standard deviation [36].Here, we calculate the temporal correlations from summer (June-July-August) mean precipitation and the associated three meteorological fields averaged over the EASM region (0-50 • N, 100-140 • E) for all re-analyses relative to the reference data (Figure 2).For precipitation, the re-analyses show a large spread, with normalised standard deviations (NSD) from 0.87 to 1.52 and a correlation range from 0.37 to 0.66 for the GPCP.The JRA-55 has the highest correlation (0.66), while the ERA-Interim exhibits the lowest correlation (0.37) with the observed precipitation.The JRA-55 and the NCEPI demonstrate a smaller inter-annual variation than the reference data, with a NSD < 1.The precipitation year-to-year variation is larger in the other six re-analyses than in the GPCP, especially the CFSR and the MERRA, with NSDs of 1.52 and 1.50, respectively.
The first-generation re-analyses exhibit the worst performance in capturing the latitude-time structure of EASM, with a lower PCC of 0.71 and 0.73 for NCEPI and NCEPII, respectively.There is no doubt that the third-generation reanalyses have better skill at representing the seasonal progress of EASM.Its PCC range is from 0.81 to 0.84, while the RMSE range is from 0.93 to 1.44.We can see that 20CR and ERA-20C exhibit a reasonable performance in describing the EASM rainfall structure.It is worth mentioning that JRA-55 is the best re-analysis dataset in capturing the pattern distribution of monsoon rainfall structure, but it demonstrates the worst performance in representing the amplitude of rainfall, due to its having the largest RMSE (1.44).

EASM Inter-Annual Variability
Taylor diagrams are a valuable tool for evaluating model data performance regarding the matching of temporal variability using temporal correlation and standard deviation [36].Here, we calculate the temporal correlations from summer (June-July-August) mean precipitation and the associated three meteorological fields averaged over the EASM region (0-50° N, 100-140° E) for all re-analyses relative to the reference data (Figure 2).For precipitation, the re-analyses show a large spread, with normalised standard deviations (NSD) from 0.87 to 1.52 and a correlation range from 0.37 to 0.66 for the GPCP.The JRA-55 has the highest correlation (0.66), while the ERA-Interim exhibits the lowest correlation (0.37) with the observed precipitation.The JRA-55 and the NCEPI demonstrate a smaller inter-annual variation than the reference data, with a NSD < 1.The precipitation year-to-year variation is larger in the other six re-analyses than in the GPCP, especially the CFSR and the MERRA, with NSDs of 1.52 and 1.50, respectively.The re-analysis datasets illustrate high consistency for the SLP and the zonal wind at 850 hPa (i.e., u850).The range of SLP correlation is 0.71 to 0.81, and NSD is 0.95 to 1.17.We found that the re-analyses have the same performance in representing the u850 as the SLP, with a high correlation (0.88-0.93) and an approximate inter-annual variation (NSD: 1.05-1.29)for the IGRA.Obviously, the meridional wind at 850 hPa (v850) presents a lower year-to-year variation than the reference data's.We observed that the CFSR and the MERRA have a worse performance in capturing v850 variation than the other six re-analyses, with NSDs of 1.48 and 1.58, respectively.
A further evaluation focuses on the re-analysis datasets and their ensemble mean in representing the inter-annual variability of the three variables over land (Figure 2b) and ocean (Figure 2c).In general, the pr and the v850 exhibit better performance (with higher correlation coefficient) over land than over oceans, while the SLP and the u850 show the opposite performance.We find that the CFSR and the NCEPII have a lower correlation and a higher NSD than the other re-analysis datasets.The re-analysis mean captures a more realistic year-to-year variation of the four variables.It shows a higher correlation coefficient than the individual re-analysis dataset.

Spatial Difference in Re-Analysis Datasets
Figure 3 presents the ensemble standard deviation (ESD) of the eight re-analyses for the four meteorological variables.It shows the inter-reanalysis difference.The precipitation spread decreases with increasing latitude.A large ESD (~3 mm day −1 ) occurs in the low latitude region, especially in the western Indo-China peninsula, where the ESD is >5 mm day −1 .For the u850 and v850, the re-analyses show high consistency, with ESD ~1 mm s −1 in the entire EASM region.There is no prominent difference among the re-analyses in representing SLP over ocean.However, we found a large ESD of SLP over land, especially in the western EASM region (i.e., Tibetan Plateau; ESD > 3 hPa).The ESD of ua200 shows a "sandwich" pattern, with high-low-high distribution in our study region.
For a specific variable (e.g., pr and SLP) and region, the re-analysis members show significant disagreement among them.We employ the BIAS analysis to quantify their magnitude of deviation from observation (Figures 4 and 5).In monsoon season, the inter-tropical convergence zone reaches its northernmost location.The water moisture is transported by northward wind from ocean to land.Two rainfall belts are located in the East Asia; the south branch stretches from the Bay of Bengal, over the Indo-China peninsula and the Philippine Sea, and the north branch occurs over the east of China, the Korean peninsula and the south of Japan (Figure 4; cf.observation).
The eight re-analysis datasets capture the major features of the spatial distribution of rainfall in the monsoon season.However, these datasets tend to generate significantly wetter conditions in the southern EASM region (i.e., the South China Sea and the Philippines) and much drier conditions over the Korean peninsula and Japan.The third generation of re-analysis datasets presents a better performance (with small BIAS) in capturing the summer rainfall, especially in mainland China, the Korean peninsula and Japan than the first-and the extended-generation re-analysis datasets.It is worth mentioning that only the ERA-Interim and the ERA-20C produce more rainfall in the Indo-China peninsula, whereas the other six re-analysis datasets generate less rainfall.The large rainfall BIAS occurs along with the lower level of general circulation (i.e., wind fields at 850 hPa) BIAS.There is a significant wind BIAS over the South China Sea and the Philippine Sea.In northern China, the 20CR shows a larger northwestward wind than the observation, but the other seven re-analysis datasets illustrate a good agreement with the observation (Figure 4).In the summer season, the land (ocean) is a heat source (sink), with lower (higher) mean sea surface pressure (Figure 5; cf.observation).The Tibetan Plateau presents a prominent low pressure centre due to its huge topography.The first-generation re-analysis datasets produce higher SLP in the Tibetan Plateau, the western part of the Indo-China peninsula and Indonesia, and lower SLP in Japan and the Sea of Japan (Figure 5; cf.NCEPI and NCEPII).The CFSR and the ERA-Interim have the same performance in presenting SLP in the Tibetan Plateau and Japan, but illustrate no significant BIAS in the Indo-China peninsula and Indonesia.However, these two datasets generate less SLP in the East China Sea.Both the JRA-55 and the MERRA show the prominent negative BIAS of SLP in the western (e.g., Tibetan Plateau, southwest of China etc.) and the northern (e.g., Mongolia, northern of China etc.) EASM region.The 20CR demonstrates positive BIAS at the centre of SLP in the south of the Tibetan Plateau, the northeast of China and the northern Sea of Japan, but produces less SLP in the western South China Sea and the Philippine Sea.Compared to the HadSLP2r, the ERA-20C generates more SLP in the Tibetan Plateau and the western Indo-China peninsula, but less SLP in the northern EASM region.At low latitudes, the re-analyses show good agreement with the observation in presenting the year-to-year variation of SLP (with RMSE ~0 hPa; Figure 7).There is no doubt that all the re-analyses have a large RMSE of SLP over the Tibetan Plateau.In northern China and Mongolia, the re-analyses exhibit a larger RMSE score (~1 hPa) than at low latitudes.The ERA-20C calculates the RMSE centre of SLP (RMSE > 2 hPa) from northern China to the northeast of China.For the u850 and v850, the re-analyses demonstrate distinct RMSE scores in the northwest of China, Mongolia, the South China Sea and the Philippine Sea.In these areas, u850 and v850 exhibit a significant difference between the re-analysis datasets, and also from the observation.

Monsoon Strength
The EASM is characterised by strong year-to-year variability.To measure the strength and study the long-term change of the EASM, more than 25 monsoon indices have been produced in the last few decades.Wang et al. [37] classified these monsoon indices into five categories and analysed their performance in capturing the main features of the EASM.They found that the Wang and Fan index [38] outperforms the other 24 monsoon indices in capturing the three-dimensional circulation and total variance of the precipitation over East Asia.Following Wang et al. [37], we select the Wang and Fan index in our further study.The definition of the Wang and Fan index is the standardised average zonal wind at 850 hPa at (5-15 • N, 90-130 • E) minus that at (22.5-32.5 • N, 110-140 • E).
Figure 8a illustrates the observed (IGRA) and the multi-reanalysis ensemble mean produced EASM index (EASMI).The re-analysis ensemble mean shows good with the observation in representing the EASMI.For the individual re-analysis, it can capture the phase of EASMI, only showing a slight difference in capturing the EASMI magnitude (Figure 8b).The range of correlation coefficient between the EASMI in observation and in the re-analyses is from 0.97 to 0.99 during the satellite era .A lower correlation coefficient (0.72-0.84) can be found in the pre-satellite era .The 20CR indicates an extremely strong monsoon year (EASMI > 1) in 1997 and 2007, but the observation and the other re-analysis datasets show a normal monsoon year (EASMI > −1 and EASMI < 1).The 20CR and ERA-20C show good agreement from the mid-1930s to the end 1950s (Figure 8c).However, the two datasets present different monsoon phases from 1900 to the early 1930s.They produce the same monsoon phases and similar monsoon magnitudes in two strong monsoon years (1904 and 1911).

Monsoon Strength
The EASM is characterised by strong year-to-year variability.To measure the strength and study the long-term change of the EASM, more than 25 monsoon indices have been produced in the last few decades.Wang et al. [37] classified these monsoon indices into five categories and analysed their performance in capturing the main features of the EASM.They found that the Wang and Fan index [38] outperforms the other 24 monsoon indices in capturing the three-dimensional circulation and total variance of the precipitation over East Asia.Following Wang et al. [37], we select the Wang and Fan index in our further study.The definition of the Wang and Fan index is the standardised average zonal wind at 850 hPa at (5-15° N, 90-130° E) minus that at (22.5-32.5°N, 110-140° E).
Figure 8a illustrates the observed (IGRA) and the multi-reanalysis ensemble mean produced EASM index (EASMI).The re-analysis ensemble mean shows good agreement with the observation in representing the EASMI.For the individual re-analysis, it can capture the phase of EASMI, only showing a slight difference in capturing the EASMI magnitude (Figure 8b).The range of correlation coefficient between the EASMI in observation and in the re-analyses is from 0.97 to 0.99 during the satellite era .A lower correlation coefficient (0.72-0.84) can be found in the pre-satellite era .The 20CR indicates an extremely strong monsoon year (EASMI > 1) in 1997 and 2007, but the observation and the other re-analysis datasets show a normal monsoon year (EASMI > −1 and EASMI < 1).The 20CR and ERA-20C show good agreement from the mid-1930s to the end 1950s (Figure 8c).However, the two datasets present different monsoon phases from 1900 to the early 1930s.They produce the same monsoon phases and similar monsoon magnitudes in two strong monsoon years (1904 and 1911).

Figure 1 .
Figure 1.Precipitation latitude-time cross section of observation ((a) Global Precipitation Climatology Project; GPCP) and different re-analysis datasets (b-i) in the East Asian summer monsoon region (0-50° N, 100-140° E) from 1979 to 2010.The number in the upper-left corner of each panel indicates the pattern correlation coefficient (left) and the root-mean-square error (right) skill of the observed precipitation.

Figure 1 .
Figure 1.Precipitation latitude-time cross section of observation ((a) Global Precipitation Climatology Project; GPCP) and different re-analysis datasets (b-i) in the East Asian summer monsoon region (0-50 • N, 100-140 • E) from 1979 to 2010.The number in the upper-left corner of each panel indicates the pattern correlation coefficient (left) and the root-mean-square error (right) skill of the observed precipitation.

Figure 2 .
Figure 2. Temporal statistics describing inter-annual variability of the re-analysis datasets and the multi-datasets ensemble mean in terms of June-July-August (JJA) mean precipitation (black), zonal winds (blue) and meridional winds (green) at 850 hPa, and mean sea level pressure (red) over the East Asian summer monsoon (EASM) region (0-50° N, 100-140° E) (a), the EASM land only (b), and the EASM ocean only (c) from 1979 to 2010.The Global Precipitation Climatology Project (GPCP) was employed as the reference data for precipitation, while the mean sea level pressure was compared by

Figure 2 .
Figure 2. Temporal statistics describing inter-annual variability of the re-analysis datasets and the multi-datasets ensemble mean in terms of June-July-August (JJA) mean precipitation (black), zonal winds (blue) and meridional winds (green) at 850 hPa, and mean sea level pressure (red) over the East Asian summer monsoon (EASM) region (0-50 • N, 100-140 • E) (a), the EASM land only (b), and the EASM ocean only (c) from 1979 to 2010.The Global Precipitation Climatology Project (GPCP) was employed as the reference data for precipitation, while the mean sea level pressure was compared by extended the Hadley Centre's monthly historical mean sea level pressure dataset (HadSLPr2), and the wind fields were evaluated by the Integrated Global Radiosonde Archive (IGRA).

Figure 5 .
Figure 5. Summer (JJA) mean sea level pressure (observation shaded) of the extended Hadley Centre monthly historical mean sea level pressure dataset (HadSLPr2) and wind fields at 850 hPa (Observation vector) of the Integrated Global Radiosonde Archive (IGRA) and the mean sea level pressure anomalies 're-analysis minus HadSLPr2; shaded', and the wind anomalies 're-analysis minus IGRA; vector' in 1979-2010.The presented anomalies of precipitation pass Student's t-test at 0.05 level.The green box represents the East Asian summer monsoon region (0-50° N, 100-140° E).

Figure 5 .
Figure 5. Summer (JJA) mean sea level pressure (observation shaded) of the extended Hadley Centre monthly historical mean sea level pressure dataset (HadSLPr2) and wind fields at 850 hPa (Observation vector) of the Integrated Global Radiosonde Archive (IGRA) and the mean sea level pressure anomalies 're-analysis minus HadSLPr2; shaded', and the wind anomalies 're-analysis minus IGRA; vector' in 1979-2010.The presented anomalies of precipitation pass Student's t-test at 0.05 level.The green box represents the East Asian summer monsoon region (0-50 • N, 100-140 • E).

Figure 6 .
Figure 6.Root-mean-square error skill score for summer (June-July-August) rainfall during 1979-2010.The Global Precipitation Climatology Project (GPCP) is employed as the reference data for rainfall.The green box is the East Asian summer monsoon region (0-50° N, 100-140° E).

Figure 7 .
Figure7.As in Figure6, but for mean sea level pressure (SLP; shaded) and lower level winds (at 850 hPa; vector).The reference data for mean sea level pressure is the extended Hadley Centre monthly historical mean sea level pressure dataset (HadSLPr2).The Integrated Global Radiosonde Archive (IGRA) is selected to evaluate the winds.The green box is the East Asian summer monsoon region (0-50° N, 100-140° E).

Figure 6 .
Figure 6.Root-mean-square error skill score for summer (June-July-August) rainfall during 1979-2010.The Global Precipitation Climatology Project (GPCP) is employed as the reference data for rainfall.The green box is the East Asian summer monsoon region (0-50 • N, 100-140 • E).

18 Figure 6 .
Figure 6.Root-mean-square error skill score for summer (June-July-August) rainfall during 1979-2010.The Global Precipitation Climatology Project (GPCP) is employed as the reference data for rainfall.The green box is the East Asian summer monsoon region (0-50° N, 100-140° E).

Figure 7 .
Figure 7.As in Figure6, but for mean sea level pressure (SLP; shaded) and lower level winds (at 850 hPa; vector).The reference data for mean sea level pressure is the extended Hadley Centre monthly historical mean sea level pressure dataset (HadSLPr2).The Integrated Global Radiosonde Archive (IGRA) is selected to evaluate the winds.The green box is the East Asian summer monsoon region (0-50° N, 100-140° E).

Figure 7 .
Figure 7.As in Figure6, but for mean sea level pressure (SLP; shaded) and lower level winds (at 850 hPa; vector).The reference data for mean sea level pressure is the extended Hadley Centre monthly historical mean sea level pressure dataset (HadSLPr2).The Integrated Global Radiosonde Archive (IGRA) is selected to evaluate the winds.The green box is the East Asian summer monsoon region (0-50 • N, 100-140 • E).

Figure 8 .
Figure 8. East Asian summer monsoon index of observation (IGRA) and multi-reanalysis ensemble mean (a), individual re-analysis dataset (b), and the 20CR and ERA-20C in 1900-1957 (c).The number following the re-analysis presents the correlation coefficient between the East Asian summer

Figure 8 .
Figure 8. East Asian summer monsoon index of observation (IGRA) and multi-reanalysis ensemble mean (a), individual re-analysis dataset (b), and the 20CR and ERA-20C in 1900-1957 (c).The number following the re-analysis presents the correlation coefficient between the East Asian summer monsoon index produced by re-analysis and the observed one during 1958-2010 (left), during 1958-1978 (middle), and during 1979-2010 (right).

Table 1 .
Basic information of re-analyses investigated in this study.

Table 2 .
Brief summary of the eight re-analysis datasets in this study with their strength and limitation. https://www.esrl.noaa.gov/psd/data/gridded/data.ncep.reanalysis.html