3.1. The Performance of ERA5 over Antarctica
The correlation coefficients between ERA5 and observations are significant for all months, and they are higher than 0.82, and those of ERA-Interim are more than 0.84, indicating a stronger linear relationship of monthly temperature. Figure 2
displays the MB between the monthly mean air temperature measured at the 41 weather stations and the ECMWF reanalysis output data, and we also plot the RSD in this figure. For ERA5, warm bias prevails between May and September. ERA5 has a monthly bias between −0.44 and 1.19 °C, and the extreme values appear in August and December, respectively. ERA-Interim has a cold bias in all months, and this characteristic is different from that of ERA5. In particular, a large contrast in ERA5 and ERA-Interim bias occurs in winter (June–August (JJA)).
As shown in Table 2
, for the whole Antarctic ice sheet, there are significant correlations between ERA5 and observations for all annual and seasonal temperature. Compared to ERA-Interim, ERA5 exhibits lower performance only in summer (December–February (DJF)). In this season, the significant correlation coefficient of ERA5 is 0.84, and the corresponding value of ERA-Interim is 0.86. Table 3
shows the bias between ERA reanalyses and observations. In general, ERA5 performs worst in summer, with the largest bias of 1.06 °C and the lowest correlation coefficient. Over the whole of Antarctica, ERA5 has good performance in autumn (March–May (MAM)) and spring (September–November (SON)), with low bias of 0.16 and 0.38 °C, and the highest significant correlation coefficients are 0.90 and 0.93, respectively. Different from the lowest bias for ERA5, which occurs in MAM, ERA-Interim has the lowest bias of 0.52 °C in JJA. The greatest bias between ERA reanalyses and observations is shown in DJF, with the cold bias of 1.06 and 0.76 °C for ERA5 and ERA-Interim, respectively. In Antarctica, a cold bias prevails for annual and seasonal mean temperatures in ERA-Interim, whereas ERA5 shows a warm bias in JJA, when the largest difference between ERA reanalyses is captured. Figure 3
shows the time series of the annual and seasonal mean temperature anomalies with respect to the 1979–2018 mean from 41 observation locations and the corresponding value of ERA reanalyses. Generally, there are no clear anomaly differences between ERA5 and ERA-Interim in Antarctica. ERA reanalyses can reflect common interannual variability, and they can capture the abrupt changes occurring in Antarctica.
The correlations, mean bias, and ratio of SDs between observations and the corresponding data from ERA reanalyses at Antarctic coastal and inland stations are shown in Figure 4
, and the correlations pass the test of significance (p
< 0.05). ERA5 presents the highest correlations in spring for both coastal and inland stations, with significant correlation coefficients of 0.93 and 0.94, respectively. Compare to ERA-Interim, ERA5 has higher correlations in MAM and JJA at coastal stations, and lower correlations at inland stations in these seasons. ERA-Interim always shows cold bias at coastal stations and warm bias at inland stations; somewhat differently, ERA5 has cold bias in DJF at inland stations, indicating that the cold biases seen previously for the whole continent are caused by the coastal stations. The mean bias at coastal stations is always smaller than that of inland stations in all annual and seasonal mean temperature, and both of the ERA reanalyses show this feature. For coastal stations, biases in ERA5 are smaller than in ERA-Interim for all seasons with the exception of austral summer. In MAM, ERA-Interim has the greatest bias of 0.99 and −3.25 °C for coastal and inland stations, respectively. For ERA5, only inland stations have an SD value higher than the observations in MAM, and the RSD of ERA-Interim is greater than 1 in autumn, winter, and annually at coastal stations.
3.2. The Performance of ERA5 over the Three Subregions in Antarctica
The trend of Antarctic temperature is unclear now, and the tendency and the number of weather stations may differ among regions, therefore, we divided Antarctica into three subregions, including East Antarctica, West Antarctica, and the Antarctic Peninsula (Figure 1
) to further explore the performance of ERA5. The correlation coefficients, bias, and ratio of the SDs of ERA5 and the differences in these variables between ERA5 and ERA-Interim for monthly temperature measurements at 41 meteorological stations are shown in Figure 5
. It is worth noting that the correlation coefficients here are all significant at the 95% confidence interval. For monthly temperature, the correlation coefficients between ERA5 and observations were generally high, with correlations higher than 0.95 at every one of the 41 stations selected, and the high correlations related to the temperature data have been assimilated into the reanalyses. The difference between the correlation coefficients of ERA5 and ERA-Interim is fairly small, and the division between them is less than 0.01. As shown in Figure 5
b, ERA-Interim shows a higher linear relationship at stations located in East Antarctica and most stations in the Antarctic Peninsula, whereas ERA5 exhibits a stronger linear relationship with the observations on the Ross Ice Shelf. For ERA5, warm bias prevails for stations located at the interior of the East Antarctica, and there is no distinct pattern of bias over the coastal East Antarctica, West Antarctica, and Antarctic Peninsula stations. Compared with ERA-Interim, lower bias in ERA5 can be found at 26 stations, and the biggest difference between ERA5 and ERA-Interim occurs at McMurdo. For ERA5, eight stations have a higher SD value than the observations (Belgrano II, Cape Ross, Marble Point, Marilyn, Mario Zucchelli, O’Higgins, Rothera, and San Martin), and only two stations have ERA5 SD values that are more than 20% lower (Dome C II and Marambio). ERA-Interim RSD values are higher than that those of ERA5 at 32 stations, and the SD value for ERA-Interim is higher than observations at 16 stations.
exhibits the correlations of ERA5 for annual and seasonal temperature and the difference between these correlations and those for ERA-Interim. Generally, the correlation coefficients between ERA5 and observations are fairly high for annual and seasonal temperature, with significant correlation coefficients (p
< 0.05) higher than 0.80 at most stations. Compared with ERA-Interim, ERA5 exhibits winter temperature better, with relatively high correlation coefficients at 21 stations. ERA5 always has a stronger linear relationship with observations at Esperanza, Schwerdtfeger, and Scott Base, and the biggest difference between ERA5 and ERA-Interim occurs at Scott Base for annual temperature, where the correlation coefficient of ERA5 is 0.23 higher than that of ERA-Interim. Figure 7
shows the same content as Figure 6
but for bias. In general, the bias values (<2.00 °C) are relatively low at stations located at the coastal area of East Antarctica. For annual and seasonal temperature, ERA5 always has a warm bias at 8 stations (Amundsen–Scott, Butler Island, Halley, Marambio, Mirny, Neumayer, Schwerdtfeger, and Scott Base), and a cold bias is shown at 14 stations (Belgrano II, Byrd, Esperanza, King Sejong, Manuela, Marble Point, Mario Zucchelli, Mawson, McMurdo, Novolazarevskaya, O’Higgins, Palmer, Rothera, San Martin, and Zhongshan), and ERA5 shows warm biases at the inland stations (Vostok, Amundsen–Scott, and Dome C II) for the non-summer mean temperature. The smallest bias of ERA5 occurs at Faraday in MAM, and the data from ERA5 is lower than observations only by 0.01 °C. The greatest cold and warm bias can be found at Mario Zucchelli and Halley in the austral winter, with values of −6.94 and 11.61 °C, respectively. Compared with ERA-Interim, ERA5 exhibits lower bias at 10 stations (Dumont d’Urville, Faraday, Gill, Manuela, Marilyn, Mawson, San Martin, Scott Base, and Zhongshan) for all annual and seasonal temperatures, and higher bias always be found at Esperanza, Halley, Marble Point, Mirny, O’Higgins, Rothera, and Schwerdtfeger. At Scott Base station, warm bias is shown in all annual and seasonal temperatures in ERA5, while ERA-Interim shows cold bias, and a cold bias of 3.96 °C occurs for winter temperature in ERA-Interim, whereas ERA5 has a warm bias of 2.40 °C, which is the largest contrast between ERA5 and ERA-Interim. In particular, a large contrast in bias between autumn and winter is exhibited at Mawson and McMurdo station, which represent as ERA-Interim shows large bias larger than 5 °C, but the bias in ERA5 is lower than 1 °C. ERA5 bias is bigger than ERA-Interim in JJA, and smaller bias are shown in SON and annual temperature at inland stations. Compared with ERA-Interim, the smaller bias in ERA5 occurs at five stations (Zhongshan, Casey, Dumont d’Urville, Mawson, Neumayer) at the coastal areas of East Antarctica in all annual and seasonal temperature, and higher bias always at Mirny station. At Byrd station located at West Antarctica, ERA5 represents bigger bias than that of ERA-Interim except for autumn. McMurdo Station and Scott Base, both on Ross Island and only 3 km apart. The temperature in ERA5 at McMurdo Station are always colder than observations, whereas those at Scott Base are warmer, indicating that the temperature of ERA5 reanalysis is influenced by small-scale topographical differences.
In Figure 8
, we plot six selected stations (Amundsen–Scott, Byrd, Marambio, Novolazarevskaya, Scott Base, and Vostok) located in central Antarctica, West Antarctica, the Antarctic Peninsula, the East Antarctica coast, the Ross Ice Shelf, and the interior of East Antarctica, respectively. For the six stations, ERA5 has slightly lower correlation (R < 0.8) at Amundsen–Scott and Novolazarevskaya, as shown in Figure 6
i, and the bias of ERA5 is larger than that of ERA-Interim only at Byrd. At Novolazarevskaya station, the data from ERA5 has a poor linear relationship with observations before 2000, but subsequently fit the observation data well. The temperature from ERA5 shows obvious disparity at Scott Base and Vostok relative to ERA-Interim. For Marambio, ERA5 shows the same warming as station data during the period.
ERA5 has average annual temperature biases of 0.51, −0.66, and 0.58 °C in East Antarctica, West Antarctica, and the Antarctic Peninsula (Table 3
), respectively. The corresponding biases from ERA-Interim are 0.97, 0.03, and 0.23 °C, which indicates a more accurate performance of ERA5 for annual temperature over East Antarctica. Table 2
summarizes the correlation between ERA reanalyses and observations over the Antarctica and its three subregions for annual and seasonal temperature means. Generally, we can conclude that ERA5 performs better over East Antarctica with exception of DJF, and it exhibits a high linear relationship with observations and small bias in other cases. Compared with ERA-Interim, ERA5 has a lower correlation and higher bias in DJF in East Antarctica, and the highest bias in ERA reanalyses occurs in this season, with the cold bias of 1.53 and 1.25 °C for ERA5 and ERA-Interim, respectively. Over West Antarctica, warm bias and higher correlation coefficients prevail for annual and seasonal temperature means in ERA5. Especially, the highest correlation coefficient of ERA5 is 0.97, occurring in West Antarctica in SON, and the corresponding value of ERA-Interim is 0.96. The biggest difference between temperature from ERA5 and observations in West Antarctica is found in JJA, with a warm bias of 1.25 °C. In the Antarctic Peninsula, ERA5 always shows cold bias, and ERA-Interim exhibits a warm bias in JJA. In general, a higher bias and lower correlation coefficients of ERA5 relative to ERA-Interim are observed in this area. For ERA5, lower performance in the Antarctic Peninsula is shown in DJF, with high bias values and low liner relationship with stations records. We conclude that ERA5 performs well in representing East Antarctic and West Antarctic temperature, especially in SON, with the bias lower than 0.80 °C and correlation coefficients higher than 0.90. The anomalies from ERA5 and ERA-Interim coincide with common variability from observations over the three subregions, especially in the Antarctic Peninsula (Figure 3
). The mean autumn temperature over East Antarctica exhibits a shift change in circa 2002, and the change is also found at Antarctica and West Antarctica in MAM. However, this phenomenon does not occur in the Antarctic Peninsula.
In Figure 9
, we compare the spatial trend of austral seasons for the whole of Antarctica from ERA5 and ERA-Interim during the period 1979–2018. In MAM, the trend of ERA5 is in broad agreement with that from ERA-Interim; both of them show significant warming trends in the western Antarctic Peninsula, while the trend of ERA-Interim in this region shows greater warming than that of ERA5. In JJA, there is an obvious warming trend on the Ross Ice Shelf in ERA-Interim, while the trend value of ERA5 is lower than that of ERA-Interim. Significant warming trends are prevalent in East Antarctica in SON, and the trend is stronger and broader than shown by ERA-Interim. Over all seasons, the greatest difference between the trend of ERA5 and ERA-Interim occurs in DJF. In this season, ERA5 reveals a slight warming trend over the area in the interior of East Antarctica, whereas ERA-Interim shows a significant cooling trend in that region. The annual and seasonal mean temperature trends are summarized for the Antarctic subregions in Table 4
. Annual and seasonal temperature trends in ERA5 and ERA-Interim are statistically significant in East Antarctica, and the cooling trends in observations in SON do not pass the significance test. The trends in ERA reanalyses and observations are all negative in East Antarctica in all annual and seasons, and the fastest cooling trend appears in MAM, and the cooling rate of this season is more than 1 °C per decade. In West Antarctica, the ERA5 trends are similar to observation trends, whereas there is a difference between ERA5 trends and ERA-Interim in SON, as reflected in a warming trend in ERA-Interim while a cooling trend is observed in ERA5. ERA5 exhibits a significant cooling trend in annual data, MAM, and JJA, and the trends from ERA-Interim always fail to pass the significance test. It is also worth mentioning that the ERA5 shows a faster cooling rate than ERA-Interim and observations in West Antarctica. Over the Antarctic Peninsula, trends of annual and seasonal temperature means in ERA reanalyses and observations are not significant. ERA5 presents a warming trend with the exception of DJF, as is the case for ERA-Interim and station records. Compared with ERA-Interim, the difference between ERA5 trends and observations in the Antarctic Peninsula is relatively small in JJA and DJF.
illustrates the regression trends of seasonal mean temperature from six select stations from the ERA reanalyses and Antarctic observations as illustrated in Figure 8
. At the Byrd, Marambio, Novolazarevskaya, and Scott Base stations, ERA5 exhibits the same trend as ERA-Interim in seasonal temperature. At Amundsen–Scott and Vostok, a warming trend is captured by ERA5 and observations in SON and DJF, whereas a cooling trend is found in ERA5 in MAM and JJA, in contrast to the warming trend in observations. There is no significant trend at Byrd, and ERA5 displays a different sign compared with observations in JJA and DJF. At Marambio, the same sign is captured by ERA5 and observation with exception of DJF. At Novolazarevskaya and Scott Base, the warming trend in ERA5 is prevalent for all seasons, while observation shows a cooling trend in MAM and DJF. At Vostok, ERA5 agrees well with the observed warming trend in SON, and divergence is shown between ERA-Interim and station records, except for austral spring.