Microscopía Electrónica de Transmisión (TEM)

The representation of 2-m temperature is evaluated by season due to the large differences between summer and winter insolation in Antarctica, which result in significant seasonal temperature variations. During winter the absence of insolation allows radiative cooling of the surface to dominate, generally resulting in the formation of a strongly stable, cold boundary layer (King, 1990). In summer the boundary layer is warmer and tends to be weakly stably-stratified or even slightly unstable (Mastrantonio et al.,1999).

Figure 3.2: The magnitude of seasonal temperature biases for each of the reanalysis products at the Bear Peninsula AWS.

As an example the seasonal temperature biases from BP AWS are shown in Fig.3.2.

At BP (and across the other AMRC sites) ERA-I records its smallest bias in the austral summer and its largest bias in the austral winter (Fig.3.2). Table3.3shows that across the AMRC sites both ERA-I and JRA-55 show a marked improvement in reproducing 2-m temperatures in summertime. Summer biases for ERA-I and JRA-55 respectively are −0.23^◦C and −1.91^◦C compared with −3.70^◦C and −3.89^◦C wintertime biases.

This suggests ERA-I and JRA-55 have more skill capturing the weakly stable or even unstable summer boundary layer, whereas CFSR and MERRA temperature biases show little seasonal variability (e.g. see Fig.3.2).

50

ANEVALUATION OF FOUR GLOBAL REANALYSIS PRODUCTS USING IN SITU OBSERVATIONS IN THEAMUNDSENSEAEMBAYMENT

Table 3.3: Statistical comparison of reanalysis near-surface temperatures to observed seasonal temperatures. A negative bias indicates that the reanalysis product is colder than the observations. SON, DJF, MAM and JJA indicate the season with DJF corresponding to austral summer. The unit for bias and RMSE is^◦C . * The NYU data set only covers a 13-month period.

Product Stats AMRC 3 site average NYU*

SON DJF MAM JJA SON DJF MAM JJA

Bias -2.27 -0.23 -2.66 -3.70 1.35 0.36 3.02 2.83 SD ratio 1.17 1.09 1.29 1.25 0.88 0.90 0.97 0.95 ERA-I R² 0.79 0.72 0.74 0.73 0.93 0.83 0.88 0.86 RMSE 4.37 2.22 5.16 6.34 2.83 2.06 4.44 4.32 Slope 1.04 0.93 1.12 1.07 0.85 0.82 0.91 0.89 Bias -2.82 -1.91 -3.40 -3.89 0.07 -0.18 1.74 1.55 SD ratio 1.01 1.14 1.03 0.92 0.81 0.84 0.82 0.74 JRA-55 R² 0.82 0.74 0.80 0.80 0.83 0.69 0.84 0.76 RMSE 4.17 2.88 4.68 5.29 3.74 2.70 4.24 4.77 Slope 0.91 0.98 0.92 0.82 0.74 0.70 0.75 0.64 Bias -2.88 -2.63 -2.96 -2.68 -0.48 -2.50 1.52 1.79 SD ratio 1.20 1.37 1.26 1.18 1.04 1.29 1.00 1.06

CFSR R² 0.75 0.62 0.75 0.73 0.78 0.59 0.84 0.79

RMSE 5.02 4.02 5.19 5.45 4.31 4.75 4.13 4.71 Slope 1.04 1.08 1.09 1.01 0.92 0.99 0.92 0.94 Bias -6.62 -6.86 -7.87 -6.89 -4.57 -6.00 -4.21 -3.11 SD ratio 1.06 1.28 1.12 0.99 0.95 1.04 1.01 0.98

MERRA R² 0.82 0.73 0.79 0.80 0.85 0.75 0.83 0.81

RMSE 7.31 7.29 8.57 7.72 5.69 6.52 5.78 4.94 Slope 0.96 1.09 1.00 0.89 0.88 0.90 0.92 0.88

The results in Table3.3show that the AMRC mean biases (across EK, TI and BP) are negative for all four reanalysis products. Reanalysis 2-m temperatures are lower than those observed. The weighted (by length of time series) mean temperature bias for all four sites shows that ERA-I has the smallest bias(−1.81 ^◦C ), compared with CFSR (−2.50 ^◦C ), JRA-55 (−2.62 ^◦C ) and MERRA (−6.80 ^◦C ). For MERRA the bias is significantly larger than the −1.6 ^◦C average bias found at coastal East Antarctic stations by Bracegirdle and Marshall (2012), suggesting MERRA may have a very strong regional bias in West Antarctica.

The all-season bias for ERA-Interim at each site is: −1.40^◦C at EK, −2.98^◦C at TI,

−2.27 ^◦C at BP and +1.89^◦C at NYU. These figures can be directly compared to the biases found at coastal AWS sites around Antarctica byJones and Lister(2015), where a group of AWSs on the Ross Sea coastline showed a similar pattern. At Marble Point,

3.4.AWS COMPARISON 51

Manuela and Scott AWSs, ERA-Interim showed temperature biases of −2.4^◦C , −1.4^◦C and −1.4^◦C respectively. These were some of the largest biases at coastal AWSs they examined and are of comparable magnitude to the ERA-I biases found at three of the sites used in the present study. Furthermore, a manned coastal station at McMurdo recorded a larger cold bias (−4.8^◦C ) than those seen at the three AMRC AWSs (Jones and Lister,2015). Similarly to the NYU AWS in our results there were also two AWSs on the Ross Sea coastline where ERA-Interim displayed a warm bias for the 2002-2013 period, at Arelis and Cape Ross. The results presented here combined with those from Jones and Lister(2015) suggest a systematic cold bias in ERA-I 2-m temperatures (of approximately −1.5^◦C ) extending around West Antarctica from the Ross Sea to the western side of the Antarctic Peninsula, there are of course a few AWS stations where ERA-I shows a warm bias within this sector.

The NYU AWS — located on PIG and at a relatively low altitude — is notably different with positive biases in the comparison with ERA-I, JRA-55 and CFSR reanalyses (Table 3.3). The linear regression slope values are all less than 1 due to a warm bias at low temperatures. The NYU AWS is the only site located on an ice shelf and so cold-air drainage during katabatic flows may be more prevalent — a phenomena that is difficult to accurately model (Renfrew,2004), perhaps leading to the warm bias. MERRA remains colder than NYU AWS observations, although the magnitude of the bias is smaller than that at the other sites.

The R²values show little variation between the four reanalysis products. R²values are typically between 0.7 and 0.85 (see Figs.3.3and3.4), comparable with correlation coefficients at AWSs used in the Antarctic wide evaluation byJones and Lister(2015).

In the summer season there are occasions where the R² value drops below 0.7.

The CFSR scatter plot in Fig. 3.3c indicates that the reduced correlation coefficient is primarily caused by large RMSEs when observed summertime temperatures fall below −10 ^◦C . Larger biases and RMSEs are seen in the wintertime scatter plots (Fig. 3.4) than the summertime (Fig. 3.3), as discussed previously this seasonal difference is markedly larger for ERA-I and JRA.

The standard deviation ratios show that both CFSR and ERA-I produce larger standard deviations than observed across all seasons at TI and BP (see Table A.2, Figs. 3.3a, 3.3c, 3.4a and 3.4c). For example, Fig. 3.3c shows that CFSR produces temperatures up to 10 ^◦C cooler than those observed. ERA-I also produces

52

ANEVALUATION OF FOUR GLOBAL REANALYSIS PRODUCTS USING IN SITU OBSERVATIONS IN THEAMUNDSENSEAEMBAYMENT

−30 −25 −20 −15 −10 −5 0 5

Figure 3.3: Summertime temperatures at BP AWS compared to 2-m temperature from reanalyses, with statistics shown in text. Top left (a) ERA-I; top right (b) JRA-55, bottom left (c) CFSR; and bottom right (d) MERRA.

anomalously cold temperatures and this causes the larger than observed standard deviations seen in Figs.3.3a and 3.4a. MERRA has a large SD ratio in the summer months, as shown in Fig.3.3d, in summer MERRA displays more variability around the mean than was observed.

• All four reanalyses produce temperatures that are colder than observed, the magnitude of this bias varies between ∼ 2^◦C (ERA-I) and ∼ 6^◦C (MERRA).

• ERA-I and JRA-55 show larger cold biases in the austral winter, suggesting the products are struggling to simulate the strong surface based inversion that is

3.4.AWS COMPARISON 53

Figure 3.4: Wintertime temperatures at BP AWS compared to 2-m temperature from reanalyses, with statistics shown in text. Top left (a) ERA-I; top right (b) JRA-55; bottom left (c) CFSR; and bottom right (d) MERRA.

characteristic of Antarctic winter.

• ERA-I temperature biases and correlation coefficients are comparable with those found in the Antarctic wide study of Jones and Lister(2015). However MERRA contains a much larger temperature bias than was found byBracegirdle and Marshall(2012) when compared with East Antarctic AWS.

54

ANEVALUATION OF FOUR GLOBAL REANALYSIS PRODUCTS USING IN SITU OBSERVATIONS IN THEAMUNDSENSEAEMBAYMENT