In contrast to the ERA5 product from 1979 onward (hereafter referred to simply as ERA5), the ERA5 back extension (1950-1978; hereafter referred to as ERA5 BE) benefits from the assimilation of 6-hourly tropical cyclone best track pressure reports from the International Best Track Archive for Climate Stewardship (IBTrACS; Knapp et al. 2010). These reports were also ingested into two previous centennial reanalyses produced by ECMWF, ERA-20C and CERA-20C, in which only surface observations were assimilated. However, strict quality control led to a rejection of many of these observations, with the result that many tropical cyclones were not well represented.
With the aim to improve on this situation in ERA5 BE, quality control on these observations was switched off. Initial tests showed a good representation of several tropical cyclones. However, later, during the production of ERA5 BE, it was discovered that for quite a number of tropical cyclones the ERA5 central pressure was significantly lower than that of the observations. The reason for this discrepancy was twofold: (1) by mistake too much weight was given to these observations in the ERA5 BE data assimilation system; and (2) the data assimilation system was not provided with the information that these observations actually represent a minimum in the pressure field. As a result, the analysis fit at observation location is usually good, but often the assimilation system places the minimum pressure away from the observations and this minimum is then typically deeper.
Therefore, the compromised quality of the tropical cyclone analyses in ERA5 BE is not related to the quality of the pressure estimates from the IBTrACS observations, but a result of the sub-optimal method used to assimilate them. This method not only leads to an overestimation of the intensity of tropical cyclones in ERA5 BE, but it also results in a discontinuous behaviour in comparison with the ERA5 segment from 1979 onwards where the intensity of tropical cyclones is generally underestimated.
As a result many tropical cyclones are unrealistically intense in ERA5 BE. Therefore, ERA5 BE should not be used for the analysis of tropical cyclones.
Effect on hourly products
In general the impact is limited to the vicinity of tropical cyclones and has at most a small influence on the ERA5 BE reanalysis products elsewhere.
Figure 1 displays the time evolution of the estimated central pressure of Typhoon "Doris" in May 1950 from the IBTrACS observations alongside the analysed mean sea level pressure (mslp) at the observation locations and the minimum central pressure of the storm developed in ERA5 BE. It can be seen that while along the cyclone path the analysed mslp values are higher than or close to observations, the IFS model develops a much stronger storm away from the path with mslp reaching a minimum of 873 hPa at 12 UTC on 11 May 1950. The map with the position of the typhoon in ERA5 BE for the same date is shown in Figure 2.
Figure 1. Time evolution of the estimated central mean sea level pressure (mslp) from IBTrACS (black line) and the ERA5 analysis (red line) for Typhoon "Doris". The analysed mslp values at the observation locations are plotted in blue.
Figure 2. Mean sea level pressure map showing the position of Typhoon "Doris" at 12UTC, 11 May 1950 in the ERA5 analysis. The red line shows the path of the typhoon, reconstructed from the 6-hourly IBTrACS observations. Isobars are every 10hPa.
An example of a tropical cyclone relatively well represented in ERA5 BE is Hurricane "Charley" in August 1951. Its particularity is that, as shown in Figure 3, it intensified three times. Based on IBTrACS observations, the first intensification was in the Caribbean Sea. The second one was while approaching the Yucatan Peninsula, during the night of 19 to 20 August 1951. It weakened slightly after it made landfall, and then it re-intensified over the Gulf of Mexico prior to reaching the mainland Mexican coast. Figure 3 shows that ERA5 BE was able to capture two of the intensification stages. The map with the position of Hurricane "Charley" at 00UTC on 20 August 1951, in ERA5 BE, is shown in Figure 4. The tropical cyclone path is also shown in this same figure, though only as far as the Yucatan Peninsula.
Figure 3.Time evolution of the estimated central mean sea level pressure from IBTrACS (black line) and the ERA5 BE analysis at the observation locations (red line) for Hurricane "Charley" in August 1951.
Figure 4. Mean sea level pressure map showing the position of Hurricane "Charley" at 00UTC on 20 August 1951 in the ERA5 BE analysis. The red line shows the path of the hurricane, reconstructed from the 6-hourly IBTrACS observations. Isobars are every 10hPa.
Differences in monthly means between the ERA5 back extension and ERA5
Although not made publicly available, ERA5 BE was continued into 1979 and 1980 (in this section, this segment of ERA BE will be referred to as ERA5 BE79-80). This allows for a comparison with the previously released ERA5 product that starts in 1979. In particular, such comparison makes it possible to assess the impact of the assimilation of IBTrACS observations in ERA5 BE79-80 since, as mentioned in the Introduction, these observations were not ingested in ERA5. This impact will depend, in part, on the intensity and duration of the storms and the region in which they develop. Although a comprehensive assessment exceeds the scope of this page, the general characteristics and magnitude of the impact of IBTrACS observations can be understood by looking at one month (August 1980) during the 2-year period of overlap between ERA5 BE79-80 and ERA5.
Figure 5. Locations of IBTrACS observations assimilated in ERA5 BE in the North Atlantic during the month of August 1980.
Figure 6. Monthly mean differences between surface fields from ERA5 BE and from ERA5 for August 1980. The differences are shown for (a) surface pressure in hPa, (b) 10m wind speed in m/s, (c) 2m temperature in K, and (d) total precipitation in mm/day.
Figure 5 shows the IBTrACS pressure observations that were assimilated in ERA5 BE79-80 in August 1980 in the North Atlantic region. This month was marked in particular by Hurricane Allen (Cat. 5), which moved through the Caribbean Sea and the Gulf of Mexico before making landfall in Texas on 10th August. The impact of the IBTrACS observations can be seen clearly in the monthly mean differences between the ERA5 BE79-80 and ERA5 displayed in Figure 6. As expected, ERA5 BE79-80 exhibits lower surface pressure (up to -0.95 hPa; see Fig. 6a) and higher wind speed (Fig. 6b) where IBTrACS observations are assimilated. The impact on 2m air temperature (Fig. 6c) is small (typically < 0.1 K) over the ocean. Finally, larger precipitation amounts are also found in ERA5 BE79-80 along the IBTrACS observations.
Two main takeaways from Figure 6 are that, when considering the monthly means of atmospheric fields (see the impact on ocean waves below), (1) the impact of IBTrACS observations is clear but limited to the vicinity of these observations, and (2) the magnitude of this impact is of the same order or smaller than differences caused by other factors.
Effect on the ERA5 ocean-wave climatology
Ocean waves are an integrator for wind (i.e. they, accumulate, or integrate the wind) and thus provide a good measure of how the effect of the too intense tropical cyclones spreads over larger regions. The effect on ocean waves can be seen in particular by looking at the statistical distribution of significant wave height, particularly the upper tail of their frequency distribution. Here, significant wave height (SWH) refers to the average height of the highest third of surface ocean waves. Percentiles of the yearly distribution of SWH at each ice-free ocean grid point were computed for ERA5 BE and ERA5 based on hourly output. For each year, yearly maps were generated for various percentile values (90th, 95th, 99th). The maps shown in Figure 7 compare the maximum values of the yearly 95th and 99th percentiles (denoted p95 and p99) in ERA5 BE and ERA5. Figures 7c clearly shows higher p99 values in ERA5 BE in the ocean basins where tropical cyclones occur most frequently, namely the western North Atlantic, the western South Indian Ocean, and the western North Pacific. In comparison, the differences between the p95 maps (Figs. 7a and 7b) are much smaller. This suggests that the impact of the IBTrACS observations on ocean waves becomes apparent only for the very upper tail of the distribution of SWH (99th percentile and above).
Figure 7. Maps showing the maximum value of the yearly 95th percentiles (top row) and the yearly 99th percentiles (bottom row) of significant wave height (in meters) in ERA5 BE 1950-1978 (left column) and ERA5 1979-2019 (right column). Further details about the method used to generate the maps are given in the text.
Knapp, K. R., M. C. Kruk, D. H. Levinson, H. J. Diamond, and C. J. Neumann, 2010: The International Best Track Archive for Climate Stewardship (IBTrACS). Bull. Am. Meteorol. Soc., 91, 363–376, https://doi.org/10.1175/2009BAMS2755.1.