Some aspects of the quality of the ERA-40 analyses.

• General characteristics
• Surface parameters
• Forecast performance
• Humidity analysis and rainfall over the tropical oceans
• Arctic analyses since 1989
• Stratosphere since 1989
• Ozone analysis since 1989
• Radiation budget
• Ocean waves

Several aspects of the quality of the production ERA-40 analyses are discussed below. The discussion is based on ECMWF's monitoring of production and on feedback received from ECMWF's partners in the ERA-40 project and from the polar and stratospheric research communities. The aim is to help users to select the analyses for their studies from the period or from the analysis stream that would best suit their application. Emphasis is placed on some of the problems encountered in the production analyses. Rerun analyses are likely to be of significantly better quality in a few respects such as identified below, but otherwise they should be of similar quality to the initial production analyses. The existing analyses can thus be used with reasonable confidence for most studies, although for some applications it is likely to be worth waiting for the "final ERA-40 analysis dataset" that will include results of the rerun.

General characteristics.

The temporal consistency of the ERA-40 analyses on synoptic timescales is better than that of the earlier ERA-15 analyses. This is most likely due to the 3-dimensional variational data-assimilation employed in ERA-40. Departures of the model atmosphere from observations are determined using model values at the time of observation rather than from the nearest synoptic hour.

Analysis quality inevitably changes over long timescales due to changes in observing systems. These systems evolved much more substantially during the more than 40 years covered by ERA-40 than during the shorter ERA-15 period. Sensitivity to observing system changes can be seen in particular in the quality of the tropical and southern hemispheric analyses. Throughout the period 1957-2001 there is a gradual tendency for observation accuracy and coverage to improve (radiosonde coverage being an exception), but there are a number of years during which analysis quality will improve stepwise in certain characteristics due to the introduction of satellite instruments: 1972 VTPR, 1979 TOVS (MSU/HIRS/SSU), Cloud Motion Vectors and TOMS & SBUV, 1987 SSM/ I, 1991 ERS Scatterometer and Altimeter, 1998 ATOVS (AMSU-A). Major additions to the in-situ observing system have been the deployment of drifting ocean buoys, first for FGGE in 1979 and then for the TOGA experiment in the mid 1980s, and the introduction and enhancements of observations from commercial aircraft from the 1970s onwards.

There are also changes in the observing systems that have short-term effects on analysis quality. These are due mainly to instrument failures or missing data, or to undetected poor-quality data entering the production analysis system. During the assimilation detailed information on how each observation is used in the analysis is archived in so called "feedback" files. A comprehensive diagnosis of this information, which will document the effect of changes to the observing system, will be carried out when production is complete.

Surface parameters.

The severe cold bias in the ERA-15 surface and near-surface temperatures during winter and spring over northern Eurasia and America has been corrected in ERA-40. In some areas a smaller warm bias has however been noted in the ERA-40 analyses, especially during springtime. The ERA-40 analysis does not use unrepresentative surface wind observations, from isolated islands for example. This makes the surface wind analyses and the surface turbulent exchanges more realistic. A revised and more accurate surface orography description has been used in ERA-40. This causes some quite large local differences in surface pressure and other near-surface parameters between ERA-40 and ERA-15, particularly over Antarctica.

Time series of global-mean snow mass from 1989 to 1997 exhibit low values from 1992 to 1994 due to a bug introduced into the snow analysis, and the analysis up to 1997 also suffers, though to a lesser extent, from a miscoding by ECMWF of Canadian snow-depth observations that moved some observation dates to later within the same month. These defects will be corrected in the rerun.

Forecast performance.

The accuracy of the forecasts run from the analyses provides a good indication of the general quality of the synoptic analyses.

Forecasts to 36 hours ahead are run routinely as part of the production system. The forecasts of tropospheric and stratospheric winds and tropospheric temperatures from the ERA-40 analyses for 1989 have been shown to be much more consistent with corresponding verifying analyses than was the case for ERA-15 or for ECMWF operational forecasts in 1989. The ERA-40 forecasts also have the lowest temperature errors in the extratropical stratosphere below 10hPa. In verifications of short-range forecasts against radiosonde data, the ERA-40 forecasts of stratospheric temperature stand out much more clearly as the best, in the tropics as well as the extratropics.

Forecasts to ten days ahead are being run twice daily from the ERA-40 analyses. Forecasts for five years, 1958, 1959, 1973, 1989 and 1996, have been completed at the time of writing. Results from 1989 and 1996 indicate similar performance in the two years, a little better than ECMWF operations in 1996 and substantially better than ECMWF operations in 1989. In the northern hemisphere there is a clear improvement in forecasts from 1958 to 1973 and from 1973 to 1989. The 1958 and 1959 forecasts are nevertheless quite good in the medium range, where they typically lose skill not much more than one day earlier than the 1989 forecasts. Verification of the forecasts from earlier periods over much of the southern hemisphere is more problematic due to the paucity of observations and model-dependence of verifying analyses, but the indications are of a much poorer analysis quality for the earlier years. The 1958, 1959 and 1973 forecasts lag in skill by up to three days in the medium range over a relatively well-observed region including Australia and New Zealand.

Humidity analysis and rainfall over the tropical oceans.

The most serious problem diagnosed in the ERA-40 analyses is excessive tropical oceanic precipitation in later years, particularly in stream 1 after 1991. The stream-1 analyses are moistened over tropical oceans by the assimilation of HIRS and SSM/I data. This moistening is rejected by the assimilating model in the subsequent background forecasts, leading to higher rainfall rates over the tropical oceans than produced by the model when run either in climate-simulation mode or in the stream-2 (pre-satellite) data assimilation. A substantial increase in rainfall rates from the second half of 1991 onwards was due in part to effects of volcanic aerosols on HIRS infrared radiances following the eruption of Mt. Pinatubo. These effects were not included directly in the forward radiative transfer model used in the variational analysis. Instead they had to be absorbed into the bias corrections applied to the radiance measurements. This was a problem especially for data from the NOAA-12 satellite that became operational just around the time Mt. Pinatubo erupted. Inadequately corrected infrared radiance biases tend to result in humidity changes in the tropical troposphere, since the relatively low background errors in temperature force analysis changes to be predominantly in humidity. A further complication came from poor bias correction of SSM/I data. This was corrected for ERA-40 analyses from January 1993. A revised thinning, channel-selection and quality control of HIRS radiances has been developed and tested, giving reduced (though still relatively high) tropical precipitation (and slightly improved short-range forecast verifications). It has been used for stream 1 from 1997 onwards, and will be used in stream 3 when it has progressed to the time that HIRS data first become available.

This problem provides a major motivation for rerunning the latter part of ERA-40. The extent of the improvement that can be achieved before the rerun is not yet clear, but the main variations in precipitation arising from direct Pinatubo-aerosol effects on HIRS radiances and from the changes made to the use of SSM/I and HIRS radiances should at least be removed.

Despite this problem, use of dynamical parameters such as vertical velocity from ERA-40 in conjunction with satellite measurements of the radiation budget has been found to provide a powerful tool for diagnosing the performance of climate models in the tropics.

Arctic analyses since 1989.

A further problem of concern is cold bias in the lower troposphere (below about 500 hPa) over ice-covered oceans in both the Arctic and the Antarctic. A related problem in Arctic precipitation has also been identified. These polar cold biases arise from the assimilation of HIRS radiances. Changes to the thinning, channel-selection and quality control of the infrared data that were introduced for analyses from 1997 onwards to reduce the tropical precipitation bias have also virtually eliminated the cold polar biases. Improved pre-1997 polar analyses are thus expected from the rerun.

Stratosphere since 1989.

Although the eruption of Mt. Pinatubo caused problems in the assimilation of HIRS infrared radiances, with consequences for the humidity analysis in the tropical troposphere, the ERA-40 analyses appear to capture well the stratospheric warming that was caused by increased solar heating due to the aerosols. The subsequent cooling is also well captured. 100hPa temperature time series show a temperature maximum in 1992 and cooling thereafter. During this period, the fit of the model background to the observed microwave radiances from the MSU-4 channels on the NOAA-11 and NOAA-12 spacecraft showed little trend, whereas the radiances themselves indicate a marked warming immediately after the Pinatubo eruption, and a cooling in later years. The ERA-40 analyses thus accurately match the low-frequency variability in these data, which are representative of layer-mean lower stratospheric temperatures.

The ERA-40 analyses provide a good representation of the QBO, as judged by comparisons with wind observations and independent analyses. The assimilating model in ERA-40 tends to exhibit significant biases in upper-stratospheric temperatures, and analysed temperatures are thus sensitive to the availability and use of satellite measurements in the upper stratosphere, supplied first by the SSU instrument and subsequently by AMSU-A over the period since 1979. Questions concerning the temperature climatology of the upper stratosphere in general and of the whole stratosphere over Antarctica are under investigation. Stratospheric humidity evolves in the assimilating ERA-40 model, but no observations are assimilated. Its distribution is clearly a major improvement on the simple prescription of a uniform specific humidity of 2.5x10-6 in ERA-15, but analyses are generally drier than seen in UARS data for the 1990s and the tropical stratospheric "tape recorder" runs much too fast.

Ozone analysis since 1989

TOMS and SBUV ozone data have been assimilated in stream 1 from January 1991. The TOMS instrument cannot provide measurements in the polar night, and data are unavailable globally for several periods, notably in 1995 and 1996. The analyses draw closely to the TOMS data when they are available, capturing the observed interannual variability. They also provide plausible values for the periods when TOMS data are missing. The assimilation of SBUV data is evidently sufficient to provide a reasonable control on the assimilating model, since in the absence of data assimilation the model tends to produce ozone values that are too small in the tropics and too large in springtime at high latitudes. In October 1996 new estimates of background error covariances were introduced. This has reduced a problem of too-low values in the upper troposphere and too-high surface values. Some quality-control problems have also been encountered.

A more consistent time-sequence of ozone analyses should be a further benefit of the planned rerun.

Radiation budget.

Clear-sky fluxes appear reasonable and their climatologies should be of significant utility. The short-wave clear-sky fluxes are however dependent on the accuracy of the prescribed surface albedo and aerosol climatologies. Any spurious changes in the analysed humidity or temperature fields will affect the clear-sky long-wave fluxes.

Cloud fraction appears to be reasonably well simulated, apart from an underestimation over stratocumulus regions and possibly an overestimation over non-convective oceanic regions. Radiation budget fields suffer from deficiencies in the radiative properties of the clouds, and are not recommended for use in studies where accurate fluxes are required. It is unlikely that the planned rerun will bring substantial improvement in this area.

Ocean waves.

There were no observations of ocean waves that could be assimilated in the initial stream-1 years of ERA-40. The monthly mean wind and wave fields for this period compare well with observations, but the analyses exhibit peaks in synoptic significant wave heights that are lower than observed. ERS-1 Fast Delivery Product (FDP) altimeter wave-height data were assimilated in ERA-40 from December 1991 onwards. The data are, however, of poor quality during the first two years due to an external processing error. Assimilation of the FDP data was halted as soon as it was realised that there was a data-quality problem, production having reached May 1993. The rest of 1993 was run without assimilation of wave-height data. Assimilation was resumed from January 1994 using good but uncalibrated ERS-1 FDP data up to May 1996. The known calibration correction was not applied because, although it would have improved analysed wave heights, it would have given poorer, too high mean wave periods. FDP ERS-2 measurements of wave height have been assimilated in ERA-40 from June 1996 onwards. Results to date indicate that this assimilation has improved the wave-height analyses, but mean wave periods do not compare as well with observations as they do for the period from May 1993 to January 1994.