Implementation of IFS Cycle 47r1

Assimilation

LWDA uses first guess from Early Delivery

• The first guess used in LWDA now comes from the analysis derived in the early delivery window.
• This gives effectively 4 extra outer loops in the LWDA system, as they were computed in the early delivery window, albeit with fewer observations.

Weak-constraint 4D-Var

• A new estimate of the model error covariance matrix has been computed from a climatology of the model error vectors estimated by the current weak-constraint 4D-Var
• This new implementation of weak-constraint 4D-Var corrects the diagnosed cold and warm biases of the model over 100 hPa, reducing the mean error by up to 50%.
  

Revision of skin temperature background errors in the context of TOVS sink variable

• New spatially and in time varying back-ground errors for skin temperature have been derived based upon output from the Ensemble Data Assimilation (EDA) to improve the way the surface skin temperature is allow to adjust to the radiance assimilation scheme during the 4D-VAR. The approach enhances the characterization of the background-error variances for the skin temperature, by providing flow dependent background errors which vary spatially and in time during the 12-hour assimilation window. This is particularly important for land surfaces which can be very heterogeneous and where errors can change rapidly during the day, synoptically and seasonally.

Timestep for last 4D-Var minimisation

• The time step in the last minimisation cycle in the 4D-VAR has been set the same for the outer and inner loop to correct spurious gravity-wave-like increments generated during the 4D-Var analysis.
• This change leads to better balanced initial conditions and a statistically significant increase in forecast skill.

Observations

ATMS observation error correlations

• In Cycle 46r1 inter-channel error correlations were accounted for in Suomi-NPP ATMS.

• In Cycle 47r1 inter-channel error correlations will be introduced for NOAA-20 ATMS consistently with Suomi-NPP ATMS.

Channel-specific aerosol rejections for IR sounders

• The use of infrared (IR) data (AIRS, IASI, CrIS) is enhanced by allowing high-peaking channel radiances to be assimilated at those locations where lower-peaking channels are rejected due to being contaminated by aerosol. Previously, the aerosol detection scheme did not distinguish between affected and unaffected channels, but instead rejected full IR spectra where presence of aerosol was found.

Spline interpolation in the 2D GPS-RO bending angle operator

• A cubic spline interpolation is introduced for vertical interpolation of Log(refractivity) within the GPS-RO forward operator.
• This removes unphysical structure in the vertical leading to more realistic vertical variability of the forward modelled bending angle background departures.

Model

Surface albedo changes

• A number of improvements have been made to the specification of the shortwave albedo of the land surface, snow and sea ice. This include: 1) the use of six climatological fields from MODIS, 2) a better handling of the spectrum from the MODIS datasets and 3) the removal of the artificial adjustment for the prognostic albedo which is now obsolete because of the changes in 2).

Update to greenhouse gases and total solar irradiance

• In Cycle 47r1 the Total Solar Irradiance (TSI) and Green House Gases (GHG) timeseries are updated with more recent ones from CMIP6.

Quintic interpolation in semi-Lagrangian advection

• As the horizontal resolution increases, more small scale waves are resolved in the horizontal direction but their representation in the vertical direction poses a challenge when vertical resolution is not appropriately increased.  As increasing the vertical resolution is expensive, a more cost-effective  solution was found by increasing the order of vertical interpolation of advected fields to the departure point in the semi-Lagrangian advection of temperature from third order (i.e. cubic, where a Lagrange polynomial of degree 3 interpolates a field using 4 neighbouring points) to fifth order (i.e. quintic, where a Lagrange polynomial of degree 5 interpolates a field using 6 neighbouring points),

Drag coefficient for very strong winds

• The drag coefficient over the oceans is reduced for strong winds to account for observational evidences whereby the drag coefficient reduces sharply for high winds.

Convection scheme

• Stability corrections to the mid-level and deep convective closures (neutral) and reduced bounds for parcel perturbations
• Rewriting the diffusion momentum solver as for scalars (neutral)
• Revision of the CIN diagnostic using virtual potential temperature instead of equivalent potential temperature. The revised CIN is now much reduced and is closer to values expected by forecasters.

Tangent-linear physics

• The mass-flux limiter, used in the tangent-linear and adjoint convection schemes to ensure that the CFL criterion is fulfilled in the vertical, is now reduced for time steps shorter than 900 seconds. The mass-flux limiter is dependent on the model time step to avoid the occurrence of instabilities in the 4D-Var minimizations.

TC “size” will be represented by radii for mean 10m wind thresholds of 18, 26 and 32 m/s (34, 50 and 64 knots) to denote the furthest distance (in metres) away from the centre of the TC at which each of the wind speed thresholds are exceeded. Each of these are computed for each of four earth-relative quadrants, i.e. in NE, SE, SW and NW, delivering a total of 12 “size metrics” for each TC at each time step.

To arrive at these metrics the code scans, in outwards fashion, all model gridpoints within each quadrant. Values are computed for the HRES and ENS for all TCs that are either present in the initial conditions, or that develop during the forecast integrations (i.e. TC “genesis”).

The new values, which we will call “wind radii”, are included as supplementary information within the BUFR message that currently contains the TC tracks (pairs of latitudes and longitudes) and the TC intensities (minimum pressure and overall maximum wind speed).

Implementation of the TC wind radii product was motivated by continuing model developments such as the recent re-tuning of the Charnock parameter for very high wind speeds, which goes live in model cycle 47r1, and which improves lower tropospheric wind speeds around intense storms.

Computation of the wind radii values is executed after the TC tracking, in a post-processing step, using code developed by NOAA which is available in the public domain.

The wind radii values are non-zero wherever the wind thresholds are exceeded, otherwise they are set to zero. If the track is missing at specific time step(s) then values are replaced by a missing value indicator.  No changes were made to the TC tracker software.

Each tropical cyclone identified in an IFS forecast has 10m mean wind attributes associated. In previous cycles this was just the maximum within the cyclone circulation, now there are also TC size metrics (see above). Historically such metrics were adjusted upwards, compared to raw model values, by multiplying by 1.08, in part to provide more realistic values. This dates back to a time when model resolution was much lower than it is now. With cycle 47r1 we have chosen to remove this adjustment factor. This decision coincides with a concurrent model change that increases the 10m winds slightly in extreme conditions (see meteorological content section).

ECMWF output that references such values includes the TC BUFR files, and on the web the TC-related Lagrangian meteograms. The TC classifications used for TC strike probabilities (and their ecCharts counterparts) are also impacted.

It is unlikely that users will notice a difference.

 In the extra-tropical stratosphere, the new cycle brings large improvements, such as 2-5% error reductions for temperature and geopotential at 100 hPa and 5-15% at 50 hPa. In the tropics, there is an apparent degradation in the order of 1-3% in upper air scores when forecasts from each cycle are verified against those cycles' own analyses, but when forecasts are verified against observations the impact is neutral. Verification against observations shows that upper-air changes in the tropics are overall neutral, with small improvements and deteriorations balancing each other out. One exception is 250 hPa temperature in the tropics, where a deterioration of 1-3% is seen also against observations. This is mainly due to a small (about +0.1 K) shift in the mean, resulting from the model changes in the new cycle.

The new cycle improves forecasts of several near-surface parameters, most notably 2m temperature and humidity (by about 0.5%) both in the extra-tropics and, when verified against observations, also in the tropics. Extra-tropical 10m wind in the HRES is slightly improved, as well as total cloud cover both in ENS and HRES. Tropical 10m wind is very slightly deteriorated. Significant wave height is mostly neutral against observations and improved against own analysis.

Changes in forecast performance for tropical cyclone (TC) tracks are statistically neutral. There is a small deterioration in TC intensity in terms of central pressure, but the pressure-wind relationship has been improved, and there is a small improvement in mean maximum wind speed errors.

The impact of 47R1 on weekly mean anomalies in the extended-range forecasts is overall neutral, except for some improvement in 50hPa meridional velocity and temperature while there is a slight degradation in week 1 in the tropical troposphere. The MJO is 3-4% weaker in the extended-range in the new cycle, which implies a slight degradation.