Differences between cycles.

This section highlights the changes introduced by the latest cycle.  This helps the user to understand and take into account differences between forecast data from the earlier cycle and that from the latest cycle.

Cycle 43r3 and earlier

Large scale precipitation in cycle 43r3 and earlier

In warm airmasses, due to issues loosely related to aerosol, precipitation accumulations were:

  • rather enhanced in low lying coastal parts (See Fig9.5.1).
  • rather deficient in areas where orographic enhancement might be expected (See Fig9.5.2). 

 

  

Fig9.5.1(left):  Precipitation accumulation T+0 to T+54 from HRES data time 00UTC 10 July 2017 for the operational 43r3 run.  Anomalous enhanced precipitation totals are focused along the coastlines of England (Cornwall and Bristol Channel) and France (Golfe de Saint-Malo) just over the adjacent sea.

Fig9.5.1(right): Precipitation accumulation T+0 to T+54 from an equivalent run of the HRES, but with the physics changes incorporated into cycle 45r1.  The precipitation maxima are no longer offshore along the coast, but moved inland with a more realistic spatial structure. Compare with Fig9.5.1(left).


  

Fig9.5.2(left): Precipitation accumulation T+0 to T+48 from HRES data time 00UTC 13 May 2017 for the operational 43r3 run.  Anomalous enhanced precipitation totals are focused along the coastlines of western and southern parts of Ireland and Great Britain just over the adjacent sea. Also, apart from the high ground of southwest Scotland, rainfall is deficient over areas where orographic enhancement of precipitation occurred.

Fig9.5.2(right): Precipitation accumulation T+0 to T+48 from an equivalent run of the HRES, but with the physics changes incorporated into 45r1.  The precipitation maxima are no longer offshore and along the coast, but instead there are areas of orographically enhanced precipitation that have a much more realistic spatial structure. Rainshadow is slightly less extensive, though it is not immediately obvious whether this is better or worse. Compare with Fig 9.5.2(left).

 

Tropical cyclones in cycle 43r3 and earlier

Impact of coupling between IFS Atmospheric and Ocean models

HRES (ECMWF forecasts cycle 43r3 and earlier) was not coupled with NEMO but retained the initial sea-surface temperature anomalies throughout the forecast period.  HRES tended to deepen relatively slow-moving tropical cyclones too much.  This was due to the lack of ocean/atmosphere coupling in HRES which often kept the ocean too warm when it should be cooling.  In reality the strong winds and slow movement would induce turbulent mixing of the very warm surface waters with cooler waters from deeper in the ocean, reducing sea-surface temperatures, and hence inducing less deepening of the storm.  ENS was and continues to be coupled with the ocean and didn't suffer from this problem.

 

Fig8.1.10.11: Comparison of a coupled HRES (used in ECMWF forecasts from cycle 45r1 released June 2018) and an uncoupled HRES (used in ECMWF forecasts cycle 43r3 and earlier) forecast following tropical cyclone NEOGURI. Left: Sea-surface temperatures in °C; Right: Central pressures for NEOGOURI in hPa; Uncoupled HRES - Blue, Coupled HRES - Red, Observed values - Black. HRES data time 00UTC 6 July 2014.

The uncoupled HRES (ECMWF cycle 43r3 and earlier) uses initially high sea-surface temperatures through the forecast period maintaining high availability of heat and moisture energy throughout.  This leads to over-deepening of the tropical storm by around 20hPa.  Depth is better forecast by the coupled HRES (ECMWF forecasts from cycle 45r1 released June 2018) which uses the realistic sea-surface temperature predictions provided in part by NEMO, and consequently the forecast central pressure of the tropical storm is much nearer to reality. ENS was and continues to be coupled with the ocean and doesn't suffer from this problem.


 

 Fig8.1.10.14A: Intensity probability for NORU up to 10 August 2017 (T+240) based on ENS and HRES forecast, data time 12UTC 1 Aug 2017.

Top: ENS-based probabilities for tropical cyclone NORU to fall into each of 5 intensity categories, shown at 6hr intervals out to 10 days.

Centre and Bottom:  Lagrangian meteograms of distribution of the ENS mean (dotted line) and HRES (solid line) for the 10m wind (kn), and MSLP (hPa) at NORU's centre. The HRES forecast of central pressure is considerably lower than the ENS forecast, and the winds are consequently stronger.

HRES values used to look plausible but had to be used with caution. In the case of NORU (Fig8.1.10.14) HRES predicted the tropical storm to deepen below 900 hPa but one would have expected this to be exaggerated given the slow movement of NORU together with the absence of ocean coupling at that time in HRES (and also the fact that NORU was then remote from land).

Stochastic Kinetic Energy Backscatter (SKEB) in cycle 43r3 and earlier

SKEB provided a numerical description of the physical process of upscale kinetic energy transfer. In the real atmosphere, that process has been observed to occur at all scales. In the numerical model, there is no mechanism to enable energy to transfer from the sub-grid scales to the resolved scales. SKEB represented the uncertainty associated with this "missing" process by randomly perturbing the rotational flow (via the streamfunction), with perturbations that were modulated by an estimate of the local sub-grid dissipation rate. The SKEB scheme was switched off with cycle 45r1 in early 2018. This was because it had become ineffective since ECMWF introduced the cubic octahedral grid, due to compatibility issues.


Prior to cycle 45R1

Problems with coupling of HRES and Dynamic Ocean Model prior to cycle 45R1.

Note: Coupling of HRES with the Dynamic Ocean Model was introduced in cycle 45r1 released in June 2018.  Users should be aware when inspecting model data before release of this cycle that atmosphere/ocean coupling with the HRES was not in effect.

Upgraded Bathymetry in cycle 45r1

Upgraded bathymetry (water depth) is used the wave models in IFS Cycle 45r1(HRES-WAM and ENS-WAM). It is based on ETOPO1 (1/60 degree global data). NB: HRES-SAW withdrawn 7 Jan 2020.

Fig 12.1: New bathymetry used for HRES-WAM for Europe together with the difference with the previously used bathymetry (ETOPO2).


Fig 12.2: New bathymetry used for ENS-WAM for Europe together with the difference with the previously used bathymetry (ETOPO2 = 2/60 degree global data).

This change was in part driven by users pointing out that the previous bathymetry for the Baltic Sea was quite erroneous in a few places.  Changes in water depth will mostly affect the wave fields in coastal areas, generally resulting in higher wave heights where the water has become deeper and vice-versa.  Moreover, some WAM grid points have changed from sea to land (i.e. no waves at those points),  and vice versa.  These locations are respectively shown in the right-hand portion of Fig12.2 above, with green and black shadings (you may need to zoom into the pictures).  This change of land/sea points will be visible for some coastal locations in the Wave ENSgrams (Wavegrams) and for users relying exclusively on the wave model values at those locations.


Prior to cycle 46R1

Techniques to derive perturbations prior to cycle 46R1

Until cycle 46r1 we had less EDA members than ensemble forecast members (ENS members).  Once the different sets of SVs had been separately calculated over the northern and southern hemispheres and over the tropics between 30°N and 30°S, they were linearly combined (using coefficients randomly sampled from a Gaussian distribution) and added to the EDA perturbations to make a set of 25 global perturbations.  The signs of these 25 global perturbations were then reversed to obtain another set of 25 “mirrored” global perturbations.  This gave a total of 50 global perturbations for 50 alternative analyses and forecasts.  Consecutive members therefore had pair-wise anti-symmetric perturbations. The anti-symmetry may, depending on the synoptic situation and the distribution of the perturbations, disappear after one day or so, but can occasionally be noticed 3-4 days into the perturbed forecasts (see Fig5.3 and Fig5.4).

  • The EDA perturbations and the Singular Vectors were added together give the perturbation for each ENS member.  Thus:
    • ENS member 1 Analysis = HRES Analysis + (EDA member 1 - EDA mean) + SV Perturbation 1
    • ENS member 2 Analysis = HRES Analysis - (EDA member 1 - EDA mean) - SV Perturbation 1
    • ENS member 3 Analysis = HRES Analysis + (EDA member 2 - EDA mean) + SV Perturbation 2
    • ENS member 4 Analysis = HRES Analysis - (EDA member 2 - EDA mean) - SV Perturbation 2
      •  
    • ENS member 49 Analysis = HRES Analysis + (EDA member 25 - EDA mean) + SV Perturbation 25
    • ENS member 50 Analysis = HRES Analysis - (EDA member 25 - EDA mean) - SV Perturbation 25

This gave a set of 50 global perturbations for the 50 alternative analyses and forecasts.  Two neighbouring ENS members neighbouring couplets of ENS members were not truly independent and they shared the same perturbation, just with the sign reversed.


Fig12F.1: A comparison between ENS perturbations. Top derived from 25 EDA members using plus/minus symmetry (each EDS perturbation added to ENS member1, subtracted from ENS member2 etc and thus couplets of ENS members are not completely independent).  Bottom derived from 50 EDA members (EDS perturbation1 added to ENS member1, EDS perturbation2 added to ENS member2 etc and thus each ENS member is independent).  Diagram taken from Slide7 of Cycle46R1 Overview.


The 50 ensemble member forecasts are made from slightly different initial conditions (perturbations) which are designed to represent the uncertainties inherent in the analysis.  The initial perturbations are constructed using the singular vector (SV) technique, and perturbations generated from the ensemble of data assimilations (EDA).  

Points to note:

  • Perturbations based on singular vectors (SV) and perturbations based on ensemble of data assimilations (EDA) have different characteristics:
    • Geographically, perturbations based on EDAs are less localised than based perturbations based on SVs.  In particular, they have a larger amplitude over the tropics.
    • Spectrally, perturbations based on EDAs are smaller in scale.
    • Vertically, perturbations based on EDAs are more barotropic than perturbations based on SVs.  Perturbations based on SVs show westward tilt with height typical of baroclinically unstable structures.

    • At initial time, SV-based perturbations have a larger amplitude in potential energy than kinetic energy; perturbations based on EDA have a similar amplitude in potential and kinetic energy.
    • EDA perturbations grow less rapidly.
    • An ensemble based on EDA underestimates the ensemble spread.
  • More reliable and accurate forecasts are obtained with a combination of EDA- and SV-based perturbations (operational since Jun 2010).

Note that: 

  • the perturbations are constructed so that all perturbed members are equally likely.
  • all perturbations are flow-dependent: they are different from day to day.


Once the different sets of SVs have been separately calculated over the northern and southern hemispheres and over the tropics between 30°N and 30°S, they are linearly combined (using coefficients randomly sampled from a Gaussian distribution) and added to the EDA perturbations to make a set of 25 global perturbations.  The signs of these 25 global perturbations are then reversed to obtain another set of 25 “mirrored” global perturbations.  This gives a total of 50 global perturbations for 50 alternative analyses and forecasts.  Consecutive members therefore have, pair-wise anti-symmetric perturbations. The anti-symmetry may, depending on the synoptic situation and the distribution of the perturbations, disappear after one day or so, but can occasionally be noticed 3-4 days into the perturbed forecasts (see Fig5.3 and Fig5.4).


Fig5.1.1: 1000hPa perturbed analyses and forecasts of members 1 and 2 from 12UTC 13 August 2010; the positive and negative perturbations in red and blue dashed lines respectively.  At initial time the perturbations are pair-wise anti symmetric, weakening or deepening a shallow low-pressure system on the westernmost Atlantic (upper images). 24 hours into the forecast, the perturbations in member 1 have led to the low splitting into two cyclonic pressure systems, in member 2 to a significant deepening of the single low pressure system.

 

 

Fig5.1.2:  Same as Fig5.1.1 but for 00UTC 15 August 2010. In this case the anti symmetry is still clearly seen 24 hours into the forecast, member 1 having the low deepened and displaced into a slightly more westerly position, member 2 having the low weakened and displaced into a slightly more easterly position.


Prior to cycle 47R1

Problems with CIN prior to cycle 47R1

Please note that weaknesses were discovered in 2019 in the computation method ECMWF uses for its Convective Inhibition field (CIN). Stored values are often too large, substantially so in some situations. These comments pertain to the diagnostic quantity only, and do not relate to the actual handling of convection within the model. Nonetheless forecasters need to be careful to not be misled into thinking convection cannot trigger when the truth is that triggering could happen. CIN (a standard MARS parameter) can be displayed in map form in ecCharts and is also a component of ecCharts vertical profile plots. It is also a diagnostic that is available from ERA-5.

ECMWF introduced an improved computation of CIN in 2020 as part of cycle 47r1, and also intends to make available several more CAPE and CIN parameters directly from the IFS at a later date.

Note: users should desist from using this parameter (CIN) in any operational run output that has a data time before 12UTC 30th June 2020 (i.e. prior to 47r1).

The problem was rectified with the introduction of cycle 47R1 on 30 June 2020.


Integrated Forecasting System - IFS

  

Fig2.2:  Exchange of physical quantities between the atmospheric, ocean wave and ocean models (before introduction of ice cover information to the wave model in 43R1 in 2018).  All the Atmospheric models give to the Wave model information on air density, ice cover, and surface wind and gusts whilst the Wave model gives to Atmospheric models information on surface roughness (associated with the forecast waves).  Additionally, for the ENS forecast only:

  • the ENS exchanges information with the Ocean model on currents and sea-surface temperature and surface energy fluxes,
  • the ENS gains information from the Ice model on ice cover,
  • the Ocean model gains information from the Wave model on stress, drift, turbulent energy.


Prior to  cycle 45r1 in June 2018 a remotely-generated sea ice cover analysis (OSTIA) was used directly in sea-ice atmosphere assimilation in the surface analyses of the HRES 4D-Var, and in the ensemble of data assimilations (EDA).


Large Scale Precipitation

With the introduction of changes in the modelling of physics processes in cycle 45r1 released in June 2018, previously anomalous forecasts of precipitation in coastal, exposed upslope and rainshadow areas have been significantly reduced. However greater activity is seen with higher precipitation rates in active regions (e.g. in the East Asian monsoon).  Users inspecting model rainfall forecasts for data times before the release of cycle 45r1 should expect somewhat different behavioural characteristics.



EUROSIP - Multi-model Ensemble for Seasonal Products

In Seasonal forecasting the multi-model approach is commonly used to represent the large uncertainties associated with model errors.  The multi-model approach works better when the models contributing have different kinds of errors.  The EUROSIP multi-model system components include the ECMWF seasonal forecast ensemble and four other different ensemble forecasting systems (the UK, French, USA and Japanese systems).

Note: Products from the EUROSIP Multi-model Seasonal Forecasting System have been discontinued with effect from October 2019.

Additional Sources of Information

(Note: In older material there may be references to issues that have subsequently been addressed)


Forecast error statistics for pre-existing tropical cyclones

Effects of resolution on tropical cyclone forecasts.


This bit will go into the "past versions" section in Section12

Fig8.1.10.3.3: An illustration of what can happen when the model resolution is increased.  HRES model performance Aug-Nov 2015 using test runs (~9 km resolution - Red;  ~16 km resolution - Blue), both without ocean coupling. On average, ~9 km resolution forecast location error is slightly better to about Day5 but marginally worse from Day5 to Day7.  However, beyond about Day5 the low sample size makes statistics unreliable and ~9 km resolution is unlikely to be significantly different to ~16km resolution. Users should note that this diagram is included to illustrate that  resolution changes have a significant impact.  Before June 2018 Ocean-atmosphere coupling in HRES was introduced in cycle 45r1 in June 2018 a tendency to over-deepen, which is very apparent here, has been reduced.  New plots will be added to the User Guide once we have a large enough sample.