Changes & Updates

Scientific Model Upgrades

The Met Office is upgrading its operational modelling systems to Parallel Suite 47 (PS47), scheduled to go live on Wednesday 21 January 2026 when it will be switched over to become Operational Suite 47 (OS47).

The first OS47 models will come at:

• Global Deterministic and post-processing 12z - expected time 16:30 - 16:50
• UKV and post-processing 10z (delayed) - expected time 11:30 - 11:50 (delayed 25 minutes)
• Global Ensemble and post-processing - 12z - expected time 18:20 - 18:55
• MOGREPS-UK and post-processing 09z (delayed) - expected time 12:00 - 12:30 (delayed 50 minutes)
• Met Office Blended Probabilistic Forecast - 22:30z

Parallel Suite 47 (PS47) is the first significant science upgrade to our operational systems since PS45, over 3 years ago. Consequently, there are major upgrades to many of our primary models. In particular, science upgrades to our global and regional model systems (atmosphere and coupled NWP) resulting in changes to the model systematic errors which are likely to be noticeable to users. For our 'big 4' models these include:

• UK cloud forecasts (including fog) will be more accurate, which is likely to benefit, for example, aviation.
• UK winter temperature forecasts are improved benefiting, for example, the energy industry, gritting/de-icing, etc. However, summer cold bias is slightly exacerbated.
• Precipitation forecasts for the UK will look more realistic (less 'blobby'), and hence usable, by our operational meteorologists and the public.
• Improved predictions of tropical cyclone intensity, helping to save lives and livelihoods around the world, including overseas UK nationals. As the global model now has improved tropical cyclone intensities, we are retiring the tropical cyclone model which was introduced at PS45 as a 'stop-gap' solution.

PS47 marks the operational implementation of the first elements developed under the Next Generation Modelling System Programme. Specifically, our operational verification will adopt the internationally developed MET+ infrastructure, and JOPA (JEDI-based Observational Processing Application) will be adopted for the observational data processing for assimilation into our global models (coupled and marine-only).

These changes are designed to improve forecast accuracy and reliability. As a result, we do not expect significant impacts on users. However, it is likely that some of these changes will require a degree of intervention prior to our release to prevent any impacts to usage of our data.

Weather DataHub will not be releasing all PS47-related changes at this stage. However, over the coming months we will be introducing additional updates, including an extended forecast range for our atmospheric models, as well as new level types, timesteps and parameters.

We will share further information, including timescales and technical details, on the Changes & Updates page as these developments become available.

Atmospheric:

• Global Deterministic 10km timesteps: Timesteps 51 and 54 will be removed for the following parameters:
  • Precipitation accumulation (3 hour)
  • Rainfall accumulation from convention (3 hour)
  • Snowfall accumulation from convection (3 hour)
  If you still require data for timesteps 51 and 54 for these parameters, the following parameters now include timesteps 1-54:
  • Precipitation accumulation (1 hour)
  • Rainfall accumulation from convection (1 hour)
  • Snowfall accumulation from convection (1 hour)

Top Level Summary - Global NWP Models

The main components of PS47 global package include an upgrade of the global model configuration to GC5, a replacement of the global observation processing (OPS) to the NGMS Jedi-based JOPA code, addition of the new global Mode-S aircraft observations and ocean data assimilation. We expect these changes to improve our deterministic and ensemble verification metrics on average by 1% and 2% respectively.

Improvements to the model initialisation

• Major technical (but science-neutral) change to the software used to extract and process observations used in atmospheric, no impacts should be evident in the forecast performance.
• Addition of EMADDC (European) Mode-S aircraft temperature observations and the newly available global Mode-S aircraft observations. Good positive impact on forecast scores in the northern hemisphere and tropics, particularly at lead times up to 3-days.

Improvements to the model formulation

• GC5 is the latest global coupled configuration which is a combination of Global Atmosphere and Land (GAL9) and Global Ocean and Sea Ice (GOSI9).
• GAL9 includes numerous changes to almost all areas of global model science, including several enhancements to the 6A convection scheme, latent heating in the gravity wave drag scheme, 'fountain buster' scheme which adds improvements to advection conservation in conditions of near-grid-scale convergence, bimodal cloud initiation and dust dependent ice-nucleation temperature.

Impact on forecast skill

• Overall positive change in GC5 verification scores relative to GC4 with ensemble CRPS improvements greater than the deterministic RMSE scores.
• Most parameters improve across all lead-times, with improvements in global temperature and winds, subtropical jet, monsoon processes over India and Maritime Continent and their diurnal cycle in convection.
• Significant improvements in the tropical temperature at 850hPa and precipitation SEEPs score at all lead times.
• A few exceptions include tropical near-surface winds, which have higher errors that are caused by stronger windspeeds.
• Also, some mixed results for the southern hemisphere and Europe in some of the seasons tested, but nothing persistent.
• Although ensemble verification CRPS has improved widely there is also a small overall reduction in spread. This is partially expected due to lower forecast errors, but to address any under-dispersive issues, a re-tuning of the Stochastic Perturbation of Tendencies scheme (SPT) has been done to partially mitigated any spread loss. This was possible due to greatly improved model stability from the model orography filtering work.

Top Level Summary - UK NWP Models

The regional UK updates for PS47 include a significant change in the model formulation with the RAL3 package and the use of different lateral boundary conditions and ensemble perturbations with the new GC5 global implementation.

Improvements to the model initialisation

PS47 includes important technical and scientific changes to the model initialisation and includes:

• the removal of two main data assimilation methods (the latent heat nudging and the adaptive vertical grid),
• improvements in quality control for visibility,
• improvements to the existing radar reflectivity data assimilation
• and technical upgrades to prepare the move to JEDI-based Observation Processing Application (JOPA) at PS48.

Improvements to the model formulation

• The RAL3 package in PS47 introduces a new double-moment microphysics scheme, CASIM, as well as a new bi-modal cloud scheme.
• Further RAL3 updates include the boundary-layer scheme as well as land surface updates which brings this regional configuration more in line with the global land atmosphere settings.
• As a result of these changes, it has been possible to unify mid-latitude and tropical configurations.

During the R2O PS47 trialling and evaluation process, we further tested and developed the RAL3 configuration in collaboration with APP. This resulted in:

• maintaining a minimum amount of spin-up at the edges of the domain by adding stochastic boundary-layer perturbations near the boundaries.
• new land-surface ancillaries to include C4 grass and improvements to reduce the occurrence of extreme Leaf Area Index values.
• improvements for extensive cover of low cloud base heights in high profile regimes, such as Easterlies.

Impact on forecast skill

• PS47 delivers improvements in forecast skill for temperature, cloud cover, cloud base heights and winds as measured by Ranked Probability Scores.
• PS47 performance for precipitation shows small detriments in Ranked Probability Scores but improvements in spatial verification metrics, such as Fractions Skill Scores.
• Model biases are generally improved (PS47 is cloudier and shows slower winds) but temperature bias performance differs by season, with the summer daytime diurnal cycle being slightly worse with PS47. The winter temperature bias however is much reduced with PS47, together with the cold temperature lead time drift mostly noticeable in MOGREPS-UK.
• The scientific summary report also highlights the variability of PS47 performance with regimes and time series, and particularly the variability in cloud cover biases and the presence of cloud breaks in the winter in anticyclonic regimes.
• Differences in the PS47 performance between the UKV and MOGREPS-UK, particularly for visibility, are noted. While PS47 performance in visibility is positive for the UKV at all lead times, MOGREPS-UK PS47 performance in visibility is only positive at short lead times, due to the changes in the visibility quality control for data assimilation; the negative performance for PS47 for the ensemble at longer lead times is mostly associated with the higher visibility thresholds, seen as less important to the users. As the ensemble verification appears more consistent across its different metrics, we conclude that the ensemble scores are more suitable for representing probability distributions and therefore should be more trusted.
• Ensemble spread performance is neutral overall and slightly worse for summer temperature (with opposite impact from RAL3 and GC5).

The subjective evaluation does not raise any strong and specific concern to the stakeholders. Overall, the PS47 performance is preferred, but details vary. The main points to note are the generally better cloud base heights and the improved modelling of light rain in fronts. Model characteristics will be updated and disseminated.

Technical Impact

The introduction of CASIM means the cloud ice water content is now calculated differently. This was previously output as Stash 0-012 (QCF AFTER TIMESTEP). What this now contains is different and users requiring cloud ice water content should use the sum of Stash 0-012 (QCF AFTER TIMESTEP) and Stash 0-271 (CLOUD ICE (CRYSTALS) AFTER TIMESTEP). This will be handled in StaGE, so there will be no impact on those taking data from Weather DataHub.

Standard Gridding Engine (StaGE) Suite

Top Level Summary

This is a very significant release for StaGE, the first in over 2 years, and will be relevant to users of our gridded product streams.

Atmospheric data (Global, UKV, MOG-G, MOG-UK) pulled through Service Hub

Level of impact will vary significantly depending on exactly what data they take and if from object/coverage service/a middling system etc.

Data offering enhancements

Several enhancements have been made to the output of each atmospheric model, including new parameters, vertical levels, and timesteps. These have been made to meet user requirements and improve consistency; increasing the pull-through of supercomputing outputs, and providing an overall richer, more comprehensive offering of data.

It's important to note that the impact of these data offering enhancements will vary across different customer segments. Depending on the specific systems and data sources in use, some of you may benefit from these improvements sooner or in greater detail than others. However, every effort is being made to extend these enhancements as widely as possible, so that all users benefit from the comprehensive data enhancements to our atmospheric model.

Reduced precisions, smaller files

Unnecessary precision has been reduced for many parameters, keeping accuracy where it matters while cutting file sizes and improving data handling efficiency.

Improved humidity data

Calculations of relative humidity have been updated to improve consistency between surface and upper-atmosphere data, and to remove unrealistic values. The new processing chain is more aligned with the model approach, and has reduced the extent of unrealistic super-saturation, though this is still occasionally observed in very cold areas and further investigations are ongoing. The opportunity is also being taken to expose specific humidity in all models, in addition to the existing relative humidity; combined, these provide a fuller picture of atmospheric moisture for a wider range of users.

Improved vertical data quality

Data through the atmosphere (multi-level data) is now processed more accurately and efficiently, using improved interpolation methods and better alignment with model processes. This enhances the quality of parameters such as temperature, wind, and geopotential height 'on pressure levels'. It also increases flexibility in what multi-level data can be provided, without a dependency on upstream NWP model updates.

Improved data quality around coastlines We have updated the way that parameters 'over land' and 'over sea' are defined, processed and blended. This means that the differences in data along coastal boundaries are better preserved. The Global and MOGREPS-G “land sea mask” has been updated with a more accurate representation of the model coastline (giving a more realistic land-sea boundary).

Soil moisture improvement

Following feedback, the generation of soil moisture data has been updated to avoid unrealistic saturation levels, improving land-surface modelling.

Improved consistency of radiation data

A more appropriate interpolation method is now applied to all radiation parameters, leading to more accurate and representative values.

The generation of 'total downward shortwave radiation flux at surface' (the amount of shortwave radiation reaching the Earth's surface) has been made consistent across all models, and is now more representative of 'instantaneous' values rather than an 'average' over a period of time. This addresses the observed differences in the values of this parameter between different models, particularly around sunset.

Updated cloud ice mixing ratios in UK models

Following an update to model physics, the calculation cloud ice data in UK models has been updated to maintain compatibility with existing outputs. This protects users from a breaking change.

Temporary fix to erroneous UKV canopy water values

Erroneous negative values of canopy water parameter in the UKV have been discovered (these should be zero or positive). A temporary fix ensures that all negative value outputs are now set to zero. Further investigations into the cause are ongoing.

Impact on forecast skill

The changes to horizontal and vertical regridding described above, combined with the updated generation of certain parameters (including separate land/sea values) should provide improvements to the way StaGE reflects the model representation of a range of parameters. In particular: at the surface along coastal boundaries, through the vertical atmosphere, and, in the case of soil moisture, on soil depths. The humidity work is also expected to better represent the model humidity, and address problems spotted by users.

Model availability

• We're making improvements to our data processing, which may result in slightly longer processing times. As a result, there could be a delay in model availability when PS47 transitions to OS47 on 21 January and thereafter.
• For example, for the global deterministic model, we are expecting data to be available around 10-20mins later than OS46, possibly varying in this range over the lifetime of OS47. We also expect users of our gridded products to experience similar changes in data timeliness.

Met Office Blended Probabilistic Forecast:

We're introducing new and improved daily summary diagnostics for the medium-range forecast period (days 8-14). These updates will provide richer insights and help you plan further ahead with greater confidence.

Key enhancements:

• Forecast Range: New daily summary diagnostics have been introduced to support extended-range forecasting.
• Parameter Updates:

Parameter	Detail
24-hour wind gust maximum	We currently provide wind gust maxima for 1-hour and 3-hour periods. We're adding a new 24-hour maximum wind gust diagnostic for medium-range forecasts (days 1-14). This will give you a clearer view of the strongest gusts expected over an entire day. For the UK, this will be available as a single daily value (00Z-00Z). For the global domain, data will be provided every 3 hours to support different time zones.
Typical daily wind speed	We're introducing a new diagnostic that shows the typical wind speed over a 24-hour period for medium-range forecasts (days 1-14). This is calculated using percentiles to represent a range of possible conditions, not just a single value. The 50th percentile gives an idea of the wind speed you might typically experience during the day, while higher and lower percentiles show potential variation. Percentiles available: 5, 10, 15, 20, 25, 30, 40, 50, 60, 70, 75, 80, 85, 90, 95. This helps provide a more complete picture of daily wind conditions rather than relying on instantaneous values.
Sunshine duration	We're adding a new diagnostic to show how bright a day is likely to be. At this stage, it's intended only for technical prototyping and should not be used in broader applications. We plan to refine it further to make it more accurate and useful. Provided as a 24-hour value, though actual sunshine hours depend on daylight and weather conditions. Percentiles are included to show a range of possible sunshine hours.
Precipitation fraction (fraction of the day that is wet)	We're introducing a new diagnostic that shows the fraction of the day that is wet. Calculated from shorter periods (3-hour blocks) to estimate how much of a 24-hour day is likely to have precipitation. Percentiles are provided to show driest, most likely, and wettest scenarios for the day. This helps give a clearer picture of daily conditions in the medium range (days 8-14), especially when detailed weather symbols aren't available.
Probability of precipitation (POP)	We're adding a new diagnostic that shows the chance of any precipitation during the day. Calculated from the precipitation fraction diagnostic to ensure consistency. This approach avoids misleading results (e.g., a single shower giving a 100% POP for the whole day). POP represents the probability that some part of the day will be wet, giving a more realistic view of daily conditions in the medium range (days 8-14).

• More detail: Additional percentiles for better representation of uncertainty, 25^th & 75^th percentiles.
• UK & Global Coverage: Outputs available for both domains.
• Data Availability: Data will now be available earlier. For example, the 06z blend will appear around 07:00 UTC instead of 11:00 UTC.

When?

These enhancements will go live between 20 January and 4 February 2026.

Action required:

These improvements will provide a clearer view of conditions in the second forecast week, helping to support better decision-making. For the above, no action is required, and the updates will be available from the release date, 20 January 2026.

However, there is a small change to the parameter ID for freezing rain accumulation. If you are utilising this parameter, you will need to update your systems to consume the new IDs for these: