Alternative low-bandwidth version Skip to main content

Monthly Forecast Explained


Monthly forecasting covers the time range of 10 to 46 days and is really more correctly described as sub-seasonal or extended-range. It is a time scale in between medium-range weather forecasting (10-day) and seasonal forecasting (3 monthly).

Medium-range weather forecasting is mostly an atmospheric initial value problem. Seasonal forecasting, on the other hand, is justified by the long predictability of the atmospheric boundary conditions such as ocean, land or ice and by their impacts on the atmospheric circulation.

The time range of 10 to 46 days is probably short enough that the atmosphere still has a memory of its initial condition and long enough that the ocean variability could have an impact on the atmospheric circulation. Therefore, the ECMWF sub-seasonal forecasts are produced from coupled ocean-atmosphere integrations.

An important source of predictability over Europe in the 10-46 day range originates from the Madden-Julian Oscillation (MJO). The MJO is a 40-50 day tropical oscillation. Several papers (see for instance Woolnough et al. 2003) suggest that the ocean-atmosphere coupling has a significant impact on the speed of propagation of an MJO event in the equatorial Indian and western Pacific oceans. Therefore, the use of a coupled ocean-atmosphere system helps capture some aspects of the MJO variability. Other important sources of extended-range predictability include the stratospheric initial conditions (e.g.  Baldwin and Dunkerton 2001) and the soil moisture initial conditions (e.g. Koster et al. 2010)


There is a reason why our extended-range forecasts are not very detailed and often quite general. This is because forecasts beyond one week become increasingly uncertain due to the chaotic nature of the atmosphere. Small inaccuracies in the weather forecast today can become very large by next week. To try and combat this, a set (or ensemble) of forecasts is produced. This set of forecasts aims to give an indication of the range of possible future states of the atmosphere. A probabilistic forecast is born.

The extended-range probabilistic forecast can often give a trend in the weather over the next several weeks. Is it more likely to be colder or warmer than average? Drier or wetter?


Met Éireann uses data from the ECMWF (European Center for Medium Range Weather Forecasts). The ECMWF ENS/monthly forecasting system is a 51-member ensemble of 46-day coupled ocean-atmosphere integrations. The extension of ENS to 46 days is performed every Thursday and Monday. The first operational real-time monthly forecast at the ECMWF was realized on Thursday, 7 October 2004.

The extension of ENS to 46 days is performed every Thursday and Monday with data available for end-users on Tuesday morning and Friday morning of each week. Over each point of the map, atmospheric variables such as 2-metre temperature, total precipitation, mean sea-level pressure or surface temperature, have been averaged over a weekly period:



  • day 5 to 11
  • day 12 to 18
  • day 19 to 25
  • day 26 to 32
  • day 1 to 7
  • day 8 to 14
  • day 15 to 21
  • day 22 to 28


The weekly means have been averaged over the 51 members of the real-time forecast and the 660 members of the back statistics (11 members x 20 years x 3 forecast runs). The weekly anomaly charts we will display show the difference between the ensemble mean of the real-time forecast and the ensemble mean of the back-statistics or climatology. The graphical products therefore displays the shift of the forecast ensemble mean from the estimated “climatological” mean.

Meteorologists at Met Éireann analyse the forecast data throughout the week and the extended range forecast is updated on Tuesday and Friday evenings.


Monthly forecasts can at times provide an insight into weather patterns in the month ahead. However they should not be used for specific planning purposes as they have generally low skill compared with the 10-day forecast.

Sub-seasonal forecasts from the ECMWF tend to have reasonable skill at both week 1 and week 2 and can highlight trends towards certain weather patterns. At week 3 and 4, there is only marginal skill but the forecast can occasionally indicate trends towards certain types of weather patterns.

Monthly forecast verification statistics and performance (Figure 1) shows the probabilistic performance of the monthly forecast over the extratropical northern hemisphere for summer (JJA, top panels) and winter (DJF, bottom panels) seasons since September 2004 for week 2 (days 12–18, left panels) and week 3+4 (days 19–32 right panels). Forecast skill for week 2 exceeds that of persistence by about 10%, for weeks 3 to 4 (combined) by about 5%. In weeks 3 to 4 (14-day period), summer warm anomalies appear to have slightly higher predictability than winter cold anomalies, although the latter has increased in recent winters (with the exception of 2012).

In 2018, week 2 forecast skill for summer warm anomalies was unusually high, but persistence was also at its highest level within the period shown. The corresponding week 3+4 forecast skill and persistence were also near the upper end of values seen so far. Skill for winter cold anomalies in 2018 was close to average, however in week 2 there is an increasingly consistent margin relative to persistence in recent years.


Fig 2; A comparison of the RMSE and Anomaly Correlation of the monthly forecasting system (MOFC) over Europe against climatology (CLIM) and persistence (PERS). Scores are computed from forecasts on a standard 2.5° x 2.5° grid limited to standard domains. The bottom panel displays the anomaly correlation score of daily monthly forecasts of 500 hPa geopotential height over several regions. Anomaly correlation scores are spatial correlation between the forecast anomaly and the verifying analysis anomaly. The top panel displays the Root Mean Square Errors (RMSE) of daily monthly forecasts of 500 hPa geopotential height over several regions. Root Mean Square Errors are the geographical average of the squared differences between the forecast and the analysis valid for the same time.

  • The MOFC RMSE is lower than the persistence RMSE throughout the forecast period whilst is notably lower than the climatology through until c. day 16, thereafter it is similar to climatology but with a slightly smaller error.
  • The Anomaly Correlation of the MOFC is higher than both climatology and persistence through to Day 30.


A particular case when the sub-seasonal forecast proved very useful was ahead of the severe cold spell in Europe in late February 2018. Below we can see the evolution of the forecast from as early as 1 February and compare with the analysed temperature and pressure anomalies. We can see cold anomalies over Europe began to be heralded in early February. In this instance the forecaster had growing confidence that there was a risk of exceptional cold weather in the region in late February.

(Fig 3; Analysis and ECWMF ENS Forecasting system 2-metre temperature anomaly for week 26-02-2018 to 04-03-2018)

Fig 3; Analysis and ECWMF ENS Forecasting system 2-metre temperature anomaly for week 26-02-2018 to 04-03-2018



Fig 4; Analysis and ECWMF ENS Forecasting system MSLP anomaly for week 26-02-2018 to 04-03-2018

Fig 4; Analysis and ECWMF ENS Forecasting system MSLP anomaly for week 26-02-2018 to 04-03-2018

Fig 3&4: These graphics displays the weekly mean anomalies relative to the past 20 year climate. The first panel corresponds to the anomalies computed using ECMWF operational analysis and reanalysis for a given week. The other panels correspond to the eight monthly forecasts starting one week apart and verifying on that week. The model anomalies are relative to the model climate computed from the model back-statistics. The areas where the ensemble forecast is not significantly different from the ensemble climatology according to WMW-test are blanked. The time range of the forecasts is day 1-7, day 5-11, day 8-14, day 12-18, day 15-21, day 19-25, day 22-28 and day 26-32. This figure gives an idea of how well the predicted anomalies verified against the ECMWF analysis and also about the consistency between the monthly forecasts from one week to another.

You are here: