Creation of M-Climate

The M-Climate is derived from a set of medium range re-forecasts.  These are created using the same calendar start dates over several years for data times either side of the time of the ensemble run itself.  The re-forecast runs are at the same resolution as the medium range ensemble (currently 9km) and run over the 15-day medium range ensemble period.   

There is merit in examining the real-time performance of a forecasting system.   But the sample sizes created for one system are far too small to conclude anything about its true performance levels.  Re-forecasts are used to increase the available data to produce a model climate.   The results of forecast system may be compared with this model climate.  

Re-forecasts are a fundamental component of all seasonal forecasting system; they have two applications:

  • medium range forecast verification metrics are based on the re-forecasts
  • re-forecasts allow computation of the M-climate.

Selection of medium range re-forecasts

The set of re-forecasts is made up from:

  • a set of re-forecasts using the same calendar start dates for each of the last 20 years.
  • nine consecutive re-forecasts (covering a 5-week period).  The middle one corresponds to the preceding Monday or Thursday that is closest to the actual ensemble run date.
  • each re-forecast is from an 11-member ensemble (1 control and 10 perturbed members) run over the 15-day ensemble forecast period.

In total, each set of re-forecasts consists of 20 years x 9 runs x 11 ensemble members = 1980 re-forecast values.  These are available for each forecast parameter, forecast lead-time, calendar start date, location, at forecast intervals of 6 hours. 

The M-climate is used in association with the ensemble forecast:

  • to present the 15-day ensemble meteograms with the medium range climate (M-climate)
  • to deliver the extreme forecast index (EFI) and shift of tails (SOT) products 
  • to highlight significant forecast departures of 2m temperature, wind speed, cloudiness and precipitation from the norm for a given location and time of year.  

Values evaluated in M-climate

  • 2m temperature.
  • soil temperature.
  • sea-surface temperature.
  • mean sea level pressure.
  • precipitation.
  • cloudiness. 

The same M-climate set is used for 00UTC and 12UTC ensemble runs.  This is to avoid inconsistencies between the validity period of the ensemble and M-climate.  So, for example:

  • Day1 M-climate is used with T+12h to 36h forecasts from the 12UTC run and T+0h to 24h forecasts from the 00UTC run.
  • Day2 M-climate is used with T+36h to 60h forecasts from the 12UTC run and T+24h to 48h forecasts from the 00UTC run.

Limitation of twice weekly updates to M-climate

M-climate is updated twice per weaken Mondays and Thursdays.  So M-climate quantile plots for the same verifying time from two forecast runs straddling the M-climate update will be slightly different.  So, for example, the quantile plots for forecast runs on 00UTC Thursday and 00UTC Friday will not be exactly the same.  However, this limitation of twice weekly updates to the M-climate can be significant.  It can be particularly evident in spring and autumn when mean temperatures are changing most rapidly day by day.

Different reference periods for M-Climate and ER-M-Climate

ECMWF uses different reference periods but essentially the same re-forecast runs to build the M-climate and the ER-M-climate.   The key difference is that those runs are grouped and used in different ways:  

  • For shorter ranges, the priority is the best possible capture of the climatological distribution of the tails (e.g. for extreme forecast index (EFI) and shift of tails (SOT)).  This can be better achieved using a re-forecast span of 5 weeks (1980 re-forecast values).  
  • For longer ranges, the priority is the correct representation of seasonal cycles.  This can be better achieved by using a span of 1 week (660 re-forecast values).  The tails should not be so prone to having a reduced sample size.


Note before Cy41r1 in spring 2015, the M-climate was constructed from only 500 re-forecasts was more prone to sampling errors and as a result.