Introduction

At ECMWF we use the sub-seasonal re-forecasts to build the Sub-seasonal (ex extended) Model Climate (currently still referred as ER-M-climate).

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

• sub-seasonal range forecast verification metrics are based on the re-forecasts.
• re-forecasts allow computation of the ER-M-climate which allows actual forecasts to be converted into an anomaly format. Forecasts in terms of anomalies relative to a model climate (rather than relative to the observed climatology) mean that some calibration for model bias and drift into the products is incorporated.

In IFS Cycle 48r1 upgrade, we increased the frequency of the sub-seasonal (ex extended) range forecast to daily with 100 members. However, the re-forecast data remained twice per week on Monday and Thursday. 

In 49r1, the re-forecast configuration will be change significantly. Instead of production being tied to the day of the week, it will be produced on the specific days of the month.

The main advantages of the new configuration are:

• Increased sub-seasonal range frequency will benefit skill assessment and calibration
• Running fixed days of the month will allow for direct comparisons between re-forecasts produced in different years, and direct comparisons with seasonal re-forecasts.
• Common dates for medium-range and sub-seasonal range re-forecasts provide opportunities for generation of calibrated dual-resolution ensemble products
• Common dates for the two re-forecast sets also facilitate an assessment of the impact of resolution

For sub-seasonal range, we will continue to produce weekly products based on calendar weeks (Monday-Sunday). However, other options will be possible (week starting on any day of the week).

In 49R1, land initial conditions for all re-forecasts (medium and sub-seasonal ranges) are provided by a separate offline land data assimilation reanalysis, instead of using ERA5 data as is, which had been the case for previous model cycles. This change improves the consistency of land initialisation between  re-forecasts and the real-time forecasts which are initialised from the operational analysis, since the offline land data assimilation includes recent land model changes, such as multi-layer snow, which are not present in ERA5. In turn such a change naturally makes the re-forecasts and operational forecasts more consistent. This improves the quality and integrity of forecast products which use the two in tandem, such as the EFI and SOT, and indeed the vast majority of sub-seasonal products, which are presented as anomalies from a reforecast-based climatology. 

Atmospheric and wave re-forecasts change summary

From the Cycle 49r1 atmospheric and wave model are fully coupled which means the re-forecasts are produced with identical configuration

• medium range every four days starting every 1st of the month:  1/5/9/13/17/21/25/29 (excluding 29 February)
• extended range every two days starting every 1st of the month: 1/3/5/7/9/11/13/15/17/19/21/23/25/27/29/31 (excluding 29 February)

* There are no additional products from ENS-WAM Extended hindcast statistics (meaning nothing is produced in the stream . 

Sub-seasonal range re-forecast configuration

Sub-seasonal (ex-extended) range re-forecasts will be produced every other odd day of the month: 1/5/9/13/17/21/25/29 (excluding 29 February).

Mars streams in the Atmospheric model  - Dissemination data stream indicator H :

• eefh - Ensemble forecast hindcast
• eehs  - Ensemble forecast hindcast statistics

Mars streams in the Wave model - Dissemination data stream indicator Z:

• weeh - Wave ensemble forecast hindcast 

Building a model climate

We will show here how to build the sub-seasonal Model Climate (ER-M-climate) using the re-forecast data from IFS 49r1 cycle.

What data do I need?

The set of re-forecasts for the ER-M-climate is made up from:

• a set of re-forecasts using the same calendar start dates for each of the last 20 years - we refer to these as Re-forecast date
• five consecutive re-forecasts (covering a 9 days period), the middle one of which corresponds to the preceding odd day of the month that is closest to the actual ensemble run date.
• each re-forecast is formed from an 11-member ensemble (1 control and 10 perturbed members) run over the 46-day ensemble forecast period.

In total, each set of re-forecasts consists of 20 years x 5 runs x 11 ensemble members = 1100 re-forecast values.  
These are available for each forecast parameter and forecast lead-time at forecast intervals of 6 hours.  

These are used to define the ER-M-climate.

Which dates do I need to build ER-M-climate?

Data will be disseminated sufficiently far enough ahead (more on that later) for you to build a model climate with the following configuration.

For sub-seasonal re-forecasts, the model climate can be built from 5 re-forecasts (1 central + 2 ahead + 2 behind) – annotated by the red dashed line.

On even days, it’s not possible to use a centralised re-forecast so we recommend using the previous re-forecast as the ‘central’ and then 2 either side.

You can use the same model re-forecast for 2 days in a row EXCEPT at the end of a 31-day month, when we have two odd days in a row (31st and 1st).

However be careful with using the weekly steps, as you will need to use different steps on different days (to match with the corresponding forecast)

And what should I do with these dates?

ER-M-climate consists of weekly means over all 11 ensemble members and 20 years of re-forecast.

In order to compare the ER-M-climate and the forecast their steps must match. This is where it gets slightly complicated.

Since we produce the forecast every day and the end product is weekly mean where week starts on the following Monday, every day we use different steps. You can see the steps on the right → → → 

However, since we produce the re-forecast every other day, in order to build the ER-M-climate you can not use the same set of steps on the two consecutive days.

For example:

If today is Tuesday, 17th September, the forecast will have the steps 144-312/312-480 etc. So the re-forecasts used to build the ER-M-climate must use those steps.
Tomorrow is Wednesday, 18th September and we didn't produce new re-forecasts, so we use the set from the day before. But we need to use the steps from Wednesday, 120-228/228-456 etc.

Easy, so far...

Every sub-seasonal re-forecast date is being used 5 times, so in order to be able to build the ER-M-climate users need to order all the steps, because every day different steps will be used.

We are preparing a Jupyter notebook to show you how you can calculate the ER-M-climate using the re-forecast data available in the Test RCP server.

Installing your data in the Product Requirements Editor

Primary Recommendation (easiest option)

The primary recommendation is that if you receive daily Sub-seasonal Forecasts, you should also receive all the re-forecasts. 

To install this, you must apply the following to every dissemination block where you have used stream=eefh: 

Twice-weekly Recommendation 

In 48r1, it was possible to receive data twice a week on fixed days (Monday and Thursday).  In 49r1, as we produce data every other day, this will not be possible (7 does not divide by 2). 

With the current keywords, the recommendation here is to apply the following to every block where you have used stream=eefh:

With this, you will receive data on those days when they are odd. Note that the frequency is not equal per month.  An example of the days you'd receive data for 2025 is shown below: 

Note that this figure is provided as an example to demonstrate the impact of trying to receive twice-weekly reforecasts. 

Once-weekly Recommendation

As per twice-weekly, to have a re-forecast approximately once per week, you would need to apply the following line to every dissemination block where stream=eefh: