LFMIP-Pobs

The LS3MIP paper does not contain a detailed description of the seasonal predictability experiments, labeled as LFMIP-Pobs. Draft experimental designs were proposed by Constantin Ardilouze (see attachment) and Bart van den Hurk (see attachment).

The experiment replicates the findings of the GLACE2 predictability study by extending the nr of participating models, period of forecasts, and nr of start dates. As such, the reference simulation is not a transient simulation over the historical record (as is the case for the LFMIP experiments using prescribed land surface conditions), but a set of reforecasts from a seasonal forecast set-up.

For LFMIP-Pobs it is decided to use either the current Copernicus reforecast model data set as a reference forecast, or a new reforecast data set based on the LFMIP model configuration. CMCC, ECMWF and MeteoFrance do participate in this Copernicus program. The model configuration used in these seasonal forecasting programs deviates from the set-up used in the other LFMIP projects.

Experimental set-up

The LFMIP-Pobs experiments duplicate the seasonal reforecasts with the following specifications:

• Forecast period covers at least 1993-2016. Each forecast will be a four-month-long integration.
• 4 start-dates per year are selected (1 Nov, 1 Feb, 1 May, 1 Aug)
• the ensemble size should be between 20 and 25 members
• initial land states are derived from offline generated pseudo-realistic land states, similar to the LMIP set-up
• initial land states are perturbed by shuffling start years (retaining the calendar day) for every member. This set-up is chosen instead of prescribing all members with climatological initial conditions.
• all land variables are updated simultaneously. In upcoming Tier 3 simulations subsets of land states (vegetation, soil hydrology, snow) may be evaluated in isolation.

Data request and Diagnostics

The data request is formed by an overlap between the general LS3MIP data request and the output produced in the Copernicus program. Diagnostics address both (perfect model) potential predictability and real (observation/reanalysis based) predictability.

Participating models

The following modelling groups will contribute to this experiment:

• Meteo France (Constantin Ardilouze)
• CMCC (Stefano Materia)
• ECMWF (Bart van den Hurk; simulations have been done already for 1900-2010)

Further contributions are seeked from ECMWF (Andrea Alessandri) and modelling groups participating in the Copernicus seasonal forecasting service (notably UKMO and DWD).