Spinup : why, how and how long? The answer depends on the ORCHIDEE component

Author: P. Peylin, F. Maignan
Last revision: 2020/04/01, A. Ducharne

Principle

Spinup of the model is needed to find a stable initial state for historical simulations (or any simulation for which the initial conditions cannot be deduced from observations).

The principle is to start from a priori initial conditions, and run the model for a certain amount of time with boundary conditions (weather, land cover, CO2, etc.) which are representative of the period preceding the initialization of the simulation. We often speak of:

• warm up if we create initial conditions by a simulation using the preceding meteorological forcing (for example 1960-1979 to initialize a simulation starting on Jan 1st 1980),
• spin up if we recycle a meteorological forcing to get a longer preceding period (for instance, to start Jan 1st 1980 after a 20-yr spinup, you can use 20 times 1979, or 5 times 1976-1979, depending on the availability of the meteorological forcing).

A priori, the closer you are to a warm up, the more realistic your initial condictions, but it's not always possible, especially if you need a long initialization period.

The required length of the initialization period (either spin up or warm up) depends on the components of ORCHIDEE you are interested in. The goal is to bring the corresponding stocks to an equilibrium with the mean climate of the period, and the required period is all the longer as the corresponding residence time are larger:

• The equilibrium of the physics of the land surface (SECHIBA part of ORCHIDEE, for water and heat stocks) should be reached in less than 10 years with a constant climatology and a prescribed LAI. If you spin up, pay attention to the selected years because El Niño or La Niña specific years may give bad results for a global spinup. The spin up can also lead to accumulate ou melt the snow cover in an unrealistic way if the recycled period is short and includes an extreme year in some regions.
• The equilibrium of the LAI (seasonal variation of the vegetation) takes less than two decades to converge.
• The carbon stocks and fluxes take a much longer time to reach an equilibrium when starting from zero.
• Vegetation C stocks/fluxes: The spinup needs to bring the vegetation above and below ground carbon stocks at equilibrium. This is relatively short for the LAI: few decades or even just one decade are enough. For the other C pools it is linked to the turnover of the Sap/Heart/wood C pools: usually 100 to 200 years are sufficient.
• Soil C stocks/fluxes: These pools take a much longer time to be in equilibrium as the passive C pool has a turnover on the order of one thousand years. Usually we need several thousand years to reach the equilibrium. Using the in-built iterative pseudo-analytical spinup configuration reduces this step as we then only need to have the input of C to the soil in equilibrium.

Finally, the length of the initialization period should be dictated by the slower active component influencing your simulations:

• If you use ORCHIDEE with STOMATE, use a 20-yr initialization even if you are not interested in LAI, because there is a big feedback between the LAI and the hydrology and energy budgets. And if you're interested in Sap/Heart/wood C pools, use a 100 to 200 period.
• If you're interested in the soil carbon, you need a longer initialization. But it is not required for the moment if you're not interested by the soil carbon, since there is no feedback between the soil carbon pools and the LAI and the hydrology in the default sept. Nevertheless, if OK_SOIL_CARBON_DISCRETIZATION is set to TRUE then some feedbacks are activated and a full spinup is necessary whaterver the variable you are interested in

Standard protocol for Carbon-related variables

A standard protocol is provided by the TRENDY model intercomparison project. The protocol comprises:

• 1) A long spin up simulation equivalent to several thousands of years should be done, recycling the meteorological forcing (usually cycling over the first 10 years of available forcing files) and choosing the atmospheric CO2 concentration of pre-industrial conditions. This brings the above and below ground C pools into equilibrium. It should be done either with:
• The iterative pseudo-analytical spinup procedure (see below): in this case a relatively shorter simulation should first be done (around 200 to 300 years) in order to bring the above ground C pools and fluxes into equilibrium. This provides an equilibrium for the input of C to the soil; the iterative pseudo-analytical spinup configuration SPINUP_ANALYTIC_FG1 then allows to bring the soil carbon pools to equilibrium in a few matrix inversions, given that the soil carbon dynamic follows first order kinetic equations.
• The old SPINUP configuration using stomate and forcesoil (see below)
• A simple model simulation for 2-3 thousands years, cycling several years of meteorological forcing files (see below)

This first step leads to an equilibrium of all C reservoirs with the chosen climate and atmospheric CO2 concentration.

• 2) A first transient simulation without real climate is usually done still cycling the meteorological forcing files, but starting increasing the atmospheric CO2 from pre-industrial time and changing the yearly PFT maps, to the start of the period that is investigated (usually the beginning of the twentieth century) in order to have to take into account rising CO2 and land use change on the gross and net carbon fluxes and the C pools.
• 3) A second transient simulation with real climate is then done, following up the first transient simulation, using besides this time real years of the forcing files (no more cycling).

The built-in iterative pseudo-analytical spinup configuration

There is a dedicated experiment in directory SPINUP_ANALYTIC_FG1 found in config/ORCHIDEE_OL. The analytic spinup is activated through "SPINUP_ANALYTIC=y" in stomate.card. The spinup.driver will then calculate and set spinup_period in run.def using CyclicBegin? and CyclicEnd?. Note that if this spinup_period is too small it will be adjusted to a higher multiple (close to ten years) automatically by the model due to stability reasons.

• The EPS_CARBON parameter is used to automatically stops the spinup when all observed litter and soil carbon pools have a relative variation (deltaC/C) lower than EPS_CARBON. Its default value is 0.01%.
• For the TRENDY project, all pools had a relative variation lower than 0.05% after 340 years.
• When the spinup converges before the maximum number of iterations is used, LibIGCM will stop the spinup. At that time the run.card will show Fatal despite the fact that everything went well. A ticket was opened on this issue (#161) but note that it does NOT affect the results.

How to loop over years with libIGCM

To run a short initialization for "quick" varaiables (water, heat, and LAI), you can either adapt the SPINUP_ANALYTIC_FG1 to fit your boundary conditions and selected years (total years and recycled years), or define these information yourself. More about How to loop over years with libIGCM.

The old/deprecate SPINUP configuration

The method relies on the SPINUP configuration, which uses the programs forcesoil and orchidee_ol to accelerate the spinup (forcesoil was used to accelerate the convergence of the soil carbon pools only). The directory SPINUP is used to spinup with this method, read more about the method in SPINUP configuration. This configuration is needed if you activate OK_SOIL_CARBON_DISCRETIZATION. The directory SPINUP is used to spinup with this method, read more about the method in SPINUP configuration.

IMPORTANT: SPINUP directory is not maintained.