wiki:DevelopmentActivities/OasisDriver

Version 9 (modified by jpolcher, 7 years ago) (diff)

--

Introduction

Today ORCHIDEE is designed to be called by a subroutine independently on each processor. This is implemented for instance in the traditional driver (dim2_driver.f90) :

No image "Witout_OASIS.png" attached to OasisDriver

This assumes that the domain decomposition of the atmospheric model or the driver are the same. It also means that the processors which have no land point to deal with are waiting for the others to finish.

In order to avoid these issues and generalise the coupling of ORCHIDEE with other atmospheric components, the OASIS-MCT coupler can be used. In this case the atmosphere can work on any number of processors and ORCHIDEE on a distinct set of processors. This is exemplified in the driver2oasis presented below. Here, the driver (driver2oasis.f90) which mimics the atmosphere, is a single processor code which just reads some forcing files and provides that data to OASIS. It is a prototype for any atmospheric model which would send the same fields from its n processors to OASIS.

No image "With_OASIS.png" attached to OasisDriver

In the OASIS approach to coupling a number of advantages are gained :

  • The atmosphere can run on a different set of processors than the land surface and have its own domain decomposition.
  • ORCHIDEE will only execute on the processors attributed to it and only treat land surface points.
  • After the atmosphere has sent its data to OASIS it can continue to work on processes which do not require the input of the land surface.
  • In the same way, once the land surface has returned the variables needed by the atmosphere (i.e. after solving the surface energy balance) it can continue to work and prepare the next time step.
  • The land surface model is an independent executable which has its own restart mechanism and history system and can thus be developed independently from the atmospheric component.

The driver for OASIS : driver2oasis.f90

This code simulates an atmospheric model which would provide data for ORCHIDEE through OASIS-MCT.

As the original ORCHIDEE driver (dim2_driver.f90) is outdated and inflexible it has been redeveloped here.

For the moment this code only reads compressed by gathering forcing files which contain a sufficient amount of meta-data (Description_Forcing_Files.pdf).

This code is aimed at running only on one processor. It should not be a limiting factor in the execution time as it only reads once the forcing and then only does some simple time interpolations. Furthermore these interpolations are done while ORCHIDEE runs, i.e. it is not blocked by an oasis_get call.

The fields which OASIS exchanges between the atmosphere and the surface

From the atmosphere to ORCHIDEE

Name units ID on driver side ID on ORCHIDEE side
Height of T and Q m ZLTQDRIV HEIGHTTQ
Height of wind m ZLUVDRIV HEIGHTUV
Atmospheric temperature at lowest atmsophric level K TAIRDRIV TEMPLEV1
Atmospheric moisture at lowest atmsophric level kg/kg QAIRDRIV HUMILEV1
Rainfall kg/m2/dt RAINDRIV RAINFALL
Snowfall kg/m2/dt SNOWDRIV SNOWFALL
Downward solar radiation W/m2 SWDODRIV SHOWDOWN
Downward longwave radiation W/m2 LWDODRIV LONWDOWN
Solar angle ?? SOLADRIV SOLARANG
Eastward atmsopheric wind m/s UWINDRIV EASTWIND
Northward wind m/s VWINDRIV NORTWIND
Surface pressure Pa PRESDRIV SURFPRES
Surface drag s/m

From the surface to the atmosphere

Name units ID on driver side ID on ORCHIDEE side
Surface evaporation kg/m2/dt EVAPDRIV TOTEVAPS
Sensible heat flux W/m2 SENSDRIV FLUXSENS
Latent heat flux W/m2 LATEDRIV FLUXLATE
Diffuse freshwater flux to the oceans m3 /dt COASDRIV COASTFLO
River flux into the ocean m3/dt RIVEDRIV RIVERFLO
Net CO2 flux ?? NECODRIV FLUNECO2
?? ?? LUCODRIV FLULUCO2
Surface radiative temperature K TRADDRIV TSURFRAD
Surface temperature K TNEWDRIV TSURFNEW
Surface humidity kg/kg QSURDRIV QSURFNEW
Near Infra-red albedo - ANIRDRIV ALBEDNIR
Visible albedo - AVISDRIV ALBEDVIS
Surface emissivity - EMISDRIV EMISLONW
Surface roughness m ROUGDRIV ROUGNESS

dt : is the time step used by the models. We do not expect ORCHIDEE and the atmospheric model to use different time steps.

The "main" which controls the ORCHIDEE execution : orchideeoasis.f90

The main, in the FORTRAN sense is very similar to the intersurf.f90 which as been used up to now. The main difference is that it obtains the atmospheric data by issuing the needed OASIS_get calls. It will perform the following actions :

  • Set-up the domain decomposition of the grid it has been given.
  • Set-up the restart and history mechanisms.
  • Start the OASIS exchanges and obtain the atmospheric data.
  • Call sechiba_main
  • Send to the atmosphere the flux computed
  • Write to the history files
  • return to the next call to OASIS in order to obtain the atmospheric data for the next time step.

Configuration of this ORCHIDEE version

Added information in the run.def

The grid description file

The namecouple

Execution of the model coupled to the driver

As OASIS is linked into both models (driver2oasis and orchideeoasis) the execution of the coupled model is simply achieved by one mpirun command. It can have the following structure :

mpirun -n $PEDRIV driver2oasis : -n $PEORCH orchideeoasis

where : PEDRIV=1 and PEORCH=$(($NSLOTS-$PEDRIV)), i.e. only one processor is for the driver and all others work on ORCHIDEE.

Execution of the model coupled to WRF

Attachments (2)

Download all attachments as: .zip