Relocatable weather forecasting

CCAM can be used to forecast the weather and downscale the forecast over a region.  This process is based on four stages that are summarised below.

Forecast initialisation

The forecast initial conditions are crucial to determining the skill of the simulation.  CCAM does not support a data assimilation system, so the initial conditions are based on analyses from various centres.  Typical analyses used by CCAM are from NCEP GFS and Australia BoM ACCESS, although CCAM has also been initialised using analyses from CMC or ECMWF.  To initialise a forecast CCAM requires the following fields

• 3D fields – Zonal winds (u), Meridional winds (v), Air temperature (ta), water vapour mixing ratio (mix_rto) or relative humidity (rh) and geopotential height (hgt)
• 2D fields – Surface temperature (tss) for sea surface temperatures, surface geopotential height (zs), land-sea mask with sea=0 and land>0 (land) and surface pressure (ps)
• Optional fields – Mean sea level pressure (mslp), sea ice cover fraction (fracice), terrestrial snow depth (snod)

Vertical levels are usually pressure levels, but can be sigma levels.  Further details on the input file format can be found under the following link CDFVIDAR – Process Lat/Lon Input to Cubic

Lower resolution forecast

Since CCAM is a global model, the above initial conditions are sufficient to general a weather forecast.  Often this forecast is generated using a global uniform (Schmidt=1) simulation so that forecast errors are similar for different parts of the grid.  However, it is possible to use forecasts with a limited amount of grid stretching, Schmidt<=3 (i.e., CCAM code parameter schmidt>=0.3).  Higher levels of stretching run the risk of distorting the skill of the forecast.  The limitation of reduced stretching for the forecast can increase the computational expense of the simulation, so often a lower resolution forecast is used and the downscaled to higher resolution (see below).  Since the scale-selective filters used for nudging are typically operating at scales of 30deg for the nesting within the CCAM forecast, the lower resolution forecast only needs to represent the large scale synoptic situation.

There are a number of ways that soil data and aerosols can be initialised.  A soil climatology can be attached to analyses using smclim.  Alternatively, soil data from a previous forecast can be ‘recycled’ by specifying the old forecast as an input to CCAM (i.e., to override the existing soil data).  Finally users can spin-up the soil data by running a spin-up forecast that is nudged towards analyses (e.g., for the previous 24 hours).

Higher resolution downscaling

Once a forecast has been generated, it can be downscaled to finer spatial resolution.  Usually the forecast resolution is stepped down with a multiple nesting approach with the ratio of grid resolutions being typically 1:8.  This compares favourably with nesting with limited area atmospheric models that require a grid resolution ratio of 1:3 or 1:4.  Higher jumps in resolution can be used if the nudging length scale (typically the width of the high resolution panel) is well resolved by the host model (e.g., 4 grid spaces).  Initial conditions for the nested simulations can be taken from the host model (e.g,, CCAM lower resolution forecast), which allows the downscaled forecasts to start at later times than the original initial conditions.  Alternatively, the initial conditions from the analyses used for the lower resolution forecast can also be interpolated to the nested higher resolution grids.  This allows the nested forecasts to take advantage of higher resolution analyses, but also results in the nested forecasts having to start at the same time as the analysis.

Ensemble forecasting

The skill of a weather forecast can be improved using ensemble forecasting techniques.  In this case an ensemble of initial conditions based on different analyses are employed to run an ensemble of weather forecasts.  The results of these forecasts can be compared to better understand the implications of uncertainty in the initial conditions.  A cost effective ensemble forecast system can be constructed using analyses from different centres.  Alternatively, some centres provide initial conditions based on maximising the growth of different atmospheric modes while still constrained by observations.  CCAM also has been used to construct its own ensemble of forecasts based on its fastest growing modes to form a 16 member ensemble forecast.

Strategies for grid selection

The choice of CCAM grids employed can dramatically impact on the speed of the model.  Here we suggest using grids and resolution that are preselected from a list of optimal grid choices.  The first choice is the grid resolution, which can be based on the resolution of the input analyses.  The user then selects from the following list of grid resolutions

The above table assumes a Schmidt factor of 3 for the coarse grid forecast, so that the grid stretching does not degrade the forecast at synoptic scales.  It is also valid to stop at a lower resolution (e.g., 64km, 8km and 2km).  The output domain size is specified by choosing a grid size from the available list below