Synoptic scale model HIRLAM

General description of the HIRLAM model

Model equations and discretization

The HIRLAM model is a hydrostatic grid-point model, of which the dynamical core is based on a semi-implicit semi-Lagrangian discretisation of the multi-level primitive equations, using a hybrid coordinate in the vertical. Optionally, an Eulerian dynamics scheme can be used as well.

The prognostic variables horizontal wind components u,v, temperature T, specific humidity q and linearised geopotential height G are defined at full model levels. Pressure p, geopotential height Φ and vertical wind velocity are calculated at “half” levels.  For the horizontal discretization, an Arakawa C-grid is used. The equations are written for a general map projection, but in practice normally a rotated lat-lon grid projection is adopted. A fourth-order implicit horizontal diffusion is applied. More details on the dynamical and numerical aspects of HIRLAM can be found in the HIRLAM Scientific Documentation, December 2002 (Unden et al, 2002).

Physical parametrizations

A variety of sub-gridscale physical processes are taken into account by parametrization schemes. The following table gives a short overview on the schemes used operationally and on additional options implemented so far:

Radiation scheme:

Clouds and condensation:


Surface and soil processes:

Mean and subgrid-scale orography:

  • a MSO/SSO parametrization of vertically propagating buoyancy waves, resonance effects and blocked-flow drag (Rontu et al, 2002).

For more details on any of these schemes, see the HIRLAM Scientific Documentation, December 2002 (Unden et al, 2002).

Data assimilation

The default upper air data assimilation scheme in HIRLAM is 4D-VAR (Huang et al. 2002, to be updated soon). It can be run with multiple outer loops. The total cost function contains background, observation, initialization and large scale constraints. Background error statistics are calculated using the so-called NMC method. A statistical balance condition is applied.

The observation data types which are assimilated by default presently are conventional observations (TEMP, SYNOP, AIREP, PILOT, SATOB, SHIP, DRIBU) and AMSU-A / ATOVS radiances over sea. Additionally, it is possible to assimilate AMSU-A over land and sea ice, AMSU-B, geostationary and MODIS atmospheric motion vectors, SEVIRI cloud-cleared radiances, GPS zenith total delay, wind profilers, radar radial winds and profiles and Seawinds scatterometer data.

Bias corrections are applied to most satellite data. Observation screening involves logical and representivity checks, background quality checks, black-or whitelisting, multi-level and station level checks, redundancy checks and moving platform checks. Super-obbing is used to reduce the amount of observations. A variational quality control algorithm is optionally available.

Optionally, a 3D-VAR (Unden et al. 2002), 3D-VAR with FGAT or optimum interpolation scheme (Unden et al. 2002) can be applied for upper air data assimilation.
By default, a blending procedure is applied, by which large-scale features from ECMWF analyses are combined with small-scale features of the HIRLAM analysis (Yang, 2005). This procedure is to be replaced during 2009 by using a large scale Jk term in the 4D-VAR cost function.
Analysis of surface variables is done for the assimilation of screen level parameters T2m and RH2m (2D-VAR, Rodriguez et al, 2003), sea surface temperature (OI, Holmleid and De Vries 2007), snow depth (optimum interpolation, Cansado and Navascues 2003), and soil temperature and soil water content (Rodriguez et al, 2003).

Initial and boundary conditions

To reduce noise and spinup, analyses are initialized out by digital filter initialization (Lynch et al. 1999). By default an incremental DFI scheme is applied with a Dolph-Chebyshev filter.
Initial and boundary conditions are normally taken from the ECMWF model or from a larger scale HIRLAM model. Lateral boundaries are overspecified, all variables being externally prescribed by the nesting model. Normally a relaxation zone of 10 grid points is adopted. Boundary relaxation is performed after the horizontal diffusion. At the upper boundary a condition of zero vertical velocity is imposed.

Chemistry branch

A chemistry branch has been developed for HIRLAM – the socalled ENVIRO-HIRLAM system. ENVIRO-HIRLAM is a fully coupled atmospheric – chemistry transport system, including passive tracer transport. Emissions are handled as Eulerian point sources located at the lowest model level. For species transport, a Bott advection scheme (Bott, 1989a, Bott, 1989b) can be used, as well as the HIRLAM semi-Lagrangian and Eulerian dynamics. The HIRLAM convection schemes have been modified to allow for the convection of extra tracers. Urban parametrizations have been included (Mahura et al, 2008), as well as several modules for dry and wet deposition and chemical solvers. A full description of this system can be found in Korsholm et al. 2008.

The HIRLAM system

To allow the model to be used for routine operational numerical weather forecasting, the model analysis and forecast code has been embedded in a system of scripts, executables, support libraries, documentation and tools. This overall HIRLAM system must be applicable in all HIRLAM institutes for both operational and research applications. As such, portability of the code and tools included is an important issue.
There is a standard version of HIRLAM, which is referred to as the Reference System. This Reference system (which consists of code, scripts, libraries and tools) is maintained on the HIRLAM server, at ECMWF and at FMI, where it is run as the operational HIRLAM model.
The HIRLAM system contains also tools to monitor and verify the model forecasts and data usage (see the Operational HIRLAM Monitoring and Intercomparison pages, for registered users only).
To install HIRLAM on a local computer, a copy of the Reference system should be obtained. Instructions on how to do this are given here in the HIRLAM system wiki (for registered users only).