ICON (Icosahedral Nonhydrostatic) Model

Left: Schematic depiction of the icosahedral grid structure of ICON
Right: Example of an ICON grid with refinement area over Europe

ICON Model grid (Source Deutscher Wetterdienst)


DWD is one of only fourteen weather services in the world running a global numerical weather prediction (NWP) model. All other weather services operating an own forecast model currently restrict themselves to limited-area models focusing on their specific area of interest. Such models depend on lateral boundary conditions provided by a global model throughout their forecast range.

The predecessor of ICON named GME, which started operational production in December 1999, was the first operational NWP model in the world using an icosahedral grid. This basic grid structure was retained for ICON. Compared to traditional approaches such as the latitude-longitude grid, icosahedral grids provide a nearly homogeneous coverage of the globe. This avoids the so-called pole problem related to the convergence of meridians in lat-lon grids, which poses severe challenges to a computationally efficient implementation.

The grid generation starts with an icosahedron inscribed in the sphere. Connecting the 12 vertices of the icosahedron with geodesic lines yields 20 equilateral spherical triangles with a edge length of about 7054 km. Iterative subdivision (e.g. bisection, trisection) of the triangle edges then yields a model grid with the desired spatial resolution. The effective mesh size of our model grid is defined as the square root of the mean area of the spherical triangles. In the current operational version, the global ICON grid has 2,949,120 triangles, corresponding to an average area of 173 km² and thus to an effective mesh size of about 13 km. All scalar prognostic model variables (e.g. temperature, density, moisture quantities) are located in the circumcenter of the triangles, whereas the edge-normal wind components are located in the edge midpoints.

ICON orography at a mesh size of 13 km for the western part of Europe.

ICON orography (Source Deutscher Wetterdienst)

Corresponding to the mean area of the triangular grid cells of 173 km², all model variables like density, temperature, wind, humidity, cloud water, cloud ice, rain and snow have to be considered as averages over the cell area of 173 km². This also pertains to the external parameters like the orography (see figure orography of Europa). While large mountain ranges can be represented reasonably well at a mesh size of 13 km, capturing individual mountain ranges or valleys requires even finer meshes. However, resolving such features can be important for some aspects, e.g. predicting the channeling of the low-level airflow in the Rhine Valley. Therefore, DWD operates in addition the high-resolution limited-area model COSMO-DE with a mesh size of 2.8 km.

The most important prognostic variables of ICON are air density and virtual potential temperature (which allows diagnosing the pressure), horizontal and vertical wind speed, humidity, cloud water, cloud ice, rain, and snow. These variables are calculated for all grid cells on 90 terrain-following model levels extending from the surface to a height of 75 km, yielding a total of about 265 million grid points. Over land, additional prognostic equations are solved for soil temperature and soil water content for 7 soil levels. If a snow cover is present, snow water equivalent and snow density are predicted in addition. The sea surface temperature over ice-free ocean surfaces is analyzed once per day from observations and is kept constant during the ICON forecast. For ice-covered parts of the oceans, the sea ice fraction is analyzed once per day from obser­vations, whereas the ice thickness and the ice surface temperature are predicted with a simple sea-ice model.

Schematic representation of the processes included in ICON

Physical processes in ICON (Source Deutscher Wetterdienst)



Besides horizontal and vertical transport processes (so-called adiabatic processes) in the atmosphere, diabatic processes like radiation, turbulence, formation of clouds, and precipitation play a major role for NWP. Describing these processes taking place on much smaller horizontal scales than the model mesh size is the task of the physics parameterizations.

Within the operational workflow of ICON, one has to distinguish between the data assimilation cycle and the forecast mode. The purpose of the data assimilation cycle is to determine an optimal initial condition for the forecast runs. To this end, a three-hour forecast starting from the previous analysis (the so-called first guess) is combined with all observations valid for a three-hour time window centered at the analysis date using a 3D variational assimilation method. Forecast runs are performed four times per day, the forecast ranges being 180 h for the forecasts starting at 00 and 12 UTC and 120 h for those starting at 06 and 18 UTC.

In the present configuration, ICON needs about 8 minutes wallclock time per forecast day on DWD’s supercomputer and produces about 900 GB of data per 7-day forecast. Output is written hourly during the first 78 hours of the forecast in order to provide the lateral boundary conditions for the COSMO-EU, afterwards the output frequency is reduced to three-hourly. Besides the above-mentioned prognostic variables, a variety of additional diagnostic quantities is calculated at every output step and provided to the users.

In addition, the ICON forecasts provide the basis for post-processing programs like the global Lagrangian dispersion model, which provides on-demand forecasts in the case of nuclear or chemical accidents, and the ocean wave model.

Besides these DWD-internal applications, many external users need ICON forecasts as basis for their products. For instance, the regional hydrological offices use the precipitation fields for flood forecasting, and the Federal Maritime and Hydrographic Agency uses wind fields for predicting storm surges. Moreover, more than 30 national weather services, among them those of the COSMO (Consortium for Small-scale Modelling) partners Italy, Poland, Romania, and Russia, and of many developing and emerging countries like Botswana, Brazil, Israel, Kenya, Mozambique, Namibia, Nigeria, Oman, Pakistan, Philippines, Tanzania, United Arabic Emirates, and Vietnam use ICON forecasts as lateral boundary conditions for limited-area forecasts.