1. Inspecting How the Model Advances from One Time Step to the Next

The essence of a weather forecasting model is to calculate the change of atmospheric variables such as temperature, pressure, or wind in time. In our models, this change can be divided into two main contributions: the model dynamics, and the model physics. The model dynamics refer to the large-scale transport of air masses with the wind that the model is able to fully resolve. The model physics on the other hand represent processes such as turbulence, radiation or cloud physics, that are too small to be resolved by the model grid, even though AROME-Arctic employs a 2.5 km grid spacing. As a consequence, these processes need to be represented by simplified formulations, so called parameterisations.

All of these parameterised processes contribute to the change of i.e. temperature or wind in our model. We refer to these contributions as individual tendencies. Investigating these tendencies enables targeted studies of the otherwise “hidden” activity and interplay of the model’s physics. Fig.1 demonstrates this utility. Shown on the left is the physical tendency (sum of all parameterisation schemes) of temperature close to the surface (lowest model level) during a day with major cloud formation. The physics actively warm most of the model domain (red shading). However, this warming is not uniform, and distinct spatial patterns emerge. These patterns can be attributed to environmental factors, such as the sea ice edge (white line), but also to model-internal factors. We found that boundary layer types play an important role (Fig. fd1b). Boundary layer types help the model to adjust its physical package towards the plethora of atmospheric regimes that it needs to represent, such as a cloud free winter night vs. an autumn storm. Together with Wim de Rooy (Royal Netherlands Meteorological Institute), we use the new perspective obtained from tendency output to investigate the impacts of a new boundary layer package.

When working with NWP models, end-users are normally provided with hourly to three-hourly output intervals. Even though today’s NWP models internally calculate with time steps of about 60 seconds or so, the sheer number of grid points in a typical model grid (~10^8) makes it simply not feasible to store every single of these time steps. An elegant solution is provided by the DDH-tool (diagnostics on horizontal domains), developed at Météo-France. The DDH tool allows for model output to be written for every single model time step within a specified sub-domain. This enables highly detailed investigation of process representations in the full, three-dimensional model. Especially the combination of such output with observations from super-sites, such as the one in Sodankylä, Finland, yields great potential for validation and model inter-comparison purposes. The tremendous increase in detail that is achieved by the DDH tool is demonstrated in Fig. 3. DDH reveals much finer structures of the development of relative humidity in the model over the measurements site of Sodankylä. Such high-resolution output enables much more detailed process studies of the stable boundary layer and testing of the new parameterisation scheme than otherwise possible.