Mesoscale meteorological models Claire L. Vincent, Caroline Draxl and Joakim R. Nielsen
Outline Mesoscale and synoptic scale meteorology Meteorological models Dynamics Parametrizations and interactions with the surface Initial conditions and boundary conditions Examples of mesoscale model output over Denmark Probabilistic forecasts Some practicalities of mesoscale modelling Limitations: What mesoscale models can t do 2 Risø DTU, Technical University of Denmark
Uses for mesoscale models Wind Energy Resource assessment in combination with microscale downscaling : What is the wind climate at a site? What are the maximum expected wind speeds at a proposed wind farm site? Short term prediction: How much wind power will be produced tomorrow? How much wind power will be produced in the next 6 hours? Will there be large variations in wind power during the day? What is the maximum expected wind speed tomorrow? Would it be safe to do maintenance on the turbines tomorrow? Other applications: fire weather when will the wind change direction?, aviation which runway to use?, marine forecasts is it safe?, search and rescue where are the survivors likely to be found?, air pollution where will the plume end up? 3 Risø DTU, Technical University of Denmark
Synoptic scale meteorology High pressure system Low pressure system Cold front Warm front (http://www.metoffice.gov.uk/weather/uk/surface_pressure.html#view) 4 Risø DTU, Technical University of Denmark
Mesoscale Meteorology Mesoscale cellular convection Mountain and valley breezes Tornado Sea breeze circulation Picture courtesy of www.dmi.dk Thunderstorm L H Picture courtesy of www.dmi.dk H SEA L LAND Picture from http://www.news2.dk/pdf/20060224x008.pdf 5 Risø DTU, Technical University of Denmark
Time and space scale of atmospheric motion Typical sizes globalscale 2000 km synopticscale 2000 km microscale mesoscale 20 km Thunderstorms tornadoes waterspouts 2 m small turbulent eddies Land-sea breeze Mountainvalley breeze Hurricanes Tropical Storms Mid latitudes Hs & Ls fronts Long waves secs to mins mins to hours hours to days days to a week or more 6 Risø DTU, Technical University of Denmark Typical life span
What is a meteorological model? Discretize the continuous equations of motion for the atmosphere: Momentum: Mass: Heat Moisture Gas law 7 Risø DTU, Technical University of Denmark
Parameterizations There always unresolved processes that cannot be represented by a numerical model. These features are approximated through Parametrization! 8 Risø DTU, Technical University of Denmark
Parameterizations A rain shower of size several kilometres will fit entirely within a grid cell Eddies relating to turbulent fluctuations also fit entirely within a grid cell Sub-grid-scale processes influence the larger scale flow therefore they cannot be ignored in a weather model! The effect of sub-grid-scale processes is parameterized in the model For example, the transfer of momentum due to turbulent fluctuations can be related to the vertical gradient in the mean wind speed. Similar relations exist for heat and moisture. z u 9 Risø DTU, Technical University of Denmark
Fundamental Parameterizations Surface Energy Balance Sensible and latent heat flux Ground heat flux Q sfc = (1-a)Q s + Q lu -Q ld + Q H + Q E -Q G longwave up shortwave longwave down Soil-vegetation-atmosphere Vegetation, Soil moisture, Evapotranspiration Water-Atmosphere SST, Sensible and Latent heat fluxes Turbulence and diffusion Convection Storms, precipitation, gust Microphysics Particle type, size and distribution Clouds Land use categories 10 Risø DTU, Technical University of Denmark
11 Risø DTU, Technical University of Denmark
Initial conditions and boundary conditions Running a mesoscale model is like solving a large set of differential equations. Therefore, one needs: BOUNDARY CONDITIONS and INITIAL CONDITIONS Obtained from one of several global models that are run every day by national weather providers... Eg: ECMWF (Europe) NWS (USA) Where does the global model get its initial conditions from? Millions of measurements from all over the globe are shared via the World Meteorological Organisation, and combined with the previous forecast (data assimilation) to create an ANALYSIS 12 Risø DTU, Technical University of Denmark
Initial Conditions Instrumental errors (systematic / random) Measurement errors Representativeness error (systematic / random) Errors of human origin Observation errors (wrong time, location, uncalibrated, ) Good quality control is needed! 13 Risø DTU, Technical University of Denmark
Initial conditions and spin-up time for a mesoscale model 1. Data from a large scale analysis are interpolated to the finer mesoscale grid. 2. Extra (possibly targeted) observations could be combined with the large scale model and the previous forecast (data assimilation) 3. Boundary conditions from the large scale forecast are applied every 3 or 6 hours. Initially, the mesoscale model contains only the scales that it inherited from the large scale model. It takes time for mesoscale variance to develop in a mesoscale model: The model has a spin-up time of around 6 hours. 14 Risø DTU, Technical University of Denmark
From large scale to small scale forecasts 6 km grid 54 km grid 18 km grid 2 km grid 15 Risø DTU, Technical University of Denmark
Mesoscale modelling over Denmark and the North Sea 16 Risø DTU, Technical University of Denmark
Mesoscale modelling over Denmark and the North Sea 17 Risø DTU, Technical University of Denmark
Mesoscale modelling over Denmark and the North Sea 18 Risø DTU, Technical University of Denmark
Mesoscale modelling over Denmark and the North Sea 19 Risø DTU, Technical University of Denmark
Wind speed Observed: 60 m Model: 55m Wind direction *** Observed: 60 m Model: 55m 20 Risø DTU, Technical University of Denmark
Probabilistic forecasting There is always uncertainty in the initial state of the model. The atmosphere is a chaotic system, so initial errors can grow in a non-linear way We can benefit from knowing the uncertainty in the modelling system Uncertainty can be described by running an ensemble of models, with either perturbed initial conditions, or with perturbed parameterizations. Figure from P. Pinson, H. Madsen (2009). Ensemble-based probabilistic forecasting at Horns Rev. Wind Energy 12(2), pp. 137-155 (special issue on Offshore Wind Energy). 21 Risø DTU, Technical University of Denmark
Practicalities of mesoscale modelling Typical resolutions and domain size Mesoscale models have horizontal grid spacing ( x) between about 1km and 15km. The smallest features resolved in a model with grid spacing x have a spatial scale of about 5 x to 7 x Boundary conditions and initial conditions can be obtained in near-real time from a global model: for example, the ECMWF (European) or NWS (American) 22 Risø DTU, Technical University of Denmark
Practicalities of mesoscale modelling Typical resolutions and domain size Mesoscale models have horizontal grid spacing ( x) between about 1km and 15km. The smallest features resolved in a model with grid spacing x have a spatial scale of about 5 x to 7 x Boundary conditions and initial conditions can be obtained in near-real time from a global model: for example, the ECMWF (European) or NWS (American) Typical output from a mesoscale model Model variables at all grid points: u and v wind components, vertical velocity, pressure, potential temperature, moisture, precipitation (rain, hail, snow), cloud fraction, outgoing longwave radiation (and many more!) Diagnostics: For example, 10 metre wind speed, 2 metre temperature, mean sea level pressure. Model data is often output every hour, or every 3 hours. 23 Risø DTU, Technical University of Denmark
Practicalities of mesoscale modelling: Computing resources Numerical weather prediction models are usually run on a supercomputer or linux cluster Example: Real-time mesoscale model run twice daily at Risø DTU 48 hour simulation with 2km grid spacing. Simulation on 9 cores with a total of 36CPUs. Simulation time: 3 hours. 24 Risø DTU, Technical University of Denmark
Some things that mesoscale models can t do 1. Mesoscale models can t resolve weather features smaller than about 5 x 2. Mesoscale models can t describe topographic effects smaller than the grid spacing. A hill that fits entirely within a grid cell definitely won t be seen by the model. 3. Mesoscale models cannot explicitly describe roughness changes for example a small forest that are smaller than the grid spacing. 4. Mesoscale models cannot explicitly resolve turbulence 5. Mesoscale models generally do not explicitly resolve individual cumulus clouds. 6. Mesoscale models rely on initial conditions, boundary conditions and sometimes some extra local observations. The mesoscale model is limited by the quality of these inputs. 7. A single run of a mesoscale model cannot diagnose the likelihood of a certain weather outcome. For this, an ensemble of mesoscale models is needed. 25 Risø DTU, Technical University of Denmark
Conclusions Mesoscale models are numerical weather prediction models that have a horizontal grid spacing between about 1 km and 15 km. Mesoscale models are very good at capturing mesocale processes, but are limited by the horizontal grid spacing of the model. In a mesoscale model as with all NWP models, some scales can be resolved explicitly, while smaller scales must be parametrized. To further downscale the output of a mesoscale model to capture small scale topography or localised roughness changes, a microscale model is needed. 26 Risø DTU, Technical University of Denmark