• English 
  • Spanish 

Local Weather Forecast: Statistical and Dynamical Methods

The predictions can be interactively visualized through the iMETEO googlemap-based portal, powered by Predictia. This service also allows personalization of the view by selecting the desired panel (maps, meteograms, or tables).

Local weather forecast is one of the most challenging problems in operational meteorology from both scientific and socioeconomic reasons. Downscaling methods work by post-processing the outputs of global atmospheric numerical models (for instance, the ECMWF or the GFS models) using either dynamical or statistical methods as illustrated in the figure below.

downscaling.png

Each of these approaches has advantages and disadvantages that have been pointed out in different studies. However, although these methods are very popular and widely used, there are only a few comparison studies for short and medium-range weather forecast.

The main goal of this project is gaining experience in the day-by-day performance of both methodologies and producing a database of daily predictions to conduct an objective comparison to assess the advantages and shortcomings of both methodologies. Thus, daily dynamical and statistical downscalings are being carried out in Spain since January 2010, considering the predictions of the global model GFS(*) as boundary conditions (for the dynamical models) and predictors (for the statistical ones). The downscaling methods used are open-source tools:

  • The WRF-UC Iberia 9km multi-physics simulations are done with the WRF4G GRID-enabled version of the open-source WRF (Weather Research and Forecast) model developed by NCAR. In this case we use WRF-ARW 3.1.1. In this project we use different sets of parameterizations to run two nested grids at 27 and 9 km, for a small North-Atlantic region and the Iberian Peninsular domains shown below. For each run, a total of 108 hours are simulated daily, from 12 UTC of day+0 to 00 UTC of day+4. The proyection used is a Lambert Conformal conic proyection.
    domains_small.png

    The configurations used in the multi-physics ensemble are the following:

    • WRF-UC-Phys1: Microphysics: WSM 5-class scheme; Cumulus scheme: Kain-Fritsch; Long-wave radiation physics: rrtm scheme; Short-wave radiation physics: Dudhia scheme; Surface Layer physics: Monin-Obukhov scheme; Panetary Boundary Layer physics: YSU scheme; Surface physics: Rapid Update Cycle land-surface model .
    • WRF-UC-Phys2: Microphysics: Ferrier (new Eta) microphysics; Cumulus scheme: Kain-Fritsch; Long-wave radiation physics: rrtm scheme; Short-wave radiation physics: Dudhia scheme; Surface Layer physics: Monin-Obukhov (Janjic Eta) scheme; Panetary Boundary Layer physics: Mellor-Yamada-Janjic (Eta) TKE scheme; Surface physics: Rapid Update Cycle land-surface model.
    • WRF-UC-Phys3: Microphysics: WSM 5-class scheme; Cumulus scheme: Kain-Fritsch; Long-wave radiation physics: rrtm scheme; Short-wave radiation physics: Dudhia scheme; Surface Layer physics: Monin-Obukhov (Janjic Eta) scheme; Panetary Boundary Layer physics: Mellor-Yamada-Janjic (Eta) TKE scheme; Surface physics: Rapid Update Cycle land-surface model.
    • WRF-UC-Phys4: Microphysics: WSM 5-class scheme; Cumulus scheme: Kain-Fritsch; Long-wave radiation physics: rrtm scheme; Short-wave radiation physics: Dudhia scheme; Surface Layer physics: Pleim-Xiu surface layer; Panetary Boundary Layer physics: Asymmetric Convective Model 2 PBL; Surface physics: Rapid Update Cycle land-surface model.
    • WRF-UC-Phys5: Microphysics: WSM 5-class scheme; Cumulus scheme: Grell-Devenyi ensemble scheme; Long-wave radiation physics: rrtm scheme; Short-wave radiation physics: Dudhia scheme; Surface Layer physics: Pleim-Xiu surface layer; Panetary Boundary Layer physics: Asymmetric Convective Model 2 PBL; Surface physics: Rapid Update Cycle land-surface model.
  • PROMETEO is a statistical downscaling tool implementing several downscaling methods based on the MeteoLab open-source toolbox for Matlab. In particular two different downscaling methods are applied in this comparison project: A method of analogs using the weighted means or frequencies of the 15 nearest analog days, and a regression conditioned on 100 weather types obtained using the k-means algorithm. In both cases a grid covering the Iberia peninsula is used, including appropriate large scale circulation variables and humidity at 850mb as predictors. The downscaling is performed on a set of over 2500 observatories of the Spanish Met Service AEMET used to build the Spain02 dataset.

The results for precipitation and maximum and minimum temperature are daily produced from the 12 UTC run of the GFS with both statistical and dynamical methods and are part of the PAULA project (Predicción Atmosférica Unificada y de Libre Acceso). Moreover, additional variables such as 10m wind, sea level pressure, total cloud area, radiation, dew point, snow, etc. are also produced by the WRF-UC model and can be visualized from the iMeteo portal and freely downloaded under creative commons CC BY-SA license from the WRF-UC THREDDS server (see the data use and disclaimer).

(*) The Global Forecast System (GFS) is a T382L64 global numerical weather prediction model run by NOAA. It is run four times a day and produces forecasts up to 7 days in advance at a 35km resolution with 64 vertical layers, producing a forecast for every 3rd hour.

Licencia de Creative Commons
WRF-UC Iberia by Santander Meteorology Group is licensed under a Creative Commons Attribution-ShareAlike 3.0 Unported License. Permissions beyond the scope of this license may be available here.

AttachmentSize
iMeteo.png16.83 KB