Weather Type Forecast Tool · Wiki · Peter Hoffmann / CapTainRain

for Jordan

Quick Info

GFS¹ weather forecast fields of the past 12 days² are operationally post-processed and allocated to recurring large-scale circulation patterns. The developed software provide experts a quick overview on structural changes in the prevailing weather-system: for instance from WT01 (westerly) to WT08 (cyclonic). The respective weather icons should represent the locally expected mean weather conditions per weather-type (WT). However, this isn't finished yet.
The used methodology retrospectively train a decision tree (random forest) between daily weather-types and atmospheric reanalysis fields of the geopotential height at 500 hPa (Z500) for the period 1981 to 2020. The derived relation is finally applied to independent field from weather forecast (+14 days) to assign the respective patterns to sequences of weather-types. Such a reduction of dimensions enables a more comprehensive assessment of ensemble forecasts. We have also setup a version for the Eastern Mediterranean region using an objective classification of recurring weather-types. The software were compiled for windows systems and can be downloaded under: https://gitlab.pik-potsdam.de/peterh/gfswt/-/tree/main/execuable

Training

  graph LR;
  A(<b>Z500<sub>x,y,d</sub></b><hr>Atmospheric Fields) == 1951-1980 ==> C((<b>Decision Tree</b><hr>Training));
  B(<b>WT<sub>k,d</sub></b><hr>Weather-Types) == 1951-1980 ==> C;
  C == Entropy ==> D((<b>Decision Tree</b><hr>Predicting));
  E(<b>Z500<sub>x,y,t</sub></b><hr>Forecast Fields) == +14days ==> D;
  D == +14days ==> F(<b>WT<sub>k,t</sub></b><hr>Weather-Types);
  F --> G(run date;fcst date;cluster<br>2023-05-01;2024-05-01;WT03<br>...<br>2024-05-01;2023-05-14;WT01);
  classDef a fill:orange,stroke:black,color:black;
  class A,B,C,D,E,F a;

Fig.1: Scheme of the methodology used for the re-identification of the predicted weather-types.

Product

Charts	Tables
weather icons are dummies latest prediction objective classified pattern: WT01 - WT20

Fig.2: Hindcast/Forecast chart of predicted weather-types sequences (here Europe). The higher the number of agreements the lower the uncertainty. The resulting weather-type sequence is given in the upper row. The respective weather symbols represent the expected local weather phenomena for Central Europe.

weather-type characteristics

precipitation	temperature