Zur Seitenansicht
 

Titelaufnahme

Titel
Probabilistic Categorical Wind Gust Forecast in Complex Terrain : Case Study on Hahnenkamm, AUSTRIA
VerfasserFiegl, Gerhard
Betreuer / BetreuerinMayr, Georg
Erschienen2014
HochschulschriftInnsbruck, Univ., Masterarb., 2014
Anmerkung
Arbeit an der Bibliothek noch nicht eingelangt - Daten nicht geprüft
Datum der AbgabeOktober 2014
SpracheEnglisch
DokumenttypMasterarbeit
Schlagwörter (EN)complex topography / resolution / NWP / probabilistic / forecast / Kitzbuehel / hourly / artificial snow production / measurement / wínter / LASSO / heteroskedastic extended logistic regression / gust classes / categories / proportion correct / ensemble uncertainty information / Hahnenkamm
Schlagwörter (GND)Hahnenkamm <Kitzbühel> / / Wettervorhersage
URNurn:nbn:at:at-ubi:1-1012 Persistent Identifier (URN)
Zugriffsbeschränkung
 Das Werk ist frei verfügbar
Dateien
Probabilistic Categorical Wind Gust Forecast in Complex Terrain [3.28 mb]
Links
Nachweis
Klassifikation
Zusammenfassung (Englisch)

In complex topography, wind and especially gusts can vary dramatically within a few tens of meters. The resolution of numerical weather prediction (NWP) models is still too coarse to capture these small-scale variations. And yet, probabilistic wind gust forecasts for the downhill ski race at the Hahnenkamm in Kitzbuehel, a famous ski resort situated on the northern side of the Eastern Alps, with hourly resolution for leadtimes 0 to +24h and three-hourly for the following day were needed. The gust model is supposed to ensure safety for the athletes during the training and race and support the scheduling of artificial snow production. Data records of four measurement sites at different altitudes along the race course are available for seven winter seasons starting in November 2006. The basis of the forecast model is a combination of NWP direct model output (DMO) and observations into a subsequent statistical model. The Least Absolute Shrinkage and Selection Operator (LASSO) selected the ideal regressors for every measurement site and hour. Of several logistic regression methods, the heteroscedastic extended logistic regression (HXLR) was best in forecasting the probability of occurrence of five gust classes as measured by the Ranked Probability Skill Score (RPSS). The proportion correct (PC) of the individual hours for the first 24 hours after five-fold cross validation has a median of between 0.52 and 0.72. This compares with a PC of between 0.37 and 0.48 for bias-corrected NWP 10 m gust and a PC of 0.3 to 0.43 for a simple persistency forecast. Adding uncertainty information from the spread of an NWP ensemble increased the median of the RPSS at every site and by up to 0.04. Finally the forecast result is visualized in a compact and intuitive form to easily and quickly absorb the probabilistic information using a HCL color scheme for future operational usage at the Austrian Weather Service (ZAMG).