Foehn has an impact on people living in mountainous regions through e.g., wildfires,
agricultural conditions, and tourism. Currently, foehn time series are often short
or available for well-investigated places only while long foehn time series at a fine
spatial scale are rare. This thesis shows that foehn time series before the beginning
of observations can be reconstructed from meteorological reanalysis data. The
probability that foehn occurred at a particular location is first probabilistically
diagnosed with a statistical mixture model from the short measurement data series.
This probability becomes the response variable of a logistic regression model with
LASSO penalization. Stability selection and cross-validation automatically select
the ideal number of informative variables. The method is applied to measurements
from seven weather stations in the European Alps and ERA5 (ECMWF Reanalysis
5 th Generation) reanalysis data to reconstruct foehn at three-hourly resolution from
1979-2018. Approximately 20 wind, mass, and humidity variables were chosen for
the reconstruction. The relative weight of each category differed among the foehn
locations. The presented method needs around 7 years of standard meteorological
observations at the place of interest. Out-of-sample verification is possible as one
of the investigated locations has a long foehn history. Neither a general set of most
important variables nor a set of typical day or night variables can be obtained as
the selection varies from place to place. But an unconventional variable describing
foehn is found: ‘Northward gravity wave surface stress’ (mgws).