Simon_news

FWF project granted for research on weather forecasting

Project leader Thorsten Simon, together with Achim Zeileis, was granted a FWF research fund for their project "Multivariate Probabilistic Forecasting of Weather Using Joint Distributional Regression".

Project leader Thorsten Simon, together with Achim Zeileis, were granted a FWF research fund for their project "Multivariate Probabilistic Forecasting of Weather Using Joint Distributional Regression".

The grant sum is 317.220,75 €. See this link for the FWF project website. 

About the project

A numerical weather prediction (NWP) ensemble prediction system (EPS)
simulates the state of the atmosphere in a probabilistic manner with a
physically consistent covariance in space, time, and among atmospheric parameters.
Multivariate post-processing aims at preserving this covariance structure,
traditionally by sequentially calibrating the margins using univariate methods and
restoring the covariance by (empirical) copulas.

The proposed research aims at developing a new framework for single-step multivariate
post-processing that simultaneously models both margins and correlations of a multivariate
normal distribution. Both the marginal and the joint correlation parameters will be
conditioned on NWP-EPS output by linking them to additive predictors in a distributional
regression framework.
This complex task requires either a sparse parameterization of the correlation structure
and/or regularization in order to estimate the regression coefficients. In the proposed
framework different regularization strategies can be applied with special emphasis given
to algorithms from the gradient boosting family.

The novelty of the approach is the use of regularized distributional regression
for modeling the joint probability distribution
of atmospheric parameters over space or time.

After laying the methodological foundation for the joint distributional regression
approach (along with corresponding software), the methods will be developed to fit the
needs of multivariate temporal or spatial forecasting,
of temperature, dew-point temperature, pressure, and
wind speed dealing with different temporal scales
(meso vs. synoptical) or different spatial dimensions (horizontal vs. vertical).

Nach oben scrollen