The talk starts at 12:15.

Speakers: John Bjørnar Bremnes (meteorologisk institutt)

Location: NR and Zoom.

Title: Probabilistic weather forecasting with quantile methods

Abstract: The key component in any weather forecasting system from hours to weeks ahead is numerical weather prediction (NWP) models. These are highly complex mathematical models derived from the laws of physics and generate one or more scenarios of the state of the atmosphere given an initial state. However, since the initial state is never known exactly and the NWP models are simplifications of the atmospheric processes, weather forecasts are more or less inaccurate or uncertain. To partly account for deficiencies in NWP models, statistical methods are widely applied, in particular to optimise forecasts to specific applications. In this talk two quantile approaches will be presented. In the first, local quantile regression, local weights are introduced to allow for nonlinear relations and to avoid crossing quantiles. In the second, Bernstein quantile networks, predictive quantile distributions are specified by Bernstein polynomials and their parameters linked to input features using flexible neural networks.

Join Zoom Meeting:
https://uio.zoom.us/j/63199595088?pwd=Y05GYkJwR1dCUHFhUnlGMFpsdGc3UT09

Meeting ID: 631 9959 5088
Passcode: 302551

Welcome!
Best regards,
Thea Roksvåg and Lars Henry Berge Olsen.