Home

WP4 Task 33042 wiki

Short description of the task

The objective of Task 33042 is to provide probabilistic uncertainty quantification based on deterministic forecasts generated by the extreme DT. The extreme DT will be run on-demand at hectometric spatial resolution with domains determined by a triggering algorithm. Hence, the domains and projections will vary depending on the expected weather conditions.

In Phase 1 the focus was to implement a set of well-established post-processing methods using the yet limited data available for demonstration and testing. Post-processing methods should ideally generate scenarios in space and time of multiple weather variables to be useful for all downstream users. However, at the current stage this is not realistic. Instead the attention is paid to univariate methods that can make forecasts in terms of exceedance probabilities and quantiles. As a possible next step these can be turned into scenarios using copula methods. In phase 2 the approaches will be supplemented with a generative machine learning method that makes ensemble forecasts directly.

Link to the sandbox : https://github.com/DEODE-NWP/WP42-sandbox

List of codes with short description:

The following repositories are created for the various methods

• WP42-QRF: Codes implementing the Quantile Random Forest method.
• WP42-GRF: Codes implementing the Generalized Random Forest method.
• WP42-EGPTail: Codes implementing the Extended Generalized Pareto Tail extension method.
• WP42-GBex: Codes implementing the Gradient Boosting for Extremes method.
• WP42-nbh-unc: Neighborhood method for estimating deterministic model uncertainty.
• WP42-EMOS: Codes implementing the standard EMOS method (truncated logistic).