Deliverable No. 5.2 Strengths and limitations of state-of-the-art weather and climate prediction systems

This document provides an overview of predictive capacity over the Arctic and mid- latitudes of current state-of-the-art prediction systems ranging from numerical weather prediction (NWP) to seasonal time scales.

The assessment is mainly based on forecasting systems and climate models contributing to the APPLICATE project. This deliverable therefore provides a thorough evaluation of the forecast models included in the WP5 stream 1 experiments, and a baseline for future improvements to current systems resulting from developments in the framework of the project.

Beyond commonly used verification metrics for the evaluation of weather and climate predictions, illustrations of current systems predictive capacity are shown by focusing on specific phenomena and case studies (e.g. extreme rainfall on Svalbard). With the perspective of providing useful and reliable forecasts for potential end-users, some skill evaluations on more user-relevant metrics were included.

Results on the weather prediction time scales show the impact of horizontal resolution in better representing precipitation extremes, although some weaknesses remain in a 2.5 km resolution configuration for the Svalbard case study examined in this deliverable. More generally, high resolution limited area models show added value with respect to global models depending on the parameter and region of interest.

At the medium range (5 days), the evaluation of the European Centre for Medium- range Weather Forecasts (ECMWF) forecasts over 1990-present for geopotential height at 500 hPa shows that these have been steadily improving over the Arctic, at the same rate as the Northern Hemisphere in general. Skill and biases are found to vary according to the region and season of interest.

Seasonal re-forecasts over a common 1993-2014 period were evaluated for both atmospheric and sea ice concentration fields. The skill of the systems is quite limited, consistent with previous works. For sea ice, forecast performance for boreal summer seems to depend quite strongly on systematic errors which appear in some systems from the initialization time step. This deliverable also presents results from a statistical forecasting framework, using HighResMIP model simulations to evaluate lagged predictability of sea ice volume with sea ice volume and area as predictors. It appears from the results presented that sea ice area does not add much additional predictability to the information provided by sea ice volume.