We will show typical measures for the quality of deterministic and probabilistic hindcasts, such as anomaly correlation coefficients or Brier skill scores of standard variables such as near surface temperatures, verified against observations and reanalysis products. In this context the effect on verification measures by mimicking an observational product, e.g. HadCRUT for near surface temperatures, combining sea surface temperatures with near surface air temperatures over land, will be discussed. We will also assess the impact of initialisation on forecast reliability quantifying the statistical relationship between the predicted probabilities and the observed relative frequency of an event.
Initialized hindcasts will further be compared to an ensemble of non-initialized historical simulations assessing the potential added value of initializing the model towards the observed climate state in near-term climate predictions. Over few ocean regions such as the North Atlantic initialized predictions show significant skill beyond the relation to trends due to external forcing. We will show some examples, such as frequency of extra-tropical cyclones, where initialization can or cannot provide potential useful added value. We will also discuss the possibility of an overestimation of the added value due to different behaviour from the observed climate in non-initialized historical simulations.
- Indicates paper has been withdrawn from meeting
- Indicates an Award Winner