Increasing Subseasonal-to-Seasonal (S2S) Forecast Skill through Climate Teleconnections: A Hybrid Statistical-Dynamical Prediction System

We seek to enhance the prediction skill and probabilistic reliability of S2S forecasts derived from the output of individual ensemble model and multi-model ensemble systems of the seasonal and subseasonal North American Multi-Model Ensemble (NMME), through the use of advanced prediction methodologies developed by co-PIs to this proposal from CSIRO of Australia.

The application of multi-model ensemble (MME) forecasts to S2S timescales and in particular extreme event prediction, requires evaluation and calibration of the model probabilistic distributions of surface air temperature and precipitation, utilization ofknowledge of climate drivers, and optimization of consolidation of multi-model systems, to assure forecasts are skillful in discriminating the potential for extreme events and reliable in conveying the uncertainty associated with the predictability of such events. To improve capacity to use MME for extremes, we will apply well-tested methodologies of model calibration and combination developed by co-PIs Wang and Schepen to the seasonal and subseasonal NMME forecasts. We will use the NMME Phase-II and the Subseasonal NMME datasets for identification of changes in the potential for extremes in daily data at S2S timescales.

We hypothesize that models of the NMME simulate some of the climate phenomena and drivers of subseasonal timescale climate variability (e.g. AO, MJO, and ENSO), but may be inconsistent in representation of teleconnections between climate drivers and North American climate variability. Information on teleconnections derived from both the NMME fields and reanalysis can be used to optimally combine predictions of the impacts of climate drivers on North American climate, as represented both within the GCMs and by statistical models. We further hypothesize that optimizing the calibration using methods of the co-PIs will improve the skill of the tails of the probability distribution and prediction of extreme events.

This project will apply the calibration, bridging and merging (CBaM) method (Schepen et al. 2014) to post-process NMME forecasts and produce hybrid statistical-dynamical forecasts. Bridging is used to indirectly forecast precipitation and temperature from model forecasts of climate indices (e.g. Nino 3.4). Merging (Wang et al. 2012, Schepen et al. 2015) optimally combines calibration and bridging forecasts from one or more climate models. The Phase-II and subseasonal NMME daily temporal resolution will allow the prediction of frequencies of high-impact events or daily extremes.