A Hybrid Statistical-Dynamical System for the Seamless Prediction of Daily Extremes and Subseasonal to Seasonal Climate Variability

We propose to demonstrate the skill and suitability for operations of a statistical- dynamical prediction system that yields seamless probabilistic forecasts of daily extremes and subseasonal-to-seasonal temperature and precipitation. We recently demonstrated a Bayesian statistical method for post-processing seasonal forecasts of mean temperature and precipitation from the North American Multi-Model Ensemble (NMME). We now seek to test the utility of an updated hybrid statistical-dynamical prediction system that facilitates seamless subseasonal and seasonal forecasting. Specific updates we intend to implement for the forecast system include: 1) Aggregation of post-processed daily forecasts to enhance the skill of subseasonal forecasts on weekly and biweekly timescales; and 2) Disaggregation of seasonal forecasts to determine the probability of daily extremes. We propose to apply the method developed by the co-PIs of this proposal (Schepen et al., 2017b) to first calibrate climate model daily forecasts through Bayesian joint probability modeling and then relate these calibrated daily forecasts made at di↵erent leads through application of the Schaake Shu✏e approach (Clark et al., 2004). The calibrated and shuffed daily forecasts will then be aggregated for subseasonal and seasonal prediction. Through this approach, forecast skill that exists at shorter subseasonal leads (e.g., weeks 1-2) will be used to improve forecast skill at longer leads (e.g., weeks 3-4). Furthermore, using the methodology developed by the co-PIs (Schepen et al., 2017a), we propose to disaggregate seasonal forecasts from the NMME into distributions of daily values. We will first develop hybrid statistical-dynamical models that use skillful NMME forecasts of large scale climate patterns (e.g., ENSO) in statistical models that relate these remote climate patterns to North American temperature and precipitation variability. Forecasts from these hybrid models first will be used to predict seasonal temperature and precipitation, and then will be statistically disaggregated to generate consistent, seamless forecasts of the distribution of daily temperatures or precipitation amounts. The probability of daily extremes of temperature or precipitation during a seasonal forecast period will be produced, taking full advantage of the enhanced predictability o↵ered by interannual models of variability, such as ENSO, the Arctic Oscillation, or climate change. Importantly, this method allows for the representation of daily extremes consistent with climate conditions.

Relevance:
The proposed project is directly relevant to Competition 3 focus area 1 in that the primary deliverable will be a hybrid statistical-dynamical prediction system, applying post-processing techniques developed in the broader community for operational purposes. The project is relevant to NWS goal 3 to “Complete the seamless suite of NCEP weather and climate products by filling the week 3-4 gap.” This project also addresses NOAA’s long- term goal of a “Weather-Ready Nation: Society is prepared for and responds to weather- related events,” by providing information on the potential for daily extreme events related to climate forecasts.