Assessing Probabilistic Forecasts of Subseasonal-to-Seasonal Climate Extremes Using the North American Multi-Model Ensemble (NMME)

While extreme climate events on subseasonal to seasonal (S2S) timescales can have great impacts on health, agriculture, energy usage, and other aspects of life, few specific early warning systems exist for extremes on these timescales. This study examines the skill of probabilistic S2S extremes prediction in the midlatitudes from a multi-model ensemble of global climate models. S2S extreme 2 m temperature and precipitation events, defined here as 90^thand 10^thpercentile based on the cross-validated historic distribution, are evaluated using 36 years of probabilistic forecasts from the North American Multi-Model Ensemble (NMME). The NMME currently provides real-time guidance for NOAA’s operational short-term climate forecasts. Forecast data from 1982-2017 from seven NMME models contribute to this study: NCEP-CFSv2, Environment Canada’s CanCM3 and CanCM4, GFDL’s CM2.1 and FLOR, NASA-GEOS5, and NCAR-CCSM4. Subseasonal (month-1) and seasonal (3-month-mean) probabilistic forecasts are examined at leads of zero to five months. Forecast quality is assessed using attributes diagrams, and forecast skill is assessed via probabilistic skill scores and contingency table statistics including the symmetric extremal dependence index (SEDI). Forecasts for temperature extremes show aggregate skill for northern and southern midlatitudes. Prediction of precipitation extremes is lower skill in general, but successful forecasts may be possible for some regions and seasons.