Seasonal Prediction of an Extreme Temperature Index over the United States

Heat and cold extremes have the potential to cause significant societal harm; advanced warming of the frequency of these events for a given season could help with planning and mitigation measures. While an individual temperature extreme event may not be predictable on seasonal time scales, the number of such events could be. In this study, we use an index representing the number of extreme days per season to examine the ability of NASA’s Goddard Earth Observing System (GEOS) sub-seasonal to seasonal (S2S) prediction system to predict temperature extremes over the continental United States. The warm extreme index is defined as the number of days per season where the daily mean temperature is above the calendar-day 90^th percentile, and the cold extreme index is defined as the number of days per season where the daily mean temperature is below the calendar-day 10^th percentile. These indices are computed for 1991-2020 using the GEOS-S2S seasonal retrospective forecasts as well as the Modern Era Retrospective Analysis for Research and Applications Version 2 (MERRA-2) for validation. Temporal correlations are assessed over this period at the grid point scale, indicating significant correlation between the retrospective forecasts and MERRA-2, particularly in the western half of the United States. Spatial correlations between the retrospective forecast and MERRA-2 indices are computed over the United States for each year revealing year-to-year variability in the forecast skill. We investigate local and remote processes associated with increased prediction skill of temperature extremes in a given year.