Variations in the Predictability of Extremes in Subseasonal Multi-Model Ensemble Forecasts

Thursday, 18 December 2014
Dan C Collins, Climate Prediction Center College Park, College Park, MD, United States
Recently there is much interest in bridging the gap between daily weather forecasts out to two weeks lead-time and seasonal climate forecasts and prediction of extremes or rare events such as for heat and precipitation at longer lead times. While predictability, defined as the magnitude of the climate signal relative to the magnitude of noise is small for subseasonal forecasts, climate extremes result from both an increase in the magnitude of the forced climate signal and chance occurrence from random noise. Although the skill of subseasonal forecasts is limited, the greater magnitude of the signal of climate extremes may result in increased predictability, if uncertainty due to noise increases more slowly than the signal magnitude. Using subseasonal ensemble model forecasts (CFSv2 and ECMWF for weeks 2 through 4, or NMME models for months 1 and 2), this study will examine the strength of the signal in subseasonal MME forecasts for extreme events relative to the strength of the signal for non-extreme events. Variations in ensemble spread or model-predicted uncertainty, as well as correlation and mean square error between extreme event and non-extreme event forecasts will be examined. MME forecasts will be partitioned by both observed and predicted extremes.

While the skill of weather forecasts rely to a great extent on the initial state of the atmosphere, seasonal forecasts derive signals from the evolution of slowly varying boundary conditions, such as sea surface temperatures, and processes of climate variability. Climate change on multi-decadal timescales provides an additional source of predictability, as prediction of subseasonal climate variability will depend on signals due to any climate process with a subseasonal timescale or longer. Subseasonal predictability has been shown to arise from MJO and ENSO (Johnson et al., 2014). Differences in MME spread, correlation, and mean square error will be used to examine changes in predictability related to ENSO and climate change, while assessing the ability of models to reproduce observed signals related to climate processes.

Johnson, N. C., Collins, D. C., Feldstein, S. B., L’Heureux, M. L., & Riddle, E. E. (2014). Skillful Wintertime North American Temperature Forecasts out to 4 Weeks Based on the State of ENSO and the MJO. Weather and Forecasting, 29(1), 23-38.