Evaluation of heat wave forecasts seamlessly across subseasonal timescales

We develop an extreme heat validation approach for medium-range forecast models and apply it to the NCEP coupled forecast model, for which we also attempt to diagnose sources of poor forecast skill. A weighting strategy based on the Poisson function is developed to provide a seamless transition from short-term day-by-day weather forecasts to expanding time means across subseasonal timescales. The skill of heat wave forecasts over the conterminous United States is found to be rather insensitive to the choice of skill metric; however, forecast skill does display spatial patterns that vary depending on whether daily mean, minimum, or maximum temperatures are the basis of the heat wave metric. The NCEP model fails to persist heat waves as readily as is observed. This inconsistency worsens with longer forecast lead times. Land–atmosphere feedbacks appear to be a stronger factor for heat wave maintenance at southern latitudes, but the NCEP model seems to misrepresent those feedbacks, particularly over the Southwest United States, leading to poor skill in that region. The NCEP model also has unrealistically weak coupling over agricultural areas of the northern United States, but this does not seem to degrade model skill there. Overall, we find that the Poisson weighting strategy combined with a variety of deterministic and probabilistic skill metrics provides a versatile framework for validation of dynamical model heat wave forecasts at subseasonal timescales.