The two- to four-day predictability of midlatitude cyclones: Idealized and real-world case studies

Abstract

The upscale-error-cascade paradigm introduced by Lorenz (1969) led to a focus on potential weather-forecasting challenges arising from initial-condition errors on very small spatial scales. Yet, recent studies have suggested that initial-condition errors on synoptic scales are more important for present-day numerical weather forecasts when averaged over the midlatitudes and over many forecast cycles. These studies are not, however, well suited for investigating the predictability of isolated, coherent structures like midlatitude cyclones. In this dissertation, we investigate the dependence of the 2–4-day predictability of midlatitude cyclones on the scale, magnitude, and structure of initial-condition errors, focusing on those that generate "forecast busts" in which the high-impact weather is significantly altered. Using novel and highly realistic convection-permitting idealized simulations of moist midlatitude cyclones, we compare the growth of synoptic-scale perturbations derived from an adjoint model with the growth of equal-energy-norm perturbations at the smallest resolved scale. For initial magnitudes comparable to those from present-day data assimilation systems, the adjoint perturbations bust the forecast by significantly changing the intensity and location of the cyclone and its accompanying precipitation. In contrast, the upscale growth of the small-scale-wave perturbations via moist convection is too slow to significantly change the cyclone through 2–4-day lead times. These results suggest that a sensitive dependence on the synoptic-scale initial conditions, analogous to the Lorenz (1963) model, may be more relevant to 2–4-day midlatitude-cyclone forecast busts than the upscale-error-cascade paradigm from Lorenz (1969). We also use an adjoint model to obtain the optimal initial-condition perturbations that minimize the 48-h forecast errors associated with the 15 November 2018 East Coast cyclone. We find that these perturbations extend throughout much of the troposphere and across much of North America. We also investigate the most important components of these perturbations by truncating them in physical and wavenumber space and rescaling them to be equal in an energy norm to the full perturbations. Retaining all of the perturbations except those in a localized region or retaining only scales larger than 1000 km both lead to substantial impacts on the forecast, whereas confining the perturbations to a localized region or retaining only scales smaller than 1000 km have weaker impacts. These results suggest that the 15 November 2018 2–3-day forecast bust was strongly sensitive to widespread, large-scale changes to the initial conditions, rather than to uncertainties in localized regions that could more feasibly be reduced via targeted observations.