Scale Interaction and Predictability of Weather

Simple estimates of the predictability of the atmosphere based on the theory of scale-invariant turbulent cascades give a limit of 2-4 weeks for skillful weather forecasting. In reality the atmosphere flow is not scale-invariant, and is strongly influenced by error growth on two dominant scales: approximately 1000 km (cyclones) and 10 km (thunderstorms). The predictability of the state varies from day to day depending on relative degree of instability on these scales and the processes that transfer energy between them. This talk will explore some recently proposed theoretical constraints on these processes, using numerical simulations and simple stochastic models. Implications for probabilistic weather forecasting and data assimilation will be briefly discussed.