The Forecast Skill Horizon
Head of Predictability Division, Research Department European Centre for Medium-Range Weather Forecasts,
Giovedì 2 Ottobre, ore 15,00
Aula 2.4, Coppito 1
Abstract: Can we aim to make skilful monthly and seasonal forecasts? More generally, where is forecast predictability horizon? At 2 weeks or beyond?
1970s' and 1980s' results led to the conclusion that there is a predictability limit for instantaneous and local forecasts at about 2 weeks, and that skilful monthly forecasts would always be a dream. To the contrary, the past two decades has seen centres like ECMWF extending the limit of skilful forecasts beyond 2 weeks, thanks to a combination of improved model components and algorithms used to estimate the initial and forecast states, the use of more and better observations, and the adoption of finer meshes. Atmospheric processes have been made more realistic, and ocean processes, aerosols and chemical species have been included in the forecast model. Algorithms have been upgraded to provide more accurate and reliable estimates of initial and forecast states, taking into account all possible sources of uncertainties.
These advances have led to the extension of ensemble numerical weather predictions out to the sub-seasonal and the seasonal time ranges, and skilful forecasts are now issued weeks ahead. But so far, to our knowledge there has not been a systematic and consistent investigation of where is the forecast skill horizon, looking at a range of variables, areas, and fields averaged over different 4-dimensional (4D) space-time volumes. In this talk, the “Forecast Skill Horizon” (FiSH) framework is introduced, and the FiSH length is computed for instantaneous and local forecasts, and for forecasts averaged over increasingly coarse 4D space-time volumes. Results will allow us to estimate where is the forecast skill horizon today, and to discuss how we can aim to further extend it.