Reducing the spin-up of a regional NWP system without data assimilation

In both operational and research settings, kilometre-scale regional numerical weather prediction (NWP) models are often initialised by interpolating analyses from a global modelling system on to the finer regional grid. When initialised in this way (known as a cold start), the first few hours of the forecast are characterised by a rapid growth of precipitation as the system develops convective-scale structures. This is the well-known spin-up problem. Spin-up effects limit the utility of cold-start models for short-range forecasting of severe weather events and model development activities. In this article, we present a method for reducing the spin-up of regional NWP models, without data assimilation (DA). The basis of our method is periodically to insert large-scale information from global model analyses into a continuously cycling regional model, thus updating the large scales whilst preserving fine-scale structures. We refer to this as a warm start. Our method is tested in a regional model over Darwin, for a 10-week period in the 2016/2017 wet season. It is shown that warm-starting indeed reduces the spin-up of precipitation significantly in the first 6–12 hr of the forecast compared with cold-starting. Full objective verification against both radar and satellite observations reveals that precipitation forecast skill is also improved during this time. In addition, the large scales remain closer to global analyses (taken as a best guess of reality) than in a free-running model in which the large scales are updated only via the lateral boundary conditions. Our warm-start technique thus provides a considerable improvement on standard cold-starting and could also be integrated into a regional modelling system with full convective-scale DA as a means of reducing analysis errors on large scales.