Using Numerical Weather Prediction Models for Climate Modeling

The concept of seamless prediction atmospheric models designed both for numerical weather forecasting and climate change modeling is considered. It consists of the fact that there are no artificial time boundaries in the atmosphere that separate synoptic, seasonal, and interannual scales. Due to the nonlinearity of the atmosphere, the processes of all spatial and time scales interact with each other. Thus, an atmospheric model focused on reproducing any phenomena must adequately reproduce phenomena of all time scales. It is wrong to talk about seamless atmosphere models as universal models that function at any possible mesh sizes. The same model cannot be used at the horizontal resolution of about 1–2 and 50–80 km, since the processes of convective precipitation formation are mainly described explicitly in the first case, while the deep convection is a subgrid scale process that has to be parametrized in the second case. This article provides an overview of the implementations of multiscale models with the example of some foreign (the unified model (UM) of the UK MetOffice and the European model EC Earth 3) and Russian SL-AV (Semi-Lagrangian, based on the Absolute Vorticity equation) atmosphere general circulation model. An example of a modification to the deep convection parameterization developed for the climate version of the SL-AV model is given that significantly reduced the errors of the medium-range forecast in the tropics. The use of the same atmosphere model for numerical weather prediction, for the probabilistic forecast of large-scale weather anomalies at monthly and seasonal scale, and for the reproduction of the modern climate (as a part of the Earth system model) is quite possible and gives good results. The further application of the seamless prediction concept is increasingly based on the Earth system models, including the global ocean model, the sea ice model, small gas components model, etc. Developing systems for assimilating observational data for such coupled models is an important task.