The scientific goal of achieving seamless predictions – the integration of weather and climate, between the science and users and between nations on observing the atmosphere – is an important objective for the next decade. Seamless weather forecasts and climate predictions could evolve towards seamless weather-climate-impacts forecasting. Increasingly sophisticated numerical models will incorporate more and more of the Earth system’s components and processes. In addition to the atmosphere and oceans, they will integrate increasingly accurate information on topography, land-use change, vegetation, rivers, lakes, clouds and socio-economic trends to provide user-specific decision-support services that will touch almost every part of our lives.
The main scientific challenges for future global Numerical Weather Prediction relate to the recent advancements in physical process parameterization, analysis and forecast uncertainty formulation through ensembles, and the provision of physically consistent initial conditions for forecasts using observations.
Regarding physical parameterizations, the challenge is to determine whether running global models at a one kilometre scale also eliminates all convection-related uncertainties and produces a fundamental stepping stone for reducing model biases and enhancing predictive skill at all forecast ranges. As these high resolutions are not yet in achievable, convection parameterizations will remain crucial for global weather and climate modelling for the next decade and progress in this area will require joint efforts from the weather and climate communities.
More observation of physical processes will be needed because of the coupled modelling of the atmosphere with ocean, land surface and sea-ice models, some of which are already in operational use today. Each coupled model has its own characteristic space and time scales. Coupled modelling is most beneficial beyond the 3–7 day range as ocean, sea-ice and land surface processes are relatively slow and mostly affect longer-term system memory. However, there are examples where the coupling of models also affects the short range: for example, when oceanic upwelling in the wake of slowly moving tropical cyclones affects their intensity or where rain- fall over land is strongly constrained by surface evaporation and thus soil moisture. In such a context, coupled data assimilation will become critical for the initialization of the future coupled models. This assimilation will need to include atmospheric composition – aerosols, trace gases, etc – as well as ocean, land surfaces and sea-ice observations.
Aerosols and trace gases are important to forecast in their own right because of their impacts on air quality. However, as atmospheric constituents they directly affect radiative heating. Aerosols can also act as condensation nuclei in cloud formation and on the heterogeneous chemistry that occurs at the surface of polar stratospheric clouds, accelerating ozone destruction. An associated challenge from adding more physical and chemical processes to models is that initial conditions for these constituents are also required and thus more and complex observations need to be assimilated.
Using more of the existing and new observations and advances in data assimilation pose more scientific challenges for Numerical Weather Prediction. Beyond the maintenance of the backbone satellite and ground-based observing systems that measure vertical profiles of temperature, moisture, clouds and near-surface weather, fundamental observables are missing. An example is the direct observation of upper-level wind with Doppler-radar technology. However, the existing backbone observations also need to be provided by a robust and resilient observing system, which requires substantial international investment and coordination (for more details The quiet revolution of numerical weather prediction, Nature, volume 525, Issue 7567, pages 47–55, published 2 September 2015).