This theme covers forecasting by coupled physical modelling systems including atmospheric physics and chemistry, ocean and the land surface, and covers modelling of floods, landslides, bushfires, pollution, etc.. While particular interactions are specific to the hazards selected in HIWeather, the principles of coupled modelling have been well established in the Earth System Modelling community that underpins current research on seasonal and climate prediction. The planned research will draw heavily on aspects of that work under several WCRP programmes and grand challenges, especially GEWEX. This theme will also work closely with the S2S project, and with the CAS Working Group on Numerical Experimentation (WGNE) and the Global Atmosphere Watch (GAW) Working Group on Urban Research, Meteorology & Environment (GURME) on modelling issues.

i. Observations & Nowcasting 

For effective and successful forecasting and warning of high impact weather events on timescales of minutes to one hour, high-resolution observations in time and space are needed. Despite major advances in NWP, computing time and model spin-up result in a gap between analysis time and the availability of useful forecasts. Future improvements in convective-scale NWP and data assimilation on the timescale of this project will shorten but not eliminate this gap. Thus observation-based nowcasting systems will remain essential for warnings at very short timescales, dovetailing with very high resolution NWP at longer timescales. To provide the best possible basis for nowcasting and NWP, new approaches to obtaining very high resolutions of observations are required, using both ground- and satellite-based remote sensing, deployment of dense networks of low cost sensors, and crowd sourcing. Research in the Predictability & Processes theme on processes leading to initiation and evolution of high impact weather events will provide the basis for more advanced nowcasting techniques. 

ii. Data Assimilation (DA) 

A major focus on initializing convective-scale models is needed to achieve the required accuracy of forecasts of high impact weather in the first day. Data assimilation for these models is in its infancy and needs to be developed so that small scales are initialized consistently with the large scales, without distorting the latter, and so that the boundary layer, in particular, is initialized consistently with the land and ocean surface, and with the atmospheric aerosol content.  

Initialization at the convective scale requires the exploitation of new observation sources, highlighted above, which have very different error characteristics from conventional observations including much higher probability of gross error, correlated error, and large biases. For remotely sensed data, the observations may be averaged over areas larger than a model grid cell. Research into the appropriate complexity of cloud assimilation will be necessary as models develop increasingly sophisticated representations of microphysics that relate in complex model-dependent ways to the observed quantities.  

Characterization of the time and space variability of the observation and forecast background errors for data assimilation are key to obtaining more accurate forecasts. In particular, development is needed in: 

a) methods suited to the non-linear and non-Gaussian errors typical of convective scale model evolution 

b) methods that handle position errors 

 iii. Model Development  

Improved forecasts of High Impact Weather depend on model improvements both to extend predictive skill of synoptic scale environments associated with high impact weather and to provide more precise and accurate small scale detail. Prediction of weather impacts requires more sophisticated coupling of NWP with physical impact models (e.g. for storm surges and floods). 

Particular challenges for model development are associated with: 

• development of a new generation of scale independent or scale adaptive parametrizations, e.g. stable boundary layer, cloud-related turbulence, cloud microphysics; 

• representation of process uncertainty through stochastic schemes;  

• development of parametrizations of partially resolved processes spanning 0.5 to 5 grid cells – the “grey zone”; 

• optimization of ocean-atmosphere-aerosol–land surface coupling strategies for small scales and short-to-medium lead time forecasts; 

• representation of weather impacts transmitted through land geophysical processes, biological processes and responses of buildings, vehicles, infrastructure, etc.

 iv. Ensemble Forecasting 

While ensemble forecasting on the synoptic scale has matured over the past two decades, there is a need for research to improve its performance in convective-scale models – both as a result of their much shorter grid-lengths and also their smaller domain size (typically national- or city-scale rather than continental-scale). 

Design of perturbations to represent initial uncertainties in the ensemble forecasts is closely linked to data assimilation (3.2.1). The size and structure of ensemble initial condition perturbations should be determined using objective ensemble data assimilation methods. 

Evolution of forecast uncertainties is governed by the representation of model errors using techniques such as stochastic physical parameterizations.  These techniques need to be designed to represent uncertainties in model physics with particular focus on high impact weather events, e.g. the impact of errors in cloud microphysics on forecasts of heavy rainfall, low visibility or temperature extremes. 

New perturbation strategies are needed to overcome under-spreading in surface weather parameters. Ensemble members currently usually share the same initial surface conditions which likely leads to significant underestimation of the uncertainties in forecasts of near-surface conditions. This may be dealt with by perturbing lower boundary specifications or by using coupled ensembles. 

Interaction with the sea surface is a special example of the influence of surface conditions.  Coupled ensembles may be needed to represent the uncertain impact of diurnal variations in coastal waters on sea-breeze fronts etc. 

A key motivation of ensemble forecasts is to capture the risks of high-impact events in the tail of the statistical distribution.  This has implications for ensemble size.

v. Post-processing, product generation and human interpretation 

Turning raw model outputs into the information required to be communicated to users requires calibration and removal of biases in the physical variables and in their probability distributions, time and space aggregation or downscaling to match user requirements, and diagnosis of ancillary variables of interest to the user that are not part of the model.  

Calibration of direct model output is important for high impact weather forecasts. While the value of using reforecast datasets to calibrate products such as the Extreme Forecast Index has been demonstrated for medium range forecasts, more research is needed to explore strategies appropriate for convective-scale models. 

A key step in the use of ensemble prediction systems is to relate the ensemble sample frequency to the probability of the event occurring. This is dependent on the perturbation strategy, the size of ensemble, the spread/skill relationship of the system, and may also be dependent on weather regime. Experience with medium range ensemble prediction systems needs to be tested at the convective scale, with special emphasis on the performance in the wings of the distribution. There is also a need for work on ensemble post-processing methods that preserve the forecast covariance structure, e.g. for use in flood forecasting. For some applications, there is a need for scenarios with user-defined characteristics rather than probabilities. Methods for extracting these from the ensemble distribution need to be developed. 

Compact presentations of raw NWP information are required for operational forecasters and there is a need to investigate factors that limit the utility of automatic warnings. Methodologies are also needed to combine ensemble NWP outputs with nowcasts to achieve seamless predictions of high impact weather events form minutes to hours.