i. Observations & Nowcasting
Current operational weather observing networks are unable to capture the detail of the weather that causes hazards, the hazards themselves, or the impacts of those hazards. New developments in ground- and space-based remote sensing, in the widespread deployment of low cost sensors, and in crowd sourcing, need to be harnessed to provide the required information. The major challenges are in sourcing and quality controlling these observations. The next generation of coupled physical models that will be used for hazard prediction will need observations to initialise the processes that are necessary for skilful predictions. We need to identify observation sources that are able to deliver the required information. Apart from precipitation, for which FDPs have provided valuable inter-comparison opportunities, most nowcasting methods have been developed for specific locations or specific applications, with little analysis of the relative performance of different approaches across a range of applications. The development of rapid-update convective-scale DA/NWP systems offers an alternative, more physically-based approach to nowcasting a wide range of hazards. We need to establish the best approaches to nowcasting weather-related hazards, whether observation-driven, NWPdriven, or a hybrid of the two.
ii. Data Assimilation
Traditional approaches to data assimilation at synoptic- and meso- scales rely on weak nonlinearity of the equations. This no longer holds for convective-scale motions. We need to develop data assimilation methods that can cope with this stronger non-linearity. Hazard forecasts must be updated more frequently than traditional synoptic and mesoscale forecasts, perhaps every hour or less. This places demands on the design and efficiency of data assimilation that are different from those placed on traditional 6hourly cycles. Data assimilation systems that are able to meet these needs require development. Many observation types with comparable resolution to that of global models, e.g, most satellite data, become relatively coarsely spaced on the convective scale. New observation operators that account for this ‘error of representativeness’ are required. The assimilation process is strongly influenced by the specification of model error. Given that both synoptic-scale and convective-scale structures will need to be accounted for, new approaches to the specification of model error are required. Component models in coupled prediction systems are usually initialised independently. However, some coupled processes need to be initialised consistently. We need to develop methods of achieving this.
iii. Model Development
Routine hazard forecasting requires the development of coupled prediction systems involving the atmosphere and its composition, land, ocean and ice. Current initiatives are based either on using Earth System Models designed for climate prediction, or on opportunistic coupling of components that happen to be available. We need to identify the aspects of coupling that enable accurate hazard prediction and to develop interface schemes that provide the best forecasts. Forecasting of convection-related hazards depends on accurate representation of convective initiation. We need to develop model parametrizations that produce reliable accuracy in this respect. Convective-scale models have been developed with widely different levels of detail in the microphysics parametrizations. We need a better understanding of the benefits of more detailed schemes, and to use them to improve simpler parametrizations. The land surface interacts with the atmosphere at both local and synoptic scales. Models often use parametrizations of gravity wave and frictional drag that are designed to provide the correct synoptic scale forcing. We need to understand how to accurately parametrize local interactions without degrading the synoptic scale response.
iv. Ensemble Forecasting
Global ensemble prediction systems have achieved a useful level of consistency between predicted uncertainty and observed error for smoothly varying atmospheric variables. We need to achieve similar reliability and to produce well-calibrated probabilities for hazard-related variables. Consistent with solving the multi-scale and coupled model data assimilation challenges, we need to develop ensemble perturbation techniques that reproduce observed error growth rates in multi-scale coupled prediction systems.
v. Post-processing, product generation & human interpretation
Extracting information from raw NWP output that users will find helpful and easy to understand remains a difficult problem except in some highly structured application areas. Together with the Communications theme, we need to develop more effective products. There are particular problems in communicating ensemble-based uncertainties, which will be compounded in convective-scale models where the uncertainties are coming from processes at very different scales. We need to develop better ways of using ensembles and more user-oriented means of communicating uncertainty. Some locally developed diagnostic outputs have potentially much greater value if their applicability could be extended globally. Examples include fire risk indices and ocean & river flood estimation methods. We need to develop the global parameter sets that will enable these products to be generated in a seamless way across the world.