Tim Palmer

Abstract:

The use of stochastic processing hardware and low precision arithmetic in atmospheric models is investigated. Stochastic processors allow hardware-induced faults in calculations, sacrificing bit-reproducibility and precision in exchange for improvements in performance and potentially accuracy of forecasts, due to a reduction in power consumption that could allow higher resolution. A similar trade-off is achieved using low precision arithmetic, with improvements in computation and communication speed and savings in storage and memory requirements. As high-performance computing becomes more massively parallel and power intensive, these two approaches may be important stepping stones in the pursuit of global cloud-resolving atmospheric modelling. The impact of both hardware induced faults and low precision arithmetic is tested using the Lorenz '96 model and the dynamical core of a global atmosphere model. In the Lorenz '96 model there is a natural scale separation; the spectral discretisation used in the dynamical core also allows large and small scale dynamics to be treated separately within the code. Such scale separation allows the impact of lower-accuracy arithmetic to be restricted to components close to the truncation scales and hence close to the necessarily inexact parametrised representations of unresolved processes. By contrast, the larger scales are calculated using high precision deterministic arithmetic. Hardware faults from stochastic processors are emulated using a bit-flip model with different fault rates. Our simulations show that both approaches to inexact calculations do not substantially affect the large scale behaviour, provided they are restricted to act only on smaller scales. By contrast, results from the Lorenz '96 simulations are superior when small scales are calculated on an emulated stochastic processor than when those small scales are parametrised. This suggests that inexact calculations at the small scale could reduce computation and power costs without adversely affecting the quality of the simulations. This would allow higher resolution models to be run at the same computational cost. © 2013 The Authors.

Abstract:

The reliability of multi-year predictions of climate is assessed using probabilistic Attributes Diagrams for near-surface air temperature and sea surface temperature, based on 54 member ensembles of initialised decadal hindcasts using the ECMWF coupled model. It is shown that the reliability from the ensemble system is good over global land areas, Europe and Africa and for the North Atlantic, Indian Ocean and, to a lesser extent, North Pacific basins for lead times up to 6-9years. North Atlantic SSTs are reliably predicted even when the climate trend is removed, consistent with the known predictability for this region. By contrast, reliability in the Indian Ocean, where external forcing accounts for most of the variability, deteriorates severely after detrending. More conventional measures of forecast quality, such as the anomaly correlation coefficient (ACC) of the ensemble mean, are also considered, showing that the ensemble has significant skill in predicting multi-annual temperature averages. © 2012. American Geophysical Union. All Rights Reserved.

Abstract:

There is no more challenging problem in computational science than that of estimating, as accurately as science and technology allows, the future evolution of Earth's climate; nor indeed is there a problem whose solution has such importance and urgency. Historically, the simulation tools needed to predict climate have been developed, somewhat independently, at a number of weather and climate institutes around the world. While these simulators are individually deterministic, it is often assumed that the resulting diversity provides a useful quantification of uncertainty in global or regional predictions. However, this notion is not well founded theoretically and corresponding 'multi-simulator' estimates of uncertainty can be prone to systemic failure. Separate to this, individual institutes are now facing considerable challenges in finding the human and computational resources needed to develop more accurate weather and climate simulators with higher resolution and full Earth-system complexity. A new approach, originally designed to improve reliability in ensemble-based numerical weather prediction, is introduced to help solve these two rather different problems. Using stochastic mathematics, this approach recognizes uncertainty explicitly in the parametrized representation of unresolved climatic processes. Stochastic parametrization is shown to be more consistent with the underlying equations of motion and, moreover, provides more skilful estimates of uncertainty when compared with estimates from traditional multi-simulator ensembles, on time-scales where verification data exist. Stochastic parametrization can also help reduce long-term biases which have bedevilled numerical simulations of climate from the earliest days to the present. As a result, it is suggested that the need to maintain a large 'gene pool' of quasi-independent deterministic simulators may be obviated by the development of probabilistic Earth-system simulators. Consistent with the conclusions of the World Summit on Climate Modelling, this in turn implies that individual institutes will be able to pool human and computational resources in developing future-generation simulators, thus benefitting from economies of scale; the establishment of the Airbus consortium provides a useful analogy here. As a further stimulus for such evolution, discussion is given to a potential new synergy between the development of dynamical cores, and stochastic processing hardware. However, it is concluded that the traditional challenge in numerical weather prediction, of reducing deterministic measures of forecast error, may increasingly become an obstacle to the seamless development of probabilistic weather and climate simulators, paradoxical as that may appear at first sight. Indeed, going further, it is argued that it may be time to consider focusing operational weather forecast development entirely on high-resolution ensemble prediction systems. Finally, by considering the exceptionally challenging problem of quantifying cloud feedback in climate change, it is argued that the development of the probabilistic Earth-system simulator may actually provide a route to reducing uncertainty in climate prediction. © 2012 Royal Meteorological Society.