Fenwick Cooper

Abstract:

Existing weather models are known to have poor skill at forecasting rainfall over East Africa. Improved forecasts could reduce the effects of extreme weather events and provide significant socioeconomic benefits to the region. We present a novel machine learning based method to improve precipitation forecasts in East Africa, using postprocessing based on a conditional generative adversarial network (cGAN). This addresses the challenge of realistically representing tropical rainfall, where convection dominates and is poorly simulated in conventional global forecast models. We postprocess hourly forecasts made by the European Centre for Medium-Range Weather Forecasts Integrated Forecast System at 6-18h lead times, at 0.1° resolution. We combine the cGAN predictions with a novel neighbourhood version of quantile mapping, to integrate the strengths of machine learning and conventional postprocessing. Our results indicate that the cGAN substantially improves the diurnal cycle of rainfall, and improves predictions up to the 99.9th percentile (∼ 10mm/hr). This improvement extends to the 2018 March–May season, which had extremely high rainfall, indicating that the approach has some ability to generalise to more extreme conditions. We explore the potential for the cGAN to produce probabilistic forecasts and find that the spread of this ensemble broadly reflects the predictability of the observations, but is also characterised by a mixture of under- and overdispersion. Overall our results demonstrate how the strengths of machine learning and conventional postprocessing methods can be combined, and illuminate what benefits ma38 chine learning approaches can bring to this region.

Abstract:

Stochastic parametrization provides a methodology for representing model uncertainty in ensemble forecasts. Here we study the impact of three existing stochastic parametrizations in the ocean component of a coupled model, on forecast reliability over seasonal timescales. The relative impacts of these schemes upon the ocean mean state and ensemble spread are analyzed. The oceanic variability induced by the atmospheric forcing of the coupled system is, in most regions, the major source of ensemble spread. The largest impact on spread and bias came from the Stochastically Perturbed Parametrization Tendency (SPPT) scheme - which has proven particularly effective in the atmosphere. The key regions affected are eddy-active regions, namely the western boundary currents and the Southern Ocean where ensemble spread is increased. However, unlike its impact in the atmosphere, SPPT in the ocean did not result in a significant decrease in forecast error. Whilst there are good grounds for implementing stochastic schemes in ocean models, our results suggest that they will have to be more sophisticated. Some suggestions for next-generation stochastic schemes are made.

Abstract:

The characteristics of the diurnal, semidiurnal, and terdiurnal tides (zonal wave numbers -5 to +5 in temperature and zonal wind) in the polar mesosphere and lower thermosphere region as simulated by the extended Canadian Middle Atmosphere Model are examined. The most significant diurnal, semidiurnal, and terdiurnal tides in the polar regions are Ds0, Dw1, and De1; Sw3, Sw2, Sw1, Ss0, Se1, and Se2; and Tw3, Ts0, and Tw1, respectively, and their latitudinal structures, seasonal variations, and hemispheric asymmetries noted. Of these components, Ds0, Tw1, Ts0, and Tw3 exhibit a seasonally symmetric variation with both hemispheres strengthening simultaneously. On the other hand, Dw1 strengthens asymmetrically so that when one hemisphere is strong, the other is weak. The remainder show no seasonal tendency but vacillate on shorter than seasonal time scales in a symmetric or antisymmetric manner at different times of the year. Global-scale correlations of the amplitudes of the migrating tides Dw1/Sw2 and the stationary planetary wave 1 and the assumed “child” nonmigrating tides are also examined. The results indicate that the correlations are highly time scale-dependent and the significant correlations seen with the original time series are mainly due to longer-term variations (>18 days). There are no consistent global correlations associated with the short-term variations (<18 days) among these waves.

Abstract:

The fluctuation-dissipation theorem (FDT) has been proposed as a method of calculating the response of the earth's atmosphere to a forcing. For this problem the high dimensionality of the relevant data sets makes truncation necessary. Here we propose a method of truncation based upon the assumption that the response to a localised forcing is spatially localised, as an alternative to the standard method of choosing a number of the leading empirical orthogonal functions. For systems where this assumption holds, the response to any sufficiently small non-localised forcing may be estimated using a set of truncations that are chosen algorithmically. We test our algorithm using 36 and 72 variable versions of a stochastic Lorenz 95 system of ordinary differential equations. We find that, for long integrations, the bias in the response estimated by the FDT is reduced from ∼75% of the true response to ∼30%. © 2013 Author(s).