Quarterly Journal of the Royal Meteorological Society Q. J. R. Meteorol. Soc. 142: 147 – 159, January 2016 A DOI:10.1002/qj.2640 Improved stochastic physics schemes for global weather and climate models Claudio Sanchez, a * Keith D. Williams a and Matthew Collins b a Met Office, Exeter, UK b Exeter Climate Systems, Exeter University, UK *Correspondence to: C. Sanchez, Met Office, FitzRoy Road, Exeter, Devon EX 1 3PB, UK. E-mail: claudio.sanchez@metoffice.gov.uk This article is published with the permission of the Controller of HMSO and the Queen’s Printer for Scotland. The importance of probabilistic weather predictions and climate projections is growing. One of the key elements of the former is stochastic physics, schemes that perturb some uncertain processes in a general circulation model (GCM), such as physical parametrizations or diffusion. They help to increase the ensemble dispersion of ensemble prediction systems (EPS) and in some cases improve certain atmospheric processes by noise-induced drifts. We have developed a new configuration of stochastic physics schemes for the Met Office Unified Model (MetUM). It consists of an improved Stochastic Kinetic Energy Backscatter v2 (SKEB2), plus the Stochastic Perturbation of Tendencies (SPT). The improvements to SKEB2 remove spurious physical artefacts, e.g. a spurious wave caused by low-wave-number perturbations, and improve the resolution sensitivity of the scheme. The SPT produces a larger ensemble spread in the Tropics than present schemes, but its impact on long-term climate budgets makes the use of conservation constraints for water vapour and energy essential. The new configuration produces a higher impact in the Tropics, increasing the ensemble spread and improving some long-standing climate biases in areas of excessive convection, whilst minimizing the negative impact on tropical processes like tropical convective waves. Key Words: stochastic physics; ensemble prediction; climate modelling; tropical biases Received 2 March 2015; Revised 6 July 2015; Accepted 17 July 2015; Published online in Wiley Online Library 18 September 2015 1. Introduction Prediction of the evolution of atmospheric flow is one of the most important and challenging problems of our times. An accurate prediction of devastating phenomena like hurricanes or runaway effects in the climate system is shaped by a myriad of processes, the intricate interactions of which encompass a wide range of temporal and spatial scales. The tool to provide these predictions is the general circulation model of the atmosphere (GCM). Their progress has been dramatic in recent decades (Simmons and Hollingsworth, 2002; Reichler and Kim, 2008). However, they still have major deficiencies, e.g. poor simulation of the Madden–Julian Oscillation (MJO: Zhang, 2005). GCMs contain many assumptions and crude approximations. Probably the most important is the representation of the bulk effects of processes below the truncation scale, so-called parametrizations. In many cases, the number of subgrid events per grid box is not large enough to permit the existence of a meaningful statistical equilibrium (Williams, 2005); furthermore, the energy spectrum indicates that there is no true separation of scales in the atmosphere (Nastrom and Gage, 1985). In addition to parametrizations, the dynamical core also employs some artificial approaches like numerical diffusion to prevent computational instabilities (Shutts, 2013). The uncertainties of a forecast are represented by ensemble prediction systems (EPS). This system produces several predictions for the same forecast. Differences amongst ensemble members, so-called ensemble spread, may emerge from various sources, such as perturbations to the initial conditions, the use of different parametrizations or even different models and, last but not least, stochastic perturbations that represent uncertain processes. The popularity and importance of EPS are growing as they are employed operationally to produce probabilistic forecasts of harmful extreme events (Hamill et al., 2012; Neal et al., 2013). The usefulness of EPS is currently limited, as they cannot in general produce enough ensemble spread (Buizza et al., 2005), thus creating a less reliable ensemble. The lack of spread is associated with the lack of multiscale coupling between resolved and parametrized scales (Palmer et al., 2005). In an EPS context, the subgrid effects on large scales should be represented by a probabilistic value, rather than the most likely value provided by the current deterministic parametrizations. Including stochastic perturbations to the ensemble could simulate some of the unrepresented subgrid variability, an idea that has been widely discussed (Palmer, 2001; Palmer et al., 2005; Williams, c  2015 Crown Copyright, Met Office. Quarterly Journal of the Royal Meteorological Society c  2015 Royal Meteorological Society