A note on the dependence of atmospheric predictability on baroclinic development

✍ Scribed by P. J. Everson; D. R. Davies

• Publisher: John Wiley and Sons
• Year: 1971
• Tongue: English
• Weight: 270 KB
• Volume: 97
• Category: Article
• ISSN: 0035-9009
• DOI: 10.1002/qj.49709741207

✦ Synopsis

Abstract

A two‐level, quasi‐geostrophic model, described by Everson and Davies (1970), is used to study the dependence of duration of predictability (as defined by, e.g., Charney et al. 1966) on the phase, at some initial time in the integration, of global scale baroclinic development (expressed in the model in terms of fluctuations in eddy kinetic energy, K').

Using firstly an invariable subgrid scale coefficient in a control run of 120 days, sinusoidal perturbations, wave length approximately 6,000 km, of the geopotential heights are inserted at all the grid points at (a) day 72, corresponding to a maximum in model K', and (b) day 84, corresponding to a minimum in model K'. The consequent ‘perturbed’ integrations are compared to the control values by taking the r.m.s. of the difference in the 500 mb calculated temperature distributions, and determining the time taken for this to increase to the ‘persistence’ value, i.e. the maximum r.m.s. difference between two randomly selected model states. The doubling time of error in (a) was seen to be 12 days leading to a model predictability of 24 days, and in (b) it was seen to be 17 days, leading to a predictability of 34 days: these values are related to an initial perturbation amplitude of 1°C.

These experiments were then repeated using a variable subgrid scale coefficient, depending on grid scale horizontal temperature gradient as described by Everson and Davies (1970). The results were; (a) working from a day of maximum K', the doubling time of error was reduced to 8 days, approaching the value obtained by primitive equation numerical models and corresponding to a predictability of 16 days; (b) working from a day of minimum K', the doubling time of error was 13 days, corresponding to a predictability of 26 days. The results in both cases show that predictability, as calculated from some initial phase of baroclinic development, is 50 per cent higher as measured from a maximum in K'compared with a minimum in K'. It also drops sharply with increase in the degree of model sophistication (expressed in the system discussed in this paper as variability of the subgrid scale coefficient).