Investigating atmospheric predictability on Mars using breeding vectors in a general‐circulation model

Abstract A breeding vectors approach is used to investigate the hypothesis that the Martian atmosphere is predictable at certain times of year, by identifying the fastest‐growing modes of instability at different times in a Mars general‐circulation model. Results indicate that the period from northern mid‐spring until mid‐autumn is remarkably predictable, with negative global growth rates for a range of conditions, in contrast to the situation on the earth. From northern late autumn to early spring growing modes do occur, peaking in northern high latitudes and near winter solstice. Reducing the size of the initial perturbations increases global growth rates in most cases, supporting the idea that instabilities which saturate nonlinearly at lower amplitudes have generally faster growth rates. In late autumn/early winter the fastest‐growing modes (‘bred vectors’) are around the north pole, increase with dust loading, and probably grow via barotropic as well as baroclinic energy conversion. In northern late winter/early spring the bred vectors are around the north pole and are strongly baroclinic in nature. As dust loading (and with it the global circulation strength) is increased their growth rates first decrease, as the baroclinic mode is suppressed, then increase again as the fastest‐growing instabilities switch to being those which dominated earlier in the year. If dust levels are very low during late northern autumn (late southern spring) then baroclinic modes are also found around the spring pole in the south, though for a slight increase in dust loading the dominant modes shift back to northern high latitudes. The bred vectors are also used as perturbations to the initial conditions for ensemble simulations. One possible application within the Mars model is as a means of identifying regions and times when dust‐lifting activity (related to surface wind stress) might show significant interannual variability for a given model configuration, without the need to perform long, computationally expensive multi‐year model runs with each new set‐up. This is tested for a time of year when previous multi‐year experiments showed significant variability in dust storm onset in the region north of Chryse. Despite the model having no feedbacks between dust lifting and atmospheric state (unlike the original multi‐year run), the ensemble members still show maximum divergence in this region in terms of near‐surface wind stress, suggesting both that this application deserves further testing, and that the intrinsic atmospheric variability alone may be important in producing interannual variability in this storm type. Copyright © 2004 Royal Meteorological Society