Abstract/Summary

Predicting the weather accurately at increasingly long lead times is incredibly useful, and relies on numerical models realistically representing the atmosphere. Here the character, cause and consequence of systematic biases in humidity and temperature are investigated as improved knowledge of these biases can help ameliorate them. Through comparison to in-situ observations, the vertical structures of an analysis moist bias and forecast cold bias in the extratropical lowermost stratosphere are determined. The moist bias has a maximum of around 170% of observed values 1km above the tropopause, and is the dominant cause of the cold bias through additional longwave cooling. Correction of the initial state is insufficient to correct forecasts, as the moisture bias is reintroduced with half-life ≈ 8 − 9 days through a combination of vertical diffusion and advection. Despite the importance of diffusion and advection the re-moistening has negligible dependence on horizontal resolution, suggesting poor resolution of processes in the vertical is responsible. Through comparison of corrected ensemble forecasts to a control, it is shown that a bias correction to the moisture field seen by the radiation scheme leads to significant skill increases at around three weeks, particularly over the North Atlantic and Europe, and a reduction in seasonal average error globally. The proposed mechanism for these results is via improvement of the representation of the tropopause, and consequently Rossby wave propagation along it and the jet stream. When initially corrected, the moist bias is reintroduced by the model on a shorter timescale than that at which we see significant forecast improvements; therefore it is necessary to correct the bias both in the model and in the analysis. Modern radiosondes have value above the tropopause if their data can be assimilated. To reduce forecast error, further work is required to improve vertical diffusion and advection, through increased vertical resolution or otherwise.