Some idea of the relative magnitude of sources of error in predictions of wind storms on the Canterbury Plains, New Zealand, is gleaned by perturbing numerical weather prediction (NWP) model simulations in two cases where moderately strong winds were observed. This complements a previous study of uncertainty due to local effects (Purnell and McGavin 1999). Under 45 per cent of the total variance in gust maximum aerodynamic load and speed appears to be predictable several days ahead, rising to a limit under 60 per cent on the day of the event. In other words, the results suggest there would be little scope for improvements in any prediction service that was already nearly this accurate. There is definitely some potential to improve the NWP model used here to provide predictions of maximum gusts via the statistical model of Part 1. The most significant predictors were the pressure-difference across the Southern Alps and its variation, but the model values of these were little better than half their observed magnitudes.
D.K. Purnell (Mon,) studied this question.