Statistical post-processing of weather forecasts and why crowd-sourced observations are a game-changer

Speaker: Kate Saunders
Affiliation: Queensland University of Technology

Abstract

To mitigate the impacts of extreme events we need reliable weather forecasts and timely warnings. Forecasts from numeric weather models however contain errors, such as in bias and dispersion, as numerical weather models approximate the complexity of the real world. To improve the forecast skill, we use statistical post-processing methods to adjust the forecasts relative to observations. For compound events, assessing the forecast skill of different post-processing methods is not trivial. Compound events involve complex dependence interactions between meteorological variables. These interactions can occur spatially, temporally, between different variables and combinations thereof. It is therefore important to assess the forecast skill relative to the driving mechanisms of that compound event, and relative to how mitigating decisions are made using that forecast. In the first half of this talk, using an example of rainfall and storm surge, I will share improvements in statistical post-processing and scoring methods for compound events. I will then show that through my work developing quality control methods for crowd-sourced wind observations, that there is further scope to improve our forecasts of extreme events. What this means for Australia is that crowd-sourced observations could be a game-changer for building resilience in the face of worsening bushfire seasons.