Ross Bannister, National Centre for Earth Observation
Unexpected background error statistics of convective-scale atmospheric models
Abstract:
Data assimilation is often used to help make sense of observations made from the traditional meteorological observing systems, from new radar networks, and from space. Effective data assimilation relies on accurate uncertainty statistics of all the data that are used, including those of observations and from the numerical model's forecasts used as the prior (background) state.
The background error covariance scheme used in the Met Office's current variational data assimilation system is designed to represent errors in large-scale processes, even though many of the models that it serves allow small-scale (convective-scale) process. The work reported here looks at how well the Met Office's current scheme copes with fine-scale features of numerical models and reveals some unexpected characteristics.