Helen Buttery, Met Office

Investigations into the use of the FSO method with the Met Office’s 1.5-km UKV model

Tim Payne


The FSO (Forecast Sensitivity to Observations) method is a way of calculating the impact of observations on the forecast. It uses information from the data assimilation system - specifically the linear Perturbation Forecast (PF) model and its adjoint - to calculate the impact of each observation simultaneously. This differs from traditional observation impact assessments using data denial experiments (OSE – Observing System Experiments) because these require a new trial for each data type. The FSO method has been used successfully and run operationally at the Met Office in the global model. However, extending this to a convective-scale limited-area model (such as the Met Office's 1.5km UKV), is not straightforward. The PF model can only provide a valid evolution of perturbations to the forecast for much smaller timescales in the case of the UKV than for the global model because of the presence of small-scale highly non-linear processes such as convection. In order for the FSO to be run successfully, the PF model must have small linearisation errors for at least 3 hours (the time between two adjacent model runs). Initial tests with the UKV showed the linearisation error would remain small enough in about 40% of cases, suggesting that it could be a useful tool to assess observations but not an operational system. Recent improvements to the PF model code to deal with an instability in its microphysics scheme have increased the number of useful cases slightly. These results will be presented. In addition, for the UKV an analysis-based total energy forecast error norm cannot be used because the analysis used would be too close to the time of the original forecast and therefore not independent. An observations-based forecast error norm can be used instead. Investigations into this will also be presented.