A New Method for the Estimation of Analysis and Forecast Error Variance

We define errors in Numerical Weather Prediction (NWP) analyses or forecasts as the difference between the NWP fields and the state of the natural system interpolated onto the model grid. In the evaluation of NWP forecasts, observations or analysis fields are used as proxy for reality. Since the actual error in either the observations or analysis fields is unknown, a level of ambiguity permeates all, and especially analysis and short range forecast error assessments.

While the actual errors in analyses and observations are unknown, in recent studies we explored an alternative Statistical Analysis and Forecast Error estimation method (SAFE, Pena and Toth 2014, Feng et al. 2017) for the unbiased estimation of analysis and short range forecast error variance. The method uses measurements of perceived error variance between forecasts and verifying analysis fields, along with basic assumptions about the behavior and growth of errors in the data assimilation - forecast system (DAFS).

In this presentation, SAFE will first be demonstrated in simulated environments with simple and intermediate complexity dynamical systems. Results from real world DAFS will also be reviewed, including a spatially extended assessment of the NCEP DAFS, a comparison with metrics for other leading NWP centers, and an exploration of the role of fast growing and random errors in DAFS systems.