Developments in the ECMWF humidity background errors

Elias Holm
  3 Oct, Noon, in 2155

The background error used in 4D-Var and the EDA has unbalanced relative humidity as control variable, with analytic balance operator between humidity and temperature in cloudy conditions. Recent developments have added humidity background error variances taken directly from the Ensemble of Data Assimilations (EDA), like for all other variables. This replaces earlier background and level dependent statistical estimates of the humidity variances. This change improved forecast scores and background fit to observations, including winds, and went operational 11 July 2017. I will show some selected results from this change and discuss some further developments. These include single observation experiments with a new analytic balance operator coupling humidity and dynamic background errors and the sensitivity of the analysis to using stratospheric humidity variances from the EDA versus tapering them to very small values, as done operationally.