Leveraging Highly Accurate Data in Diagnosing Errors in Atmospheric Models

Abstract

curate data can serve the numerical weather prediction, climate prediction, and atmospheric reanalysis communities by better enabling the diagnosis of model error through the careful examination of the diagnostics of data assimilation, especially the firstguess departures and the analysis increments. The highly accurate data require no bias correction for instrument error, leaving the possibility of confusion with error in forward models for observations as the lone hindrance to the diagnosis of model error. With this scenario in mind, we conducted numerical experiments to investigate the potential confusion using the data assimilation system at the European Centre for Medium-Range Weather Forecasts. We found that large-scale systematic model error can be misattributed to error in the forward models for observations, thereby reducing systematic firstguess departures and impeding the mitigation of model error. The same large-scale model error generated a 20% increase in analyzed specific humidity near the tropopause, suggesting that current observational data cannot constrain the upper tropospheric humidity in current models, which contributes substantially to greenhouse forcing of the climate. We expect that the confusion of model error for an error in the forward models for observations occurs regardless of the objective method used to diagnose model error. Our findings underline the importance for continued improvement in radiative transfer calculations and highlight the value of multiple sources of accurate data that are redundant in their sensitivity to atmospheric variables yet orthogonal in their radiation physics.