Model Improvement via Systematic Investigation of Physics Tendencies

Deficiencies in forecast models commonly stem from inadequate representation of physical processes; yet, improvement to any single physics component within a model may lead to degradations in other physics components or the model as a whole. In this study, a systematic investigation of physics tendencies is demonstrated to help identify and correct compensating sources of model biases. The model improvement process is illustrated by addressing a commonly known issue in warm-season rainfall forecasts from parameterized convection models: the misrepresentation of the diurnal precipitation cycle over land, especially in its timing. Recent advances in closure assumptions in mass-flux cumulus schemes have made remarkable improvements in this respect. Here, we investigate these improvements in the representation of the diurnal precipitation cycle for a spring period over the United States, and how changes to the cumulus scheme impact the model climate and the behavior of other physics schemes. The modified cumulus scheme improves both the timing of the diurnal precipitation cycle and reduces midtropospheric temperature and moisture biases. However, larger temperature and moisture biases are found in the boundary layer as compared to a predecessor scheme, along with an overamplification of the diurnal precipitation cycle, relative to observations. Guided by a tendency analysis, we find that biases in the diurnal amplitude of the precipitation cycle in our simulations, along with temperature and moisture biases in the boundary layer, originate from the land surface model.