Increasing model spatial resolution fails to reduce simulated storm biases

Science

Accurately predicting impacts from storms depends on accurately simulating their growth as a function of atmospheric conditions. Using a model setup like those used for operational forecasting, results show that total storm rainfall over a large area is reasonably predicted. However, heavy rain rates were too frequent and light rain rates were too infrequent at a local scale when compared to observations, meaning the balance between rainfall frequency and intensity is incorrectly predicted. This is caused by an excessive number of simulated storms, a model bias that worsens as the atmosphere becomes more stable. Increasing model resolution to better resolve storm circulations does not reduce these biases, indicating model representation of precipitation formation and growth in storms requires improvement.

Impact

Previous research shows that models predict too much heavy rainfall and not enough light rainfall in storm systems, which undermines their application to predicting storm impacts. It is hypothesized that a contributor to this bias is insufficiently resolved storm circulations, but this study shows that this is not the case. Precipitation forms and grows too easily regardless of model spatial resolution, leading to an excessive number of relatively shallow precipitating storm cells that do not contribute to the growth of widespread light precipitation. This work directs future research to evaluate and better represent precipitation formation and growth. Accurate representation of storms and their rainfall is important for improving weather and climate predictions.

Summary

The ability of a storm-resolving weather model to predict the growth of storms over central Argentina was evaluated with data from the Clouds, Aerosols, and Complex Terrain Interactions (CACTI) field campaign in central Argentina. Although the model accurately predicts the total amount of rain, it produces too much relatively heavy rainfall and not enough light rainfall, a bias seen in many previous studies. This research uniquely showed that the overestimation of heavy rainfall is caused by more than twice as many predicted storm cells as observed despite similar observed and simulated cell growth processes. The excessive frequency of storm cells is most prominent for relatively shallow cells that prevent the formation of widespread lighter rainfall that was much more frequent in observations. This bias was also shown to worsen as the atmosphere became less favorable for intense storms and relatively light rainfall contributes more to total rainfall. Unexpectedly, increasing the spatial resolution of the model to better resolve storm circulations did not improve predictions. This suggests that model representation of storm precipitation formation and growth processes requires improvement beyond model resolution to better predict storm rainfall intensities.

Previous studies have shown overestimated heavy and underestimated light rainfall in simulated storms to be a common bias across a variety of atmospheric conditions and models. This indicates that the results of this study potentially represent a widespread storm precipitation prediction problem that cannot be solved by simply increasing model resolution. Future work can confirm whether this is the case or not by extending analyses to additional model configurations and locations with contrasting storm environments and requisite measurements.