Abstract
Because many forecasting applications rely heavily on statistical parameterizations of poorly understood processes, research results may lead to forecast improvement without a concomitant improvement in physical understanding. Carefully defining research goals as either forecast improvement or physical understanding is an important step toward interpreting results and designing studies. To approach an optimal estimate, we must seek ways to reduce model error, the impact of simplifying assumptions, and sensitivity to the chosen norm. The scientific community will no doubt continue improving models by reducing deterministic model error, but gaps will always exist in our understanding and computational capabilities.
