A Decision Tree Algorithm for Investigation of Model Biases Related to Dynamical Cores and Physical Parameterizations

Climate models are conceived in terms of the resolved fluid dynamics (i.e. the dynamical core) and subgrid, unresolved physics represented by parameterizations. In this study, we focus on analyzing how the choice of dynamical core impacts the representation of precipitation in the Pacific Northwest of the United States, Western Canada, and Alaska. Spectral and finite volume dynamical cores are considered within the Community Atmosphere Model. We develop model evaluation strategies that identify like “objects” – coherent systems with an associated set of measurable parameters. This makes it possible to evaluate processes and identify the sources of uncertainty in models without needing to reproduce the time and location of, for example, a particular observed cloud system.

In this talk we describe a decision tree algorithm to identify objects. In previous studies the object identification was achieved via setting one or more threshold values and checking if the grid value precipitation amount is below or above that threshold. That approach often led to misclassification, as the magnitudes of objects are more variable when compared to the limited threshold values. We aim to overcome this misclassification problem by employing a decision tree algorithm which recursively partitions the initial precipitation field by setting a different threshold value between each partition. The algorithm calculates simple statistics before partitioning each node and makes a decision of the new threshold to be applied. This dynamic thresholding approach leads to a more generic applicability of our object-based method, and can be applied without extensive subjective input. We show that this identifies significant differences between the finite-volume and spectral dynamical cores in the patterns of orographic precipitation.