Performance metrics and skill assessment methods demonstrated using numerical model results for the Gulf of Mexico

Specific formulations for performance metrics to be used to compare model fields with various types of observation are proposed, including metrics that are appropriate for current meter data, hydrographic cast data, Lagrangian drifter data and satellite sea surface height and sea surface temperature data. The key element in the formulation of each of these metrics is the recognition of the fact that the salient features of the actual circulation may exist in the model fields but not always in the correct geographical location. Therefore, the proposed metrics record both the accuracy with which the model reproduces the feature, in the form of a correlation, and the relative location of the model simulated feature, in the form of a displacement. The result is considerably more informative and useful than the conventional comparison where a correlation between the data and the model field at the same geographical location is shown to be small and, therefore, the skill of the model is judged to be low. Also addressed is the essential difference between the data from in situ observations and variable fields calculated using a numerical model. An approximate method of treating the in situ data is proposed to make the comparison between in situ data and model results more meaningful.