Process relationships useful for evaluating model performance in the Arctic

Model validation with observational data is frequently done through comparisons of horizontal fields and time series of state parameters using point or mean statistical parameters such as bias, root-mean-square error, correlation coefficients, and Index of Agreement. Frequently, however, such evaluations are unsatisfying if one is trying to understand whether a model is capable of representing key processes, and hence capable of producing correct responses if the forcing is changed. Frequently, compensating errors may produce good statistical results despite poor representation of the processes; similarly, improvements affecting one process may lead to worse statistics despite a better representation of the physics. In such situations, analyses of the model must be done differently.

Different environments have different key processes, which may be related to terrain, the proximity to large gradients in surface characteristics, diurnal or seasonal cycles. The near-surface atmospheric conditions over the Arctic sea ice, which is a dominant surface feature north of 65° N, is unique and is dominated by a few key processes. These processes are related to radiative forcing by clouds, surface turbulent heat flux governed by the cloud and surface thermal and radiative properties, the highly stable atmosphere, the general suppression (though not a total lack of) of a diurnal cycle, and the forcing of surface conditions by synoptic and mesoscale atmospheric disturbances. When evaluating a model, a few key relationships evident in the observational data can be used to evaluate the model's ability to reproduce the processes. One example is the bimodal distribution of wintertime net longwave radiation, and its forcing of turbulent sensible heat flux. Another is the dominance of the reduction of turbulence over the increased vertical temperature inversion for reducing the magnitude of sensible heat flux allowing the Arctic surface layer to become very stable in wintertime conditions.

Additional examples of such relationships will be given using observational data from SHEBA, AOE-2001, and SEARCH observatories. The processes producing these relationships will be discussed, and the presence of these relationships is examined in model-based data sets such as ERA-40 and simulations using the MM5 and WRF mesoscale models and other models.