Impact of data resolution on meteorological fields over the polar regions

With respect to numerical weather and climate prediction models, the polar regions provide particular challenges including a relative lack of observational data and, for some types of models, numerical effects related to the convergence of meridians at the pole. A combined result of these aspects of the polar regions is that the relative significance of individual data points is greater than elsewhere on the globe. As such the representativeness error characteristics for such sparse polar observations become very important and must be carefully considered if meaningful data assimilation is to take place. We consider such error characteristics in this paper.

In this study, we utilize a spectral approach within a data assimilation framework to estimate valid spatial and temporal scales of the representativeness error structures as well as the sensitivity of the potential errors to various resolutions of the observations and model grid. In our approach, virtual observation data (the so-called “true fields) are generated through a simulation with the PSU/NCAR MM5 modeling system. White noise is then added to the “true" fields, and then certain points are removed to create a sparser “degraded” network of observations.

Results indicate that the largest scale field structures are correctly described even with the “degraded” observation network. However, at finer scales there are sufficient deviations from the “true” solution using the “degraded” observation network; the divergence in model solution begins almost immediately after the start of the simulation. Further, an effort to tune the model variables to the “degraded” observation locations leads to a disturbance of the fields by introducing new artificial sub-structures that do not dissipate and which arise from a combination of out-of-balance conditions and discrepancies in the local lower boundary conditions.

We conclude that in certain areas, such as the Polar region, where observation data is generally sufficiently sparser in spatial scale than either the model horizontal and vertical grid resolutions, data assimilation procedures must be properly designed and tuned so as not to overcorrect the spatial and temporal structures of meteorological fields at observation locations and thereby contribute to the drift of the model realization. This fact should be kept strongly in mind when developing an appropriate reanalysis for the Arctic regions.