Evaluating the Antarctic Observational Network with the Antarctic Mesoscale Prediction System (AMPS)

Observing networks vary vastly in spatial coverage, and the appropriate level of coverage needed to capture weather and climate variability is often an open question. In Antarctica, large parts of the continent are unknown in terms of weather and climate variability. The Antarctic Mesoscale Prediction System (AMPS) has utilized the Polar Weather Research and Forecasting (WRF) model since 2008 to provide forecasts over Antarctica. Here we evaluate the current Antarctic network, using temperature and surface pressure, and perform tests on how well the observations are represented by AMPS by varying time scales, regions, and seasons. We do this by comparing the archived surface observations to the archived 15 km AMPS grids from October 1, 2008 through October 31, 2012. As a validation test, we evaluate the spatial correlations represented by AMPS to those in the observations, and find good agreement.

We evaluate spatial coverage, as indicated by variance explained by AMPS grid cells corresponding to observing locations, and perform the analysis separately for East Antarctica, West Antarctica, and the Antarctic Peninsula. Preliminary results for the whole continent show that the temperature correlation length scales are larger in winter (April-September) when compared to summer (October-March), and that the opposite is true for surface pressure. Observing stations vary in the amount variance they explain throughout the continent, especially for temperature. Some have a large spatial influence but the signal is not isotropic. We find that there are large gaps in spatial coverage at daily time scales, in particular for temperature observations. Although still incomplete, coverage improves substantially at longer time scales. In addition to identifying gaps in the coverage by the existing network, we conclude by identifying the specific location where one new observation would add the most value to the network. A companion abstract (Hryniw et al.) describes data denial experiments that are used to evaluate the value of existing stations.