Optimal network design applied to monitoring and forecasting surface temperature in Antarctica

Abstract

As harsh weather conditions in Antarctica make it difficult to support a dense weather observing network there, it is critical to place new weather stations in locations that are optimal for a given monitoring goal. Here we demonstrate a network design algorithm that uses ensemble sensitivity to identify optimal locations for new automatic weather stations in Antarctica. We define the optimal location as one that maximizes the reduction in total variance of a given spatial field. Using WRF Model forecast output from the Antarctic Mesoscale Prediction System (AMPS), we identify the best locations for observations across the continent by considering two spatial fields: (i) the daily 0000 UTC 2-m temperature analysis field and (ii) the daily 0000 UTC 2-m air temperature 24-h forecast field. We explore the impact of spatial localization on the results, finding that a covariance length scale of 3000 km is appropriate for these metrics. We find optimal locations assuming that no stations exist on the continent (blank slate) and conditional on existing stations (CD90). In the "blank slate" scenario, the Megadunes region emerges as the most important location to both monitor temperature and reduce temperature forecast errors, with the Ronne Coast and the Siple Coast following. Results for the monitoring and forecasting metrics are similar for the CD90 subset as well, indicating that additional stations could benefit multiple performance goals. Considering the CD90 subset, Wilkes Land-Adelie Coast, Ellsworth Land, and Queen Maud Land-Interior are identified as regions to consider installing new stations for optimizing network performance.