Abstract:
|
Over the past four decades, the Arctic ice sheet has been steadily shrinking, with satellite observations indicating downward trends in sea ice extent across all months. These changes have stimulated demand for accurate estimates and forecasts of sea ice thickness. We propose a spatiotemporal geostatistical model for generating probabilistic forecasts of sea ice thickness. The new approach uses a Gaussian process model for sea ice thickness, to generate forecasts on subseasonal to seasonal time scales, at locations throughout the Arctic. The new forecasts improve upon existing physical model ensemble forecasts and simple statistical reference forecasts, as evidenced by reductions in root mean squared error and improvements in empirical coverage rates. The new approach also provides estimates of the spatial and temporal correlations in forecast errors, which allows for probabilistic forecasts of aggregate sea ice metrics such as sea ice volume.
|