Meteorologische Zeitschrift (Sep 2018)
Temperature and sea ice hindcast skill of the MiKlip decadal prediction system in the Arctic
Abstract
In this study, hindcast skill for near-surface air temperature (TAS), sea surface temperature (SST), sea ice concentration, and sea ice area is assessed for the Arctic region using decadal simulations with the MiKlip decadal prototype prediction system. The prototype MiKlip system is based on the low-resolution version of the MPI‑ESM model. In the simulations, a full field initialization of atmospheric and oceanic variables was used, but sea ice was not initialized. The hypothesis is that the increase in hindcast skill due to initialization found for TAS and SST in the North Atlantic in the prototype system compared to the historical simulations leads to enhanced skill also in the Arctic. However, the skill enhancement compared to the uninitialized experiments in the Arctic is generally weak. The hindcast skill only increases for SST and sea ice concentration along the east coast of Greenland and in the Fram Strait in lead years 2–5. Initialization additionally improves the skill in regionally integrated sea ice area (detrended) in the Greenland Sea, but only in lead year 1 and only in winter, and not in other Arctic regions. In order to assess whether additional initialization of sea ice concentration improves skill, we also analyse hindcasts and historical simulations performed with the MiKlip preoperational system that is based on the high-resolution version of the MPI‑ESM. These simulations have nonetheless a negative bias in sea ice area in late summer of 1 to 3 million km2. Noting that this is a much smaller ensemble than for the prototype system, the hindcast skill in North Atlantic TAS and SSTs is significantly reduced and not present when evaluated against ERA-Interim instead of HadCRUT4 data. Accordingly, in the Arctic, no additional skill compared to the prototype hindcasts is found. Our results underline the importance to assess the robustness of skill with different observational datasets and metrics. For future MiKlip simulations, we recommend to additionally initialize sea ice thickness or age, and to initialize the simulations in a different month to potentially enhance sea ice skill in the Arctic.
Keywords
