Sea ice thickness and volume
SIT differences remain large among different ocean-sea ice reanalyses (
Chevallier et al., 2017), which may be caused by differences in sea ice models and how SIT is updated when ingesting observations by data assimilation. In this section, we analyze improvements to SIT due to the adjustment of the control variables and compare them with the TOPAZ4 and PIOMAS reanalyses.
The cost of SIT is reduced by ~60% in the three chunks (Figure 2), despite its very little contribution to the total cost. INTAROS-opt (Figure 5b) underestimates mean SIT in the central Arctic Ocean and in the region north of Greenland, which is covered by multiple-year sea ice, and overestimates mean SIT over seasonal sea ice extent regions. INTAROS-opt reduces mean SIT errors significantly (Figure 5c), and the root mean square error (RMSE) is reduced from 0.54 m in INTAROS-ctrl to 0.40 m in INTAROS-opt (Figure 5a).
The PIOMAS reanalysis shows a slightly larger RMSE of SIT (0.46 m) than INTAROS-opt (0.40 m) and TOPAZ4 (0.41 m, see Figure 5a). TOPAZ4 shows larger errors in October and November than both INTAROS-opt and PIOMAS, but the SIT errors are quickly reduced as the model assimilates observations sequentially, resulting in smaller RMSE than the other two products at the start of the next year. Mean SIT errors remain in the three products. Negative SIT errors up to -0.6 m exist in the central Arctic Ocean and the Eurasian Basin, extending to the northeastern Greenland coast. In the Beaufort Sea, all three reanalyses overestimate SIT by ~0.2-0.4 m with TOPAZ4 performing best. In the seasonal sea ice extent regions, including the marginal seas and around Greenland, TOPAZ4 data shows smaller mean SIT errors than the other two reanalyses.