Observations and prior uncertainties
Both in-situ and remote sensing observations are used to constraint the model simulation. Data sets that are assimilated and their sources are listed in Table 1.
SST is based on optimal interpolated microwave and infrared data. SST data is not available over the sea ice-covered region, but we assume SST is at freezing temperature (-1.96 °C) where sea ice is observed but not simulated. Along-track sea level anomaly observations from satellite altimetry are available over ice-free regions. With specific algorithms, sea level could be retrieved over the sea ice-covered areas with significantly larger errors (
Armitage et al., 2016;
Rose et al., 2019). Additionally, the model is constrained to the WOA18 climatology and the mean dynamic height from
Rose et al. (2019). Hydrographic profiles from EN4 data (
Good et al., 2013) and UDASH data (
Behrendt et al., 2018) are used, while duplicated data are removed. On average, the Atlantic sector of the pan-Arctic Ocean is observed more than once per month on a 3°´3° box during April-October (Figure 1a) while once per two months during November-March (Figure 1b). However, the Arctic Ocean is severely under-sampled (Figure 1a, b). In the vertical, the profiles cover mainly the top 800 m (Figure 1c) and more observations are available in the summer season than in the winter season.
Uncertainties of along-track SLA, temperature and salinity profiles, temperature and salinity atlas are the same as in
Köhl (2015). Uncertainties of SST consist of errors from the data and representative errors based on the method of
Oke and Sakov (2008). Uncertainties of SIT and mapped SLA over sea ice-covered regions are provided by the data sets. SIC uncertainties are assumed to be geographic dependent and are computed using the method of
Fenty and Heimbach (2013). Mean dynamic height uncertainties were set to 1 cm. Since the data of the temperature and salinity climatology are interpolated to the finer grid, thereby inventing additional data points, we reduced the weight of the temperature and salinity climatology cost component by a factor of 10 and 50, respectively. Due to the low number of hydrographic profiles, we increased the weighting of the hydrological profiles component by a factor of 10 to increase their relative importance with limited iterations.