Chouai, Mohamed; Mieruch-Schnülle, Sebastian; Behrendt, Axel; Vredenborg, Myriel; Rabe, Benjamin (2024): UDASH-AI: Unified Database for Arctic and Subarctic Hydrography Optimized for Artificial Intelligence Applications [dataset]. PANGAEA, https://doi.org/10.1594/PANGAEA.973235

UDASH-AI represents an updated version of the UDASH dataset, that has been created to develop an artificial intelligence algorithm, that we name SalaciaML-Arctic to support the visual/human quality control of the data. UDASH-AI can be directly used with our algorithm, provided under the DOI: https://doi.org/10.5281/zenodo.11535790 and the respective GitHub repository, to reproduce our results, extend the methods and more. Additionally, we have implemented SalaciaML-Arctic as an user-friendly app at https://mvre.autoqc.cloud.awi.de. Following steps have been applied on the original UDASH dataset to create UDASH-AI:

• Concatenation of the single, annual txt files into one single csv file.

• The original encoding of missing time and day information in the date/time string as 'T99:99' and '-00T' have been changed to ISO8601 conformity: 'T00:00' and '-01T'. To not loose this information we have added a quality flag ('QF_time') in the column next to the date/time with following encoding:

◦ 0: No missing data (good quality).

◦ 1: Missing day.

◦ 2: Missing time.

◦ 3: Missing day and time.

• Further we have included the two temperature gradients and the density gradient described in the original UDASH paper as extra columns:

◦ Depth over temperature gradient, denoted as 'd/d_Temp_Depth_[m_°C^-1]'.

◦ Temperature over depth gradient, denoted as 'd/d_Depth_Temp_[°C_m^-1]'.

◦ Density gradient as 'd/d_Depth_Dens_[kg_m^-4]'.

• The missing value is marked with an indicator: NaN.

• We added the quality flags for temperature from the classical/traditional UDASH automatic checks: outlier and spike (flag=4), density inversion (flag=3) and suspect gradient (flag=2) as an extra column named 'QF_trad'.