A machine learning approach for supporting marine data quality control.
SalaciaML is a Deep Learning algorithm developed by Dr. Serder Demirel and Dr. Sebastian Mieruch-Schnülle at the Alfred-Wegener-Institut (AWI) in Bremerhaven, Germany. SalaciaML supports ocean scientists in detecting erroneous measurements. The algorithm is based on the Keras library.
Photo by Florian Weihmann from Pexels
Marine data quality control is heavily performed manually/visually by ocean experts. Regarding the increasing size of the data, algorithmic support is urgently needed.
Photo by Pauline Paul from Pexels
The Name
SalaciaML is named after Salacia (/səˈleɪʃə/ sə-LAY-shə, Latin: [saˈlaːkɪ.a]), the roman goddess of the sea waters.
Photo by Pauline Paul from Pexels
Dresden Hackathon
At the Dresden Deep Learning Hackathon in September 2019, Serdar and Sebastian together with Steffen, were honored with the prize for the "most accurate team". Foto: Steffen Seitz (TU Dresden), Sebastian Mieruch-Schnülle (Alfred-Wegener-Institut), Serdar Demirel (Alfred-Wegener-Institut)
A paper on our work has been published at Frontiers of Marine Sciences: This work would have not been possible without the help and support of many colleagues and friends. We like to acknowledge Prof. Reiner Schlitzer, Jörg Matthes, Peter Steinbach, Steffen Seitz, Axel Behrendt, Simona Simoncelli and the SeaDataCloud consortium.
The data
The training and evaluation of our algorithm SalaciaML is based on ocean temperature profiles provided by the SeaDataNet infrastructure, which manages ocean data from more than 100 data centers in Europe. The large data collections include approx. 2 million temperature and salinity datasets, which contain approx. 9 million ocean profiles.
SalaciaML can detect erroneous ocean data. The figure shows temperature profiles from the south eastern Mediterranean Sea. Blue dots indicate "good" data, whereas red dots highlight potentially "bad" data and give scientists hints for looking closer on these measurements.