Naldmann et al.
tests vary depending on the scientific party providing the data.
Additionally, definitions for the used codes vary, although there
are a number of similarities between the used conventions. For
example, flagging schemes based on OceanSITES (OceanSITES,
2020), ARGO (Wong et al., 2022), Copernicus (Copernicus,
2020), and SeaDataNet (2010) follow the convention that “no
quality test performed” is defined as flag=“0”, while schemes
based on GO-SHIP (Swift, 2010) use flag=1 and (UNESCO,
2013) and I0OS (Bushnell, 2020) use flag=2 for the same. This
causes major efforts regarding the mapping of flag information
between the data providers. Interpretation of quality flags
coming along with data from different data providers can thus
be very time consuming for the user. Another issue can occur in
the case of a lack of information about the results of individual
quality tests when data meant for a specific purpose do not meet
chese predefined quality criteria and are excluded although they
may be useful for other scientific questions under consideration.
Data flagging is therefore highly useful as a plausibility filter
to exclude wrong data from datasets without a detailed
knowledge of the specific sensor characteristics and
{unctionality as well as without a specific knowledge on the
later scientific question. Data flagging however cannot replace a
quality assurance procedure providing statistically robust
Juantitative information on the data’s uncertainty at a
specified confidence range.
Another contribution to the overall uncertainty budget can
be extracted from the sensor specifications determined in the
manufacturer laboratory at the time of production and
zalibration that also should find entry into the metadata
description of measured data. Most sensor manufacturers
provide initial accuracy and precision values for their sensors
and sometimes also information about the stability or drift over
:ime. Even though this information is exactly the type of
metadata required to calculate a sensor’s uncertainty or
confidence, one has to keep in mind that these manufacturer
metadata are laboratory values referring to a brand-new or re-
calibrated sensor and therefore do not take the sensor lifetime
and environmental conditions during storage, transportation
and/or deployment into account. Furthermore, it must be
zonsidered that manufacturers sometimes provide only
information for their sensors describing a typical accuracy
and/or precision for a sensor but not for a specific sensor
instance. Better qualified sensor specific metadata are only
available if the manufacturer provides a sensor specific
zalibration sheet with detailed information on the serial
number of the respective sensor or if a recalibration will be
carried out by another calibration laboratory. Therefore, we have
to consider different levels of availability and reliability of given
sensor accuracies indicating the demand for proper
documentation of sensor metadata.
As mentioned before, flags are markers for data plausibility
and provide workflow transparency. They do not include
detailed information about the significance or robustness of a
Zrontiers ın Marıne Science
ji
10.3389/fmars.2022.1002153
single data point/measurement. The manufacturers quality
parameters of a sensor’s data provide this information but
cannot be easily applied to the operational phase of a
measurement program. From the scientific point of view the
knowledge of both information, realistic uncertainty in the
operational phase of an experiment as well as flags determined
‘rom plausibility tests, would be helpful to prevent scientists
‘rom misinterpretation of data. However, while flags can be
assigned to data points independent from the operational phase
and status of a measurement, a way to assign uncertainty
information on sensor measurements in the operational phase
seem to be largely unknown and not yet widespread in
ocean sciences.
Experimental set-'.o
The experimental setup was designed to be as close as
possible to a normal monitoring program that would be
conducted in a coastal area with a duration of several months.
A balanced experimental approach with six different multi-
parameter probes [CTD: three Sea & Sun Technology
(Sea&Sun, 2022) and three Sea-Bird Scientific (Seabird, 2022)]
(see also Supplementary Materials, Appendix 1, Table Al for
more details) from four different marine institutes in Germany
were chosen. These six probes were deployed in the MarGate
underwater test site (Wehkamp et al., 2013) off Helgoland in the
southern North Sea from July 20° to November 25", 2020. This
underwater experimental field is jointly operated by the two
Helmholtz institutes, Alfred-Wegener-Institute Helmholtz
Centre for Polar- and Marine Research (AWI) and the
Helmholtz Institute HEREON (formerly HZG) as an
.nternational monitoring and test facility for marine observing
components. It has been part of the EU project Jerico-Next
Jerico-Next, 2019) where international cooperating partners
could apply for the financial cover of time slots to evaluate
marine sensors for scientific purposes.
The experimental area has a cable connected underwater
aode with ten submersed ports for continuous power and high-
speed data connection for the remote-controlled operation of
underwater sensor systems (Fischer, 2019).
The underwater field is continuously monitored for the main
essential ocean variables such as temperature, conductivity,
oxygen saturation, chlorophyll-a, turbidity, photosynthetic
active radiation (PAR), current and wave height, as well as
additional variables as pCO2 and methane concentration.
Experiments in the so called MarGate field are supported year-
round by specifically trained scientific divers who are responsible
for sensor maintenance, repairing and replacing of sensors and
new experimental set up. The area provides a highly demanding
environment with average wind speed peaks of more than 6 bft
(10.8-13.8 m/s) on more than 200 days a year and tidal currents
up to 1 m/s.
‚rontiersin.ora
