Naldmann et al. 
‚.ntroduction 
Considering the importance of judging the significance of 
observations to detect long term trends in earth systems, the 
current study appears to be timely and relevant for different 
kinds of observational activities e.g., as part of the UN Decade of 
Ocean Science for Sustainable Development (UN DECADE, U 
2021). Measuring variables in the field is fundamentally different 
from lab measurements as in-situ measurements are unique in 
space and time and are of transient character. In the ocean 
sciences, where access to environmental data is often limited due 
to required ship time or due to weather constraints, single sensor 
data without the chance of replication and limited information 
on data quality are the only available source of information. The 
-hallenge is to define the concept of “data quality” which is 
connected to “measurement uncertainty”. 
Although in other disciplines like atmospheric observations 
as conducted by the World Meteorological Organization 
(WMO) the concept of uncertainty has already found entry 
‚WMO, 2008), while ocean sciences have only dealt with 
uncertainties for specific parameters and often with a 
limited scope. 
Over the past years, various aspects on data quality have 
been considered (Wong et al., 2022), (Bushnell, 2019) including 
:he “FAIR” data concept (Wilkinson, 2016). The FAIR concept 
means that scientific data must be “findable”, “accessible”, 
“interoperable” and “reusable” including a minimum of 
associated metadata information which make data transparent 
with respect to their origin and their processing workflow. 
However, even FAIR data do not contain an adequate 
description of data quality, as is classically requested as a 
minimum standard in natural sciences. Referring to 
international conventions (UNESCO, 2013) here, data quality 
refers to the availability of “accuracy” and “precision” 
information [see Supplementary Material, Appendix 4 on 
terminology which is based on (BIPM, 2008)]. This will allow 
the calculation of a statistically robust uncertainty resp. 
confidence interval for each measured data point spanning the 
range within which the best estimate value lies with a specified 
(e.g., 95%) probability. Only when this statistical information for 
each individual data point is available can the “quality” of this 
data point be quantitatively assessed in a close context to the 
respective scientific question. While, for example, in behavioral 
ecology, temperature measurements used to determine whether 
an area is habitable for a particular species usually do not need to 
be more accurate than one degree Celsius, studies of the effects of 
climate change-induced heat content changes in the deep sea 
require uncertainties that do not exceed one hundredth of a 
degree or even less. Thus, while for the behavioral ecologist, the 
ıbove-mentioned dataset is of sufficient and thus of “high” 
quality, for the oceanographer the same data set is of 
insufficient and therefore of “poor” quality. This implies that a 
data point without information about its uncertainty is neither 
Zrontiers ın Marıne Science 
4% 
10.3389/fmars.2022.1002153 
good nor bad, but in a kind of premature raw data state that is 
not yet suitable for scientific use and publication without 
‘urther refinement. 
The study that has been carried out strives to perform an 
analysis of the relevant factors required to calculate a measured 
data points uncertainty. In a series of follow-up papers, we will 
address other related topics such as uncertainty of salinity 
uneasurements or uncertainty due to sensor drift. 
In a first step, a basic concept of uncertainty calculation for 
time-series data measured by standard oceanographic probes 
(CTD, Conductivity, Temperature, and Depth) is presented. To 
keep the analysis as comprehensive as needed, while being as 
simple as possible, we here concentrate on analyzing the variable 
temperature as an example. Nevertheless, it should be relatively 
straightforward to extend this analysis to other, more complex, 
3ssential Ocean Variables [EOVs, (GOOS, 2020)]. 
Along a data processing chain from the raw sensor output 
(e.g., T=14.345°C), we evaluate different procedures for a simple 
but statistically robust numerical uncertainty calculation of the 
measurement to finally come to an output in the form of 
T = 14.345 + 0.003°C 
or verbalized, 
Measured Value = Best estimate + Uncertainty 
(Taylor, 1997) 
The procedures presented here are not meant to replace 
existing procedures and frameworks for data quality assurance 
developed and established in ocean sciences over the last 
decades. The intention of the manuscript, however, is to 
complement the information that is provided for 
environmental data. 
A core element of “quality control” procedures are quality 
‘Jags that assign collected data into different quality categories. 
The processing and quality control results are stored and 
published alongside with the data to allow scientists to decide 
whether data are plausible within a set of mathematical and logic 
criteria. Flags assigned to data are independent from the later 
scientific question and provide information if data fulfill simple 
criteria which make them theoretically valid or invalid. The idea 
is to exclude obviously or probably wrong data from a dataset. 
Quality flags usually consist of a very basic defined code of 
numbers. A flag categorizes a data point as e.g., “good” or “bad”. 
[t can describe if data have been changed, replaced or added to 
the original raw data set during processing (e.g., “interpolated 
value”) or it can reveal certain events within a data set (e.g., offset 
detected, spike detected). Usually, a single data point is marked 
with a unique flag corresponding to a specific interpretation of 
the “quality”. Unfortunately, this marker is a combination of the 
results from different performed tests highly influenced by 
specified thresholds defined within each test. So far, there is no 
international agreement upon standards for flagging, as well as 
the choice of performed data quality tests and accompanied 
thresholds. Recommendations for necessary and optional useful 
"rontiersin.ora