Assessing the Spatial Data Quality Paradox in the Deep-sea

Abstract

Knowledge of deep-sea environments is limited by the difficulties of using traditional sampling methods in such remote areas. Sampling the deep-sea from the ocean surface rarely yields data at a spatial scale that is helpful in understanding ecological processes or meaningful for management and conservation. One way to collect better information about the seafloor is to reduce the distance between the instruments and the seafloor. This is now possible using submersible platforms. A challenge with the use of these underwater systems is the inaccuracies associated with data positioning. Positioning high-resolution datasets accurately in an underwater geospatial context is complicated by the fact that many sources of uncertainty exist, contributing to a total propagated uncertainty (TPU) on the position. These complications are acute for acoustic remote sensing systems, in which the footprint and the resulting spatial resolution of data are a direct function of depth. While quality of deep-sea data is highly variable, it is rarely assessed or explicitly considered in marine ecological studies. In this contribution, we measured the mean TPU of bathymetric data collected during surveys performed in 2010 with a remotely operated vehicle in the Northwest Atlantic, to depths down to 3,000m. We found that TPU increases with depth, leading to a " paradox of data quality " : sensors' resolution increase with depth (i.e. when reducing the distance between the sensor and the seafloor) while sensors' positional accuracy decrease with depth. We conclude that in order to be able to accurately position high-resolution datasets in the deep-sea within the same absolute reference system, the spatial resolution of the data should be larger than the TPU. Spatial data quality of underwater datasets should always be assessed, as often only the spatial resolution side of this paradox is explicitly addressed in the literature.

Full text

Assessing the Spatial Data Quality Paradox in the Deep-sea
Vincent Lecours1 and Rodolphe Devillers1
1 Marine Geomatics Research Lab, Department of Geography, Memorial University of Newfoundland,
vlecours@mun.ca
Abstract
Knowledge of deep-sea environments is limited by the difficulties of using traditional sampling
methods in such remote areas. Sampling the deep-sea from the ocean surface rarely yields data
at a spatial scale that is helpful in understanding ecological processes or meaningful for
management and conservation. One way to collect better information about the seafloor is to
reduce the distance between the instruments and the seafloor. This is now possible using
submersible platforms. A challenge with the use of these underwater systems is the inaccuracies
associated with data positioning. Positioning high-resolution datasets accurately in an
underwater geospatial context is complicated by the fact that many sources of uncertainty exist,
contributing to a total propagated uncertainty (TPU) on the position. These complications are
acute for acoustic remote sensing systems, in which the footprint and the resulting spatial
resolution of data are a direct function of depth. While quality of deep-sea data is highly
variable, it is rarely assessed or explicitly considered in marine ecological studies. In this
contribution, we measured the mean TPU of bathymetric data collected during surveys
performed in 2010 with a remotely operated vehicle in the Northwest Atlantic, to depths down
to 3,000m. We found that TPU increases with depth, leading to a “paradox of data quality”:
sensors’ resolution increase with depth (i.e. when reducing the distance between the sensor and
the seafloor) while sensors’ positional accuracy decrease with depth. We conclude that in order
to be able to accurately position high-resolution datasets in the deep-sea within the same
absolute reference system, the spatial resolution of the data should be larger than the TPU.
Spatial data quality of underwater datasets should always be assessed, as often only the spatial
resolution side of this paradox is explicitly addressed in the literature.
Key words: Acoustic remote sensing, data quality, total propagated uncertainty, deep-sea,
spatial resolution
Background and Relevance
The use of satellite remote sensing to study marine environments is limited by
the capacity of electromagnetic energy to penetrate water, resulting in a dearth of
knowledge of the marine environment under the first few metres of water (Solan et al.,
2003; Robinson et al., 2011). Marine habitat mapping aims to use knowledge of the
chemical, physical and biological properties of an area to understand biological
distribution in marine environments (Brown et al., 2011). One of the most commonly
used methods to map habitats is to sample these environmental properties and assess
how they influence species distribution (e.g. Freeman & Rogers, 2003). This knowledge
is then used in predictive modeling to estimate species distribution in unsampled areas
(e.g. Tong et al., 2013).
Previous research (e.g. Davies & Guinotte, 2011; Lecours et al., 2013; Rengstorf et
al., 2012, 2013) shows that the coarse resolution data usually available for deep-sea
environments do not always significantly explain the distribution of some biological

species and are not meaningful for purposes such as management. The marine habitat
mapping community needs higher resolution data to understand the real processes
driving species distribution. This is particularly true for multibeam bathymetric data
collected from acoustic systems, which have proven their value for habitat mapping
(Brown et al., 2011). Assuming the use of comparable acoustic systems, higher
resolution bathymetric data can be collected by decreasing the distance between the
sensor and the seafloor. Getting closer to the seafloor creates a smaller footprint and a
higher density of soundings on the seafloor, resulting in a finer spatial resolution of the
bathymetric data (Lurton, 2010). This can be done with the help of submersible
platforms such as remotely operated vehicles (ROV) or autonomous underwater vehicles
(AUV) (Wright, 1999). However, data collection using submersible platforms in deep-
sea environments presents important challenges, particularly in terms of positional
accuracy (Wright & Goodchild, 1997). The quality of the data directly impacts the
reliability of species-environment relationships measurements, habitat maps and
predictive models. Analyses of data quality are rarely performed in habitat mapping
studies (Barry & Elith, 2006), and it is estimated that the most important sources of
uncertainty come from data acquisition (Rocchini et al., 2011).
In this paper, we present a data quality assessment of multibeam bathymetric
data collected for deep-sea habitat mapping using a ROV. We measured the total
propagated uncertainty (TPU) of these data and their spatial resolution and compared
them with the depth of the surveys. Using these results, we discuss a data quality
paradox that occurs when collecting data in the deep-sea using submersible vehicles.
Methods and Data
High-resolution multibeam sonar, video and oceanographic data were collected
in 2010 off Newfoundland and Labrador, Canada, using the ROV ROPOS (Remotely
Operated Platform for Ocean Science) from the Canadian Scientific Submersible
Facility. Multibeam bathymetric data were collected using an Imagenex Delta-T system
mounted on the ROV. The ROV surveyed at heights varying from 1m to 50m above the
seafloor. Several instruments were used to estimate the position of the ROPOS. First, an
IXSEA GAPS ultra-short baseline (USBL) used an acoustic pulse travelling between the
supporting surface vessel and transducers mounted on the ROV to calculate the range
between them, from which a relative position was derived. This USBL had a 0.2% root
mean square (RMS) slant range accuracy. Then, a Workhorse Navigator Doppler
Velocity Log (DVL) tracked the seafloor when the height of the ROV was less than 30m.
Using speed measurements in all directions, the DVL derived a position relative to the
starting point. Long-term accuracy of the DVL was ±0.2% ±0.1cm/sec. An IXSEA
OCTANS fibre-optic gyrocompass and motion reference unit (MRU) also measured the
motion and speed of the ROV to determine its position relative to the starting point. The
accuracy of the gyrocompass was ±0.1° RMS and the accuracy of measurements from
the MRU was ±0.01° RMS. Finally, a Paroscientific Digiquartz depth sensor was used,
with an accuracy of 0.01% of the measured depth. The four sensors were used together
to improve the quality of position measurements, as often performed in the literature
(e.g. Rigby et al., 2006); the four sets of positions were merged using a Kalman Filter.
The configuration of the ROV is important to correct for multibeam bathymetric
data, as an inaccurate configuration can lead to errors in the geometric correction of the

sound beams. A survey was performed by the Canadian Hydrographic Service prior to
data collection to know the exact relative position of each sensor compared to the others
on the ROV.
All the information regarding the instruments’ errors and the ROV configuration
were entered in the bathymetric processing software CARIS HIPS and SIPS 9.0, which
was used to estimate the mean horizontal and vertical TPU on data from 14 dives. TPU
is a common measure to quantify the quality and accuracy of bathymetric data (Foster et
al., 2014). The theoretical resolution is dependent on the geometry of the multibeam
measurement; it is a function of the angular resolution of the multibeam system and the
distance to the seafloor (Lurton, 2010). The angular resolution depends on the number
of sound beams and their total angle. During these surveys, the angle was set to 120°
and 120 beams were used. We calculated the theoretical resolution using the mean
distance to the seafloor per transect analyzed, and compared it to the mean TPU as a
function of depth. There is a need to distinguish between theoretical spatial resolution,
which is the spatial resolution that can be reached based on the sensor-to-target
distance, and the practical spatial resolution, which is the greater value between the
theoretical spatial resolution and the positional uncertainty of data. To calculate the
practical resolution, we quantified the relationships between TPU and surveying depth
and theoretical spatial resolution and surveying depth. Using these equations, we
identified the intersections of the TPU curves with the theoretical spatial resolution
curves for depths ranging from 0 to 5,000m, which correspond to practical spatial
resolutions.
Results
Figure 1 shows that the horizontal TPU increases linearly with the mean depth of
the surveys, ranging from a TPU of 1.8m at less than 50m deep, to more than 50m at
depths higher than 2,700m. Errors associated with measurements from the Kalman
filter contributed most to the horizontal TPU. Vertical TPU did not show any specific
relationship with depth. It ranged from 0.993m to 1.12m and was mostly influenced by
the measured depth of the ROV (79 to 97%), the heave of the platform (up to 12%), and
the alignment and timing of the MRU (up to 8%).
Figure 2 illustrates how the theoretical spatial resolution increases as the sensor-
to-target distance decreases. When the ROV was very close to the seafloor (i.e. about
1m), it enabled the collection of data at a spatial resolution of less than 2cm. The
theoretical spatial resolution could not be measured for two of the dives as the ROV
surveyed higher than 30m from the seafloor, the limit at which the DVL could track the
bottom.
Using the equations generated in Figure 1 and 2, Figure 3 is an example, using a
seafloor at 3,000m deep, of how theoretical spatial resolution and horizontal TPU vary
with ROV depth. The two curves meet at 1,409m, when both the TPU and resolution are
≈28m. The practical spatial resolution at which to collect data at 3,000m is thus 28m,
and the ROV would need to be at ≈1,400m deep to keep the TPU lower than the spatial
resolution. Figure 4 extends this example to different depths and illustrates how
practical spatial resolution varies with depth.

Figure 1: Increasing propagated uncertainty (TPU) on the position of data with depth
Figure 2: Decreasing spatial resolution with distance between the sensor and the seafloor

Figure 3: Comparison between the variation of theoretical spatial resolution and TPU with
ROV depth when collecting data on a 3,000m deep seafloor
Figure 4: Variation of practical spatial resolution and appropriate surveying depth (i.e.
ROV depth) with seafloor depth

Discussion
Figure 1 illustrates that the deeper the survey, the greater the positional
uncertainty, a pattern also reported by Rattray et al. (2014) in shallow waters. TPU
values in our study are lower than values from Rattray et al. study at comparable depths,
a possible result of using the Kalman filter or using a different system. Rattray et al.
(2014) showed that their USBL and MRU contribution to the TPU increased with depth,
which is also shown in our data through the Kalman filter and the MRU. Figure 2
confirms the known relationship between the spatial resolution of the resulting data and
the sensor-to-target distance (Lurton, 2010).
When trying to accurately position datasets in the same spatial reference system,
positional accuracy needs to be analyzed as the spatial resolution of data should be
larger than the uncertainty associated with data position (Moudrý & Šímová, 2012). The
red area in Figure 3 corresponds to the depths where TPU is greater than spatial
resolution, while the opposite is observed in the green area. This constraint forms the
basis of what we call a “data quality paradox” in the deep-sea: deeper submersible
surveys increase data resolution while decreasing absolute positional accuracy. The
collection of higher resolution data is therefore limited by the sampling system, its
associated TPU and the depth of the survey: Figure 4 identifies this limit as a function of
depth for the system that we used. For instance, we can find, using the calculated
equations, that if the seafloor is at 1,000m deep, we can survey with the ROV at ≈500m
deep and collect data at a practical resolution of ≈10m while being certain that the TPU
is smaller than the spatial resolution.
When quantifying relationships between variables, the influence of uncertainty
and low positional accuracy increases with the spatial resolution of the data (Hanberry,
2013). Considering that uncertainty associated with geospatial data involves a trade-off
between data quality (i.e. accuracy and precision) and spatial scale (i.e. spatial
resolution and extent), Braunisch & Suchant (2010) tried to explore which of these
characteristics should be given priority in sampling strategy. The issue is still
unresolved: some believe that a finer spatial resolution should be targeted (e.g. Reside et
al., 2012) while others think that the focus should be lowering the uncertainty (e.g.
Braunisch & Suchant, 2010). An assessment of uncertainty should be performed in
either case and this information should be documented in metadata to enable users to
consider it during analysis. It is important to remember that for some purposes there is
only a need to position datasets in a relative spatial reference system. For instance, if
datasets are collected with different sensors onboard the same submersible at the same
time and the TPU is mostly influenced by the position of the platform rather than by the
instruments themselves, it is possible to use the theoretical spatial resolution because all
data would be in the same relative reference system. On the other hand, when datasets
are collected from different instruments on different platforms that do not share the
same reference system, there is a need to position them in an absolute frame of
reference. Such absolute positioning would then be limited by the TPU and the practical
spatial resolution should be used.

Conclusions
Combining high-resolution geospatial datasets collected from different sensors
and platforms is commonly done when performing marine habitat mapping studies. In
order to use these datasets at their highest spatial resolutions, they need to be accurately
positioned. However, an assessment of the quality of these datasets is rarely performed
in the literature. Published works often focus on the spatial resolution of data but do not
account for positional accuracy. To our knowledge, this study is the first to compare
theoretical spatial resolution and positional uncertainty of bathymetric data and to
identify the practical spatial resolution as a function of surveying depth. A potential
solution to improve data accuracy in the deep-sea is to divide the TPU into its different
components and to try to mitigate their effects both prior to and during surveys.
Surveyors need to be aware of their system’s constraints and keep their expectations
within the limits of their data. Estimations of TPU as a function of depth should be done
prior to surveying with the instruments and configuration used, which would allow the
identification of the proper spatial resolution at which to generate the final datasets (i.e.
the practical resolution). These recommendations are only valid when the purpose of the
survey is to spatially match several datasets in an accurate, absolute geospatial context.
References
Barry, S., & Elith, J. (2006). Error and uncertainty in habitat models. Journal of Applied
Ecology 43, 413-423.
Braunisch, V., & Suchant, R. (2010). Predicting species distributions based on incomplete
survey data: the trade-off between precision and scale. Ecography 33, 826-840.
Brown, C.J., Smith, S.J., Lawton, P., & Anderson, J.T. (2011). Benthic habitat mapping: A review
of progress towards improved understanding of the spatial ecology of the seafloor using
acoustic techniques. Estuarine, Coastal and Shelf Science 92, 502-520.
Davies, A.J., & Guinotte, J.M. (2011). Global habitat suitability for framework-forming cold-
water corals. PLOS ONE 6, e18483.
Foster, B., Hart, K., Froelich, G., & Lamey, B. (2014). Enhanced Total Propagated Uncertainty
(TPU) in CARIS HIPS and SIPS. Canadian Hydrographic Conference, St. John’s, NL, April
14-17.
Freeman, S.M., & Rogers, S.I. (2003). A new analytical approach to characterisation of macro-
epibenthic habitats: linking species to the environment. Estuarine, Coastal and Shelf Science
56, 749-764.
Hanberry, B.B. (2013). Finer grain size increases effects of error and changes influence of
environmental predictors on species distribution models. Ecological Informatics 15, 8-13.
Lecours, V., Miles, L.L., Devillers, R., & Edinger, E.N. (2013). Data analysis towards the
multiscale characterization of cold-water coral and sponge habitats in Canadian waters.
Technical report submitted to the Department of Fisheries and Oceans Canada, REQ. No.
F6160-120010.
Lurton, X. (2010). An introduction to underwater acoustics: Principles and applications (second
edition). Springer/Praxis Publishing, 704 p.

Moudrý, V., & Šímová, P. (2012). Influence of positional accuracy, sample size and scale on
modelling species distributions: a review. International Journal of Geographical
Information Science 26, 2083-2095.
Rattray, A., Ierodiaconou, D., Monk, J., Laurenson, L.J.B. & Kennedy, P. (2014). Quantification
of spatial and thematic uncertainty in the application of underwater video for benthic habitat
mapping. Marine Geodesy 37, 315-336.
Rengstorf, A.M., Grehan, A., Yesson, C., & Brown, C. (2012). Towards high-resolution habitat
suitability modeling of vulnerable marine ecosystems in the deep-sea: resolving terrain
attribute dependencies. Marine Geodesy 35, 343-361.
Rengstorf, A.M., Yesson, C., Brown, C., & Grehan, A.J. (2013). High-resolution habitat
suitability modelling can improve conservation of vulnerable marine ecosystems in the deep-
sea. Journal of Biogeography 40, 1702-1714.
Rigby, P., Pizarro, O., & Williams, S.B. (2006). Towards geo-referenced AUV navigation through
fusion of USBL and DVL measurements. Oceans 2006, Singapore.
Robinson, L.M., Elith, J., Hobday, A.J., Pearson, R.G., Kendall, B.E., Possingham, H.P., &
Richardson, A.J. (2011). Pushing the limits in marine species distribution modelling: lessons
from the land present challenges and opportunities. Global Ecology and Biogeography 20,
789-802.
Rocchini, D., Hortal, J., Lengyel, S., Lobo, J.M., Jiménez-Valverde, A., Ricotta, C., Bacaro, G., &
Chiarucci, A. (2011). Accounting for uncertainty when mapping species distributions: the
need for maps of ignorance. Progress in Physical Geography 35, 211-226.
Reside, A.E., Watson, I., & VanDerWal, J. (2011). Incorporating low-resolution historic species
location data decreases performance of distribution models. Ecological Modelling 222, 3444-
3448.
Solan, M., Germano, J.D., Rhoads, D.C., Smith, C., Michaud, E., Parry, D., Wenzhöfer, F.,
Kennedy, B., Henriques, C., Battle, E., Carey, D., Iocco, L., Valente, R., Watson, J., &
Rosenberg, R. (2003). Towards a greater understanding of pattern, scale and process in
marine benthic systems: a picture is worth a thousand worms. Journal of Experimental
Marine Biology and Ecology 285-286, 313-338.
Tong, R., Purser, A., Guinan, J., & Unnithan, V. (2013). Modeling the habitat suitability of deep-
water gorgonian corals based on terrain variables. Ecological Informatics 13, 123-132.
Wright, D.J. (1999). Getting to the bottom of it: tools, techniques, and discoveries of deep ocean
geography. Professional Geographer 51, 426-439.
Wright, D.J., & Goodchild, M.F. (1997). Data from the deep: implications for the GIS
community. International Journal of Geographical Information Science 11, 523-528.