3 D geospatial modelling and visualization for marine environment : Study of the marine pelagic ecosystem of the south-eastern Beaufort Sea , Canadian Arctic

Geospatial modelling of the marine pelagic ecosyste m is challenging due to its dynamic and volumetric nature. Consequently, conventional oceanographic spatial analysis of this environment is in a 2D environment, limited to sta tic cutting planes in horizontal and vertical sections to present various phenomena. In this paper, we explore the contribution of rece nt 3D development in GIS and in scientific visualization tools for representatio n and analyses of oceanographic data sets. The adva nt ges of a 3D solution are illustrated with a 3D geospatial voxel representati on of water masses distribution in the southeastern Beaufort Sea (west of the Canadian Arctic).


INTRODUCTION
Oceans cover 71% of the Earth's surface and with an average depth of approximately 4000 m, the volume of the marine pelagic ecosystem (water column) represent 99% of the biosphere (Angel, 1993).From a geospatial modelling perspective, this environment imposes different challenges compared to the terrestrial system.Whereas the latter is most often represented as an empty space filled with object (trees, houses etc...), the pelagic ecosystem is a continuous abiotic and biotic spatial geographical phenomenon in a full threedimensional (3D) environment.Study of the marine ecosystem, as well as management and conservation of marine resources, can be enhanced with adequate geospatial 3D modelling.
Traditional Geographic Information Systems (GIS), that are leading tools for the study and observation of spatial data, are not suitable to model geoscientific datasets since they have been principally designed for static and two-dimensional (2D) objects in terrestrial applications (Carette et al., 2008;Ledoux and Gold, 2008;Wright and Goodchild, 1997).Nevertheless, recent 3D GIS development justifies an update of their potential for marine pelagic geospatial modelling.Although lacking much of the flexibility in data management and ease of use of GIS, various specific scientific visualization tools have also been developed for modelling of the geologic subsoil, mainly motivated by oil and gas industry.These geomodelling tools are interesting from an oceanographer's perspective in that they treat, at least partially, the same type of continuous field as we find in the pelagic ecosystem.

*
In this paper, we first analyse and compare the capacity of different GIS tools for representation, visualization and analysis of a 3D dynamic marine environment.Specifically, we conduct a qualitative comparison between capacities of commercial GIS, commercial marine GIS, and academic prototype GIS as well as geomodelling tools (section 2).We demonstrate how recent development in 3D spatial modelling tools can improve representation, visualization and analyses of oceanographic phenomena and highlight some improvements that should be carried out to these tools in order to achieve an optimal marine spatial modelling tool.Finally, we propose an integration of the benefits from 3D geomodelling tools with advantages of GIS to improve 3D spatial modelling of oceanographic data sets.The third part of the paper is devoted to a case study proposing a 3D solution to visualization of water masses distribution in the southeastern Beaufort Sea (west of the Canadian Arctic).The data for this case study were obtained from the Malina oceanographic campaign conducted over the Mackenzie shelf between the 31 th of July and 26 th of August in 2009 (Figure 1).More detailed information about the Malina campaign can be found elsewhere (e.g.Matsuoka et al. 2012).

3D SPATIAL REPRESENTATION OF THE PELAGIC ECOSYSTEM
Pelagic marine features are characterized by their fuzzy boundaries, dynamic, and full 3D structure 1995; Shyue and Tsai, 1996).These characteristics are restrictive for data acquisition, as well as for geospatial modelling and representation.A further problem with oceanographic data sets is the frequent anisotropic distribution of data, due primarily to logistics and costs associated with expensive sampling at sea.Development of t 3D geospatial modelling of the marine pelagic ecosystem consequently challenging.This might explain why c analysis of oceanographic phenomena is traditionally in a 2D environment, limited to static cutting planes vertical sections either contoured or colour various parameters (Head et al., 1997).
Oceanic physical parameters, such as temperatur define distinct water masses with more or less fuzzy boundaries.
Vertical and horizontal distribution of these water masses influences the oceanic carbon cycles, which in turn play an important role in regulating global climate.objective of the Malina oceanographic campaign better understanding of these interactions.spatial analyses in a 3D geospatial model of these phenomena can then be of great value.Such a model could also be of good use in a resource management or conservation perspective.
A summary of some common and specialised tools reviewed in this work are listed in Table 1.These have been evaluated according to criteria for their suitability for modelling of the pelagic environment.This review indicates

Criteria ArcGIS 10
Commercial GIS 3D Interpolation -3D raster representation -3D vector representation Visualization cuts -Visualization iso-surfaces -Visualization volumes -3D statistical analyses -3D spatial analyses -Table 1. Review of five geospatial modelling tools from commercial and academic GIS as well as from geomodelling.

3D GEOSPATIAL SOLUTI REPRESENTATION OF WATER MASSES DISTRIBUTION: MALINA CASE STUDY
In the south-eastern Beaufort Sea, several types of can be identified, such as the nutrient rich pacific Halocline Water (UHW).The fractional presence of obtained for each of the 243 sampling points (x, y, z) accordingly to a method described by Lansard geospatial voxel model of this water mass was constructed Paradigm GOCAD, a scientific visualization tool developed for 3D geological spatial modelling.This spatial by a grid of 150 x 75 x 100 voxels in x, y, z direction compressed vertically between water surface and bathymetric surface.UHW values were attributed to each voxel through a 3D interpolation of sampling points with ordinary kriging.

ATION OF THE PELAGIC
Pelagic marine features are characterized by their fuzzy structure (Gold and Condal, .These characteristics are restrictive for data acquisition, as well as for geospatial modelling and representation.A further problem with oceanographic data sets is the frequent anisotropic distribution due primarily to logistics and costs associated with expensive sampling at sea.Development of tools available for of the marine pelagic ecosystem is might explain why conventional is traditionally in a 2D static cutting planes in horizontal and vertical sections either contoured or colour-coded to present Oceanic physical parameters, such as temperature and salinity, define distinct water masses with more or less fuzzy boundaries.Vertical and horizontal distribution of these water masses influences the oceanic carbon cycles, which in turn play an important role in regulating global climate.One partial objective of the Malina oceanographic campaign was to gain better understanding of these interactions.Visualization and spatial analyses in a 3D geospatial model of these phenomena can then be of great value.Such a model could also be of good resource management or conservation perspective. A summary of some common and specialised tools reviewed in listed in Table 1.These have been evaluated criteria for their suitability for 3D geospatial modelling of the pelagic environment.This review indicates that recent efforts of 3D development in the GIS field have mostly focused on object centred conceptual design using vector structures (for example: ArcGIS version 10, Fledermaus).Indeed, several research teams have adequate 3D marine GIS (Arsenault et al., 2004;Mesick et al., 2009).We consider that an adequate representation of marine pelagic continuous phenomena needs fully developed volumetric field representations.That kind of representation is more developed in geomodelling tools.The general solution the use of 3D raster-based models, commonly referred to as voxel (VOlume piXEL) structures.Although beginning of 1990, they are still GIS (the GIS open-source GRASS might however be an exception to this generality, offering It is also worth noticing that the more Voronoï tessellations, whose advantages various academic works (Beni et al., 2011;Ledoux and Gold, 2008), are absent so far in commercial geospatial model tools.A draw-back with the geomodelling tool reviewed in this paper is its limited ability for image treatment an imperative in oceanographic research considering the use of remote sensing.Another essential marine geospatial modelling tool static cuts in vertical direction.Even though might seem trivial and does not require a method, this operation is not possible at present with and very limited with EnterVol ArcGIS that permits volumetric representation Finally, none of the tools evaluated consideration the dynamic nature and fuzzy boundaries of pelagic phenomena or to assess a general of spatial 3D models, such as cross 2007).Review of five geospatial modelling tools from commercial and academic GIS as well as from geomodelling.

3D GEOSPATIAL SOLUTION TO TER MASSES MALINA CASE STUDY
types of water masses nutrient rich pacific Upper he fractional presence of UHW was for each of the 243 sampling points (x, y, z) Lansard et al. (2012).A geospatial voxel model of this water mass was constructed with scientific visualization tool developed for spatial model was built voxels in x, y, z direction and between water surface and bathymetric attributed to each voxel in the model points with ordinary sampling points that recent efforts of 3D development in the GIS field have mostly focused on object centred conceptual design using vector structures (for example: ArcGIS version 10, Fledermaus).have recognized the lack of Arsenault et al., 2004;Mesick et al., e consider that an adequate representation of marine pelagic continuous phenomena needs fully developed That kind of representation is more developed in geomodelling tools.The general solution is based models, commonly referred to as ) structures.Although in use since the ll mostly absent in commercial GRASS might however be an exception to this generality, offering limited volume rendering).
more dynamic data structureswhose advantages have been stressed in Beni et al., 2011; Ledoux and Gold, so far in commercial geospatial modelling back with the geomodelling tool reviewed in this paper is its limited ability for image treatment and analyses, imperative in oceanographic research considering the common essential function for an optimal tool is the visualization of 2D . Even though such a function does not require a true 3D interpolation this operation is not possible at present with ArcGIS EnterVol, commercial extension to ArcGIS that permits volumetric representation (Table 1).
tools evaluated permit to take into consideration the dynamic nature and fuzzy boundaries of a general predictive capability cross-validation (e.g.Foglia et al.
Review of five geospatial modelling tools from commercial and academic GIS as well as from geomodelling.
2. Kriging variance of spatial 3D model for pacific Beaufort Sea.Black dots indicate points.
The kriging variogram's dependent predictive error is presented in Figure 2. In general, the lower the error for a specific location, the better is the prediction of the spatial model.
The final spatial model permits us to visualize iso volumes as well as cuts in any plane of the water mass (Figure

DISCUSSION AND CONCL
This research has explored the potential of 3D geospatial modelling tools for the study of marine pelagic ecosystems.A review of common GIS indicates that these software mostly lack the necessary functions for volumetric representation of gradual phenomena, primordial for geospatial study of marine ecosystems.However, geomodelling tools for representation of this environment is promising, which is illustrated in this paper by a 3D solution to visualization of a water mass in the Beaufort Sea, constructed with Paradigm Gocad Integration of volumetric representation in a GIS environment is an important advance towards an optimal marine GIS must also include representation and analyses static vertical cuts.On the contrary, geomodelling tools be adapted to the marine environment by functions for oceanographic research such as image treatment and analyses.However, all spatial modelling tools conceived for the pelagic environment would also benefit from including spatial data structure that takes into consideration nature and fuzzy boundaries of the pelagic environment.
The kriging variogram's dependent predictive error is presented in Figure 2. In general, the lower the error for a specific location, the better is the prediction of the spatial model.
The final spatial model permits us to visualize iso-surfaces and volumes as well as cuts in any plane of the water mass (Figure 3) and enables spatial 3D analyses, such as volume calculation and intersection.This case study shows that recent advances in volumetric representation developed primarily for geomodelling tools can be used to extend usual interpretation of static marine pelagic phenomena from 2D stat environment.
. S indicates that these software products lack the necessary functions for volumetric rimordial for geospatial study of marine ecosystems.However, performance of geomodelling tools for representation of this environment is promising, which is illustrated in this paper by a snap-shot of a 3D solution to visualization of a water mass in the south-eastern , constructed with Paradigm Gocad (Figure 3).Integration of volumetric representation in a GIS environment is an important advance towards an optimal marine GIS.This tool and analyses functions of 2D On the contrary, geomodelling tools could by improving basic such as image treatment , all spatial modelling tools conceived for benefit from including takes into consideration the dynamic aries of the pelagic environment.
Our future research will explore geospatial modelling tools for the combination of satellite and ground data in the identification of organic carbon fluxes in the Beaufort Sea

Figure 1 .
Figure 1.Location of sampling stations for the Malina cruise in the Beaufort Sea, Canadian Arctic.

Figure 2 .
Figure 2. Kriging variance of spatial Upper Halocline Water in the Beaufort Sea.sampling points

Figure 3 .
Figure 3. Geospatial representation of Upper Halocline Water on the Mackenzie shelf approximately the upper 500 m of the constituted of UHW and voxels containing more than 60 % of UHW black dots.Vertical exaggeration in figure is Geospatial representation of Upper Halocline Water on the Mackenzie shelf in summer 2009 the water mass contained within the spatial model.Scale indicate oxels containing more than 60 % of UHW are coloured-filled.Visible sampling points Vertical exaggeration in figure is 75 times that of reality.Spatial model was constructed with Paradigm Gocad DISCUSSION AND CONCLUSIONS research has explored the potential of 3D geospatial modelling tools for the study of marine pelagic ecosystems.A