visualisation strategies for environmental modelling datakarsten/data/rink2014vis.pdf ·...

Environmental Earth Sciences manuscript No.(will be inserted by the editor)

Visualisation Strategies for Environmental ModellingData

Karsten Rink · Lars Bilke · Olaf Kolditz

The final publication is available at Springer via http://dx.doi.org/10.1007/s12665-013-2970-2

Abstract We present a framework that allows users

to apply a number of strategies to view and modify

a wide range of environmental data sets for the mod-

elling of natural phenomena. These data sets can be

concurrently visualised to find inconsistencies or arte-

facts. This ensures at an early stage that models set up

for the simulation of hydrological or thermal processes

will not give implausible results due to complications

based on input data. A number of generally applica-

ble visualisation techniques are provided by our frame-

work to help researchers detect potential problems. We

also propose a number mapping algorithms for the in-

tegration of multiple data sets to resolve some of the

most common issues. Techniques for the presentation of

input- and modelling data in combination with simula-

tion results are proposed with respect to the benefits ofvisualisation of environmental data within specialised

environments. The complete workflow from input data

to presentation is demonstrated based on a case study

in Central Germany. We identify typical problems, pro-

pose approaches for a suitable data integration for this

case study and compare results of the original and mod-

ified data sets.

Keywords Computer Graphics · Visualisation ·Virtual Reality · Hydrology · OpenGeoSys Data

Explorer

K. Rink and L. Bilke and O. KolditzDepartment of Environmental Informatics,Helmholtz Centre for Environmental Research,Leipzig, Germany,E-mail: [email protected]

O. KolditzApplied Environmental System Analysis,Technische Universitat Dresden,Dresden, Germany

1 Introduction

Simulation of natural phenomena is becoming increas-

ingly important as we are faced with problems such as

climate change, overpopulation and the need to change

from fossil fuels to green energy. For instance, compre-

hending hydrological processes is necessary to predict

floods [23, 30], develop water management schemes for

arid regions [12, 15, 41] or simulate the contamination

of drinking water [4, 31]. Likewise, an understanding

of thermal processes is needed for geothermal applica-

tions such as extraction of geothermal energy [16, 38]

or the safety of nuclear waste deposits [35,42]. In order

to run simulations for a better understanding of such

phenomena in general or specific predictions for certain

regions, it is necessary to set up models based on a

large number of data sets. While basic simulations can

be calculated with only a few data sets, additional infor-

mation is often required to account for as many of the

natural characteristics for a given process as possible.

These data sets often differ significantly in spatial and

temporal resolution, structure, area of influence and re-

liability. A possible classification based on the structure

of input data sets is the following:

– Time series data: Sensors measuring temperature,

precipitation, soil moisture, etc. at a specific loca-

tion.

– Geometric/Vector data: Courses of streams acquired

via GPS, boundaries (e.g. of model regions, river

catchments, etc.) or borehole data.

– Raster Data: Maps, climate data such as weather

surveillance radar, remote sensing data acquired via

satellite or airplane (e.g. digital elevation models

(DEM), hyperspectral imagery, etc.), or data ac-

quired using geophysical monitoring techniques such

as ground-penetrating radar, gamma spectroscopy

or electromagnetic induction.

For hydrological case studies, the model is often de-

fined by a DEM in combination with (interpolated)

subsurface information gathered from borehole strati-

graphies, with model boundaries that are either de-

fined, measured or calculated (such as the boundary

of a river catchment). Additional information might be

integrated from land use maps to include water bod-

ies, cities or agricultural areas. Based on this data,

the domain is discretised into a finite element mesh

for a subsequent simulation. Boundary conditions are

often based on time series data gathered from obser-

vation sites such as precipitation- or gauging-stations.

Using adequate process equations, models predict one

or more simulated parameters within the domain over

time. Typical results include groundwater recharge, tem-

perature, or the distribution of chemicals in the envi-

ronment.

A concurrent visualisation of all the specified data

sets is a crucial part of the workflow prior to as well as

after the actual simulation. It supports researchers in

identifying potential problems while creating the model

and will show the plausibility of the results in compar-

ison to the input data after running the simulation.

In section 2 we give an overview over existing soft-

ware and approaches for the visualisation of complex

geoscientific data. Section 3 gives an overview on how

data is visualised, modified and presented using our

framework. The application of the proposed methods

is demonstrated in section 4 for a case study in Central

Germany.

2 Related Work

Despite the complexity of geoscientific data, surpris-

ingly few tools offer an integrated visualisation of a

large number of heterogeneous data sets. A number

of Geographic Information Systems (GIS) such as Arc-

GIS [20] or GRASS [24] offer a wide range of function-

ality for 2D visualisation of raster- and vector data. 3D

GIS approaches also exist, such as ArcGIS 3D Analyst,

but do not support typical data from exploration or

modelling software. A number of approaches use a 2D

GIS context and employ the third dimension for visu-

alisation of statistical or climate data [8, 14, 34]. Simu-

lation codes and libraries only rarely offer visualisation

functionality. Aside from our own framework, one of the

exceptions is FEFLOW [5] which provides an extensive

graphical user interface for model setup and displaying

simulation results. Due to the lack of more specialised

software, visualisation is often limited to multi-purpose

software such as ParaView [2] or VisIt [21]. These sup-

port the visualisation of a large number of graphics

file formats for 3D data as well as time discretisation.

For most cases these tools cannot be employed directly

for geoscientific or simulation data and require external

data conversion first. Visualisation techniques for com-

plex 3D or 4D data sets found in literature are mostly

concerned with applications for climate data [17,25,39].

Examples for the use of visualisation of hydrogeological

data are few. Jones et al. [18] propose an approach for

multi-scale geological models that allows handling data

sets of varying scales; Ling and Chen [22] present tech-

niques for the visualisation of contaminant distribution

in geological data; and Schuchardt et al. [33] propose

a framework for numerical model setup using hydroge-

ological data integrating VisIt for 3D visualisation. In

addition, there exist a number of commercial products

developed for use in the petroleum industry such as

GoCAD, Petrel or LeapFrog Geothermal [3], support-

ing mostly their own proprietary data formats.

3 Data Visualisation and Presentation

An adequate visualisation presents a tremendous bene-

fit for researchers as it provides a better understanding

of the data itself as well as the correlations between

variables within one or across multiple data sets. Fur-

thermore, it supports researchers in finding potential

problems for the subsequent modelling and simulation

process. Data visualisation is also an effective means

for presenting progress or results of a research study to

stakeholders or the interested public.

We have developed the OpenGeoSys Data Explorer

framework [27] for import and basic visualisation of geo-

scientific data as well as model information and simu-

lation results (Fig. 1). The Data Explorer is a platform-

independent graphical user interface for the OpenGeoSys

simulation code [19] for coupled thermal, hydrological,

mechanical and chemical processes. Display, interaction

and modification of data within the framework is im-

plemented using the Visualization Toolkit (VTK) [32],

an established library for 3D computer graphics and vi-

sualisation. For subsequent presentations, the Data Ex-

plorer also allows to export single geoscientific data sets

as well as complete scenes consisting of multiple data

sets into popular graphics formats, such as the VTK-,

VRML-, OpenSG [26] or Autodesk-format. This allows

for professional postprocessing using software such as

ParaView or Unity [9]. Additional VTK functionality

for performing transformations or emphasising certain

aspects of the data can be either applied within the

framework or after export. Likewise, after conversion


Fig. 1: The OGS Data Explorer framework displaying various input data sets from the Rappbode reservoir system,

including DEM [37], water bodies [1], observation sites, rivers, and time series data. While all spatial data sets are

visualised in 3D, the time series (in this case water temperature, water level and spectral absorption) are displayed

in a diagram. (Time series data kindly provided by Karsten Rinke)

to either OpenSG or Unity, advanced rendering meth-

ods can be applied to objects or scenes within the re-

spective frameworks. This workflow for preparation and

visualisation of models as well as simulation results has

already been applied successfully in cooperation with

experts for multiple hydrogeological studies, including

model regions in Germany [27], China [36], Brazil [10]

and the Middle East [11]. While these publications fo-

cus on the data integration and modelling aspects using

the OpenGeoSys Data Explorer, this paper is centred on

visualisation and presentation aspects.

3.1 Typical Data Integration Problems

Difficulties often arise when attempting to concurrently

visualise multiple data sets acquired under different cir-

cumstances or using different devices. Data sets might

include artefacts due to errors during data acquisition

or transfer. More frequently, problems arise because of

inconsistencies between data sets. Typical examples for

inconsistencies include differences in scale, resolution,

coordinate systems and measurement techniques. Of-

ten these issues are interconnected: if two data sets

have large differences in scale, the data set with the

smaller extent often has a finer resolution. This might

be caused by the data being acquired using a different

device (satellite imagery vs GPS) or using a different

parameterisation. Also, different acquisition techniques

often have different offsets for measured parameters or

implicitly employ certain preprocessing algorithms or

coordinate systems. Likewise, if data is generated based

on other data sets, some of the original information

might be lost. For example, geometric data created us-

ing GIS is frequently missing elevation information be-

cause this is often considered insignificant in a 2D envi-

ronment using a sequence of layers. Other errors might

be caused by artefacts, such as cloud cover in satellite

imagery, stained sensors at observation sites or extreme

weather events. Errors might even be introduced into

the data after the actual measurement process, for ex-

ample due to inaccuracy during meter-reading, using

a wrong projection when harmonising the coordinate

systems of multiple data sets, transposing digits when


manually copying or transcribing data, or OCR-errors

when digitalising data. Finally, problems might arise

in the model creation process itself when choosing a

parameterisation of the finite element mesh, merging

similar soil types, or interpolating the subsurface based

on borehole information.

Even though several of the problems described above

occur quite frequently, most are too complex to check or

fix automatically. While automated data modification

algorithms are employed for some common issues where

necessary modifications are unambiguous (see [27] for

details), for the most part we take advantage of a con-

current visualisation of relevant data sets. Experts can

often easily detect potential problems in a 3D environ-

ment either instantly or after some inspection of the

data given a set of basic tools.

3.2 Visualisation

As previously mentioned, visualisation of input-, model-

and simulation-data is a valuable help during every step

in the modelling workflow. It is an important means for

initially detecting potential problems, evaluating modi-

fied data sets and comparing them to the original data,

and verifying the plausibility of simulation results. With

some additional preparation, such an integrated scene

consisting of multiple data sets can then also be used

for presenting the progress or results of this particular

study to other scientists or stakeholders.

The OGS Data Explorer framework supports a large

number of file formats commonly used in the creation

of a simulation model, including ESRI shape files [6],

various raster- and mesh file formats as well as files

containing borehole stratigraphies or other geological

information. After loading, every data set (except time

series information provided by data loggers) is rendered

in 3D space.

The two basic classes of input data sets are vector-

and raster-data. Vector data typically consists of in-

formation given in the form of points and (poly-)lines.

While the data can be displayed in exactly this form,

it is often hard to actually see such structures in a

complex scene consisting of many data sets. To ensure

that this data (e.g. boreholes, logger positions or even

streams) remains visible, we typically represent point

data using glyphs (typically spheres or arrows) while

geometric lines (i.e. boreholes or streams) are rendered

as tubes. While a number of parameters can be adjusted

for displaying vector data in such a way, the radius r

of glyphs and tubes is by far the most important for an

adequate visualisation. The extend of model regions we

have worked with in the past varies significantly, rang-

ing from less than one hectare (ca. 2.5 acres) to more

than a million square kilometres [28]. Also, large differ-

ences of the E in different dimensions of a macroscale

hydrological modelmax{Ex,Ey}

Ez> 100 are quite com-

mon while for geothermal applications Ez might easily

be the largest extent. In our framework we have set the

default radius r = maxE150 , which has proven a good size

for structures to be instantly visible without appearing

too large or concealing to much of the surrounding area.

The user can easily adjust a number of parameters for

each data set individually, including the radius, colour,

opacity, etc. An application of these techniques is de-

picted in figure 3 where lines and points would be barely

visible without enlarging the data and using noticeable

colours. An application for the comparision of two geo-

metric data sets using different glyph size is illustrated

in figure 3c.

In contrast, raster data can contain a far wider range

of information. In general, it can be mapped as texture

onto surfaces. This is especially useful if the information

is of statistical nature such as the classification of land

use classes, or the distribution of social or economic

parameters. If the information can be interpreted in a

geometric sense, we typically use our framework to con-

struct geometric objects to represent this information.

Examples include creating surface meshes from digital

elevation models as well as cloud or humidity distribu-

tions from climate data [13]. DEMs in combination with

interpolation of borehole stratigraphies are the basis for

3D subsurface models [27] that are also used later for

the actual simulation of phenomena. Thus, simulation

related data forms a third class of data to be considered

for visualisation purposes.

As such, volume data can easily occlude a lot of

other information within the subsurface domain. We

employ transparency to avoid this issue or reduce the

visualisation to layer boundaries only. Another solution

utilises clipping-planes which can be set to see cross sec-

tions of subsurface models or simulation results. To fo-

cus on interesting subdomains within a data volume, it

is also possible to apply thresholds to a mesh based on

any parameter associated with nodes or elements, such

as materials, groundwater recharge, temperature, etc.

To intuitively represent various kinds of flow stream-

lines, iso-surfaces or arrow-glyphs can be employed (see

figure 4a). All of the above techniques can also be com-

bined to ensure a maximum of flexibility for presenting

case studies and their results.

The complete scene containing combinations of all

the types of data sets described above can be continu-

ously rotated and zoomed, each data set can be individ-

ually scaled and translated to account for differences in

offset or extent. Colours or lookup tables can be indi-

vidually chosen and opacity can be adjusted to ensure


basic visibility and make comparison between data sets

as simple as possible. Many of the potential problems

described in the last section become apparent instantly,

the most common examples being different coordinate

systems, missing elevation information in shape files,

false offsets or intersecting subsurface layers.

We employ the Visualization Toolkit for 3D visu-

alisation as it offers a wide range of functionality for

the transformation and modification of objects. In ad-

dition, the VTK library is extendable and allows for the

implementation of new functionality as well as the defi-

nition of pipelines for applying a workflow consisting of

a sequence of modifications to a given data set. While

the library implements a large number of visualisation

techniques, the algorithms described above for the var-

ious types of geoscientific data sets are the ones applied

most frequently.

3.3 Data Mapping

If a data set that might potentially present a prob-

lem during model creation or simulation, that data set

should be modified in such a way that the potential is-

sue is less likely to occur. Such modifications to the data

are always considered if experts expect the simulation

result to benefit from the change. That includes not

only modifications to data sets containing artefacts (as

shown in section 3.1) but might also result in modify-

ing actually accurate input files if they are inconsistent

with the rest of the data or might otherwise present

a problem later in the workflow. Examples are surface

triangles with a bad edge ratio that might result in os-

cillations as well as geometric information that does not

fit the finite element mesh, resulting in an incomplete

application of boundary conditions during the simula-

tion. The detection of degenerated mesh elements has

been thoroughly discussed in [27], the focus here will

be on a number mapping algorithms for various cases

to map points and lines or surfaces based on another

data set to remove inconsistencies or artefacts.

For distinction we will in the following refer to the

data set that is modified (i.e. mapped) as A and the

data set that A is mapped on as B. In general, a map-

ping of data set A can be performed if a data set B

exists such that Pxy(A)⋂Pxy(B) 6= ∅, where Pxy(X)

denotes the projection of X onto the xy-plane. The re-

sult of such a mapping operation is the modified data

set PB(A). In the algorithms proposed below a point

p ∈ A is mapped only if Pxy(p) ∈ Pxy(B). Otherwise

no value for p can be deducted from B and the point

is ignored (i.e. not modified). This allows to perform a

mapping of A unto B even if Pxy(A) 6⊂ Pxy(B) or even

Pxy(B) ⊂ Pxy(A).

Algorithm 1 - Mapping data on a raster: The ele-

vation information for data sets contained in a model is

usually extracted from a digital elevation model B. The

pixel size of DEMs can vary considerably. While free

DEMs such as NASA SRTM [7] or ASTER GDEM [37]

have a pixel size of 90 metres and 30 metres, respec-

tively, high resolution DEMs with a pixel size of one or

two metres are sometimes available from state offices or

other administrative organisations. Geometric points or

mesh nodes p ∈ A are assigned an elevation based on

the pixel q ∈ B with Pxy(p) ∈ q. To account for steep

gradients or artefacts in the DEM we do not assign the

value of q directly but use a 2D linear interpolation be-

tween q and its neighbours for calculating the actually

assigned elevation value (fig. 2a).

While Algorithm 1 correctly maps each point of

the data set based on the DEM, elevation of data sets

PB(A1) and PB(A2) need not be consistent. Polyline

data A1 representing observation sites, streams or bore-

holes might not match the surface mesh A2, even if both

have been mapped based on the same DEM B (see fig-

ure 3a). The position and local distribution of points

pi ∈ A1 may vary considerably from qj ∈ A2, therefore

the interpolated elevation might differ and the slopes

of line segment and triangle edges might vary consider-

ably depending on the position of their end points. To

minimise this effect, we generally try to integrate all rel-

evant geometric information (such as provided by A1)

during the generation of the surface mesh A2 (see [27]

for details on mesh generation). However, to get a con-

sistent result requires the typical segment length of

polylines in A1 to be roughly similar to the desired edge

length of triangles forming surface A2. If the local num-

ber of geometry points is significantly smaller or larger

than that of the mesh nodes, geometry points will form

plateaus on or span bridges across surface triangles, re-

spectively. This disparity grows with larger gradients

on the surface as well as with the difference between

the number of local geometric points and mesh nodes.

Algorithm 2 - Mapping data on a mesh: The pre-

ferred workflow consists of mapping the surface mesh

A2 based on the DEM B and afterwards mapping the

geometry A1 on C = PB(A2). A collision detection al-

gorithm will set the elevation of point p ∈ A1 to the

intersection of a vertical line l through p with C, i.e.

PB(p) = l(p)⋂

PB(A2) (fig. 2b). That way, all geomet-

ric points are matched exactly onto the surface, even if

the position of geometric points varies from the posi-

tion of mesh nodes or there are locally more geometric

points than mesh nodes. The only potential problem left

occurs if the distance of consecutive points in a poly-

line is larger than the edge length of the local triangles


(a) Mapping based on DEM (b) Mapping based on mesh (c) Additional geometric points

Fig. 2: Schematic of mapping algorithms implemented in the OpenGeoSys Data Explorer. (a) Calculation of

geometric point elevation based on weighted DEM pixels. (b) Projection of points onto mesh surfaces via triangle

line intersection. (c) Inserting additional points into geometry at intersections of geometric lines with mesh element

edges and nodes. Newly inserted points are marked in red.

forming the surface (see fig. 3b). Again, the disparity

is larger in the presence of large gradients. Therefore,

our algorithm will optionally insert additional points

into a polyline whenever the line intersects with trian-

gle edges or nodes and these additional points A+1 will

be mapped along with the original points in A1 (fig. 2c).

The resulting geometry A1 = PC(A1

⋃A+

1 ) is guaran-

teed to fit the surface perfectly (see figure 3) and the

subsequent increase of the number of geometric points

has in all test cases been insignificant in comparison to

the number of mesh nodes.

In the same way, a mesh A can also be mapped

based on a second mesh B. The mesh nodes p ∈ A are

handled the same way as geometric points in the pre-

vious example. This has been useful in the past if reli-

able information exists locally with Pxy(B) ⊂ Pxy(A).

While A is still required to represent the model area,

the topography of A can be improved by integrating in-

formation from B. An application for this algorithm is

shown in figure 8, where the values of a DEM acquired

via satellite are modified based on the much more reli-

able data of a bathymetry aquired via sonar.

Low pass filtering of surfaces: While not techni-

cally a mapping algorithm, we would like to add that

our framework also allows to apply a low pass filter on

surface meshes. If a surface triangulation A is signif-

icantly finer than the DEM B it is mapped on using

Algorithm 1, the result is an oversampling of B which

gives rise to step artefacts in PB(A) (see for example

figure 6c). This effect can be reduced or even avoided

by low pass filtering over the elevation of mesh nodes

p ∈ PB(A). The new elevation of p is given by

pz =1

|N |+ 2

(2pz +

∑x∈N

xz

)(1)

where N is the set of all nodes x directly connected to

p by a triangle edge. By weighting pz twice as much

as the neighbours of p the smoothing effect is approx-

imately that of a binomial or gaussian low pass filter.

Adjusting the influence of neighbours by distance to

p has been neglected since A is a finite element mesh,

including the common constraint that the size of neigh-

bouring elements must not vary in size by more than

30%. Application of such a low pass filter will reduce

outliers as well as step artefacts in PB(A), thus given

a better representation of the surface for presentation

purposes. The difference of simulation results using the

mapped surfaces PB(A) and the filtered surface PB(A)

have not been investigated yet but since step artefacts

are not a good representation of the actual surface of

the model region, results using A are expected to give

at least as good an approximation of simulated param-

eters as using A.

3.4 Presentation

We employ various methods for presentation of virtual

scenes of case studies that have progressed to a point

where it makes sense to discuss them in a larger group

to gain further insight or to formally present its re-

sults. The focus of our work is presentation in a vir-

tual reality centre with a 6x3 metre video wall using

stereoscopic projection and an optical tracking system

(Fig. 4a). Presentations within this environment are

used frequently for discussions between scientists with


(a) Geometry mapped based on DEM (b) Geometry mapped based on mesh

(c) Additional geometric points (d) Concurrent view of all algorithms

Fig. 3: Comparison of geometry mapping on a high-gradient mesh. The green geometry G is mapped on the

underlying DEM, also used to create the depicted surface mesh via Algorithm 1. The blue geometry B is mapped

directly on the surface via Algorithm 2. The red geometry R is mapped via the extended version of Algorithm

2 : Additional points have been inserted into R where the polyline intersects with triangle edges or nodes before

it has been mapped on the mesh. When projected into the xy-plane, the polylines of all three geometries are

identical. Given the underlying mesh, only R fits perfectly due to the mapping of the additionally inserted points.

(a) Comparison of geometries G and R, (b) comparison of geometries B and R, (c) top-down view for illustration

of additional points inserted into R in comparison to B, (d) a wider view of the data with a concurrent visualisation

of all three mappings on a semitransparent mesh. Note, that for surfaces with low gradients all proposed algorithms

give very similar results. The large differences visible in this example are due to the model being a mountainous

region with sudden changes in elevation.

a different background. We also present research results

to stakeholders during project meetings or inform the

general public during open day events.

For navigation within this environment and demon-

stration of models and simulation results we employ two

respective software systems: an OpenSG framework as

well as the Unity game engine. Special input devices

allow users to move within the data displayed in the

virtual reality environment. We integrated support for

3D mouse and flystick to make navigation within the

scene as intuitive as possible.

For both programmes we implemented functionality

typically needed during such presentations. Most im-

portantly, predefined viewpoints can be accessed that

take the user to a specific point in the scene, with a

predefined view of a certain interesting aspect in the

data. While such viewpoints could also be accessed us-

ing the aforementioned input devices, a smooth and tar-

geted transition to that point usually looks more pro-

fessional and finding that specific position and perspec-

tive is much more reliable. Viewpoints can be connected

via paths, so by defining a sequence of notable view-

points and connecting them accordingly, a presentation

can be conducted without actually using interaction de-

vices. While the OpenSG framework only supports di-

rect connections between viewpoints, Unity supports

the definition of arbitrary camera paths via polylines

or splines for convenient transit between two specific

points or the rotation around a certain object. At any

point on such a path objects can be set to appear or


Fig. 4: (left) Stereoscopic visualisation of an OpenFOAM and Min3P simulation [40] of stream flow in the hyporheic

zone in a virtual reality environment. (Data kindly provided by N. Trauth), (right) Visualisation of the Rappbode

reservoir in Unity, including camera paths and visual effects, employed for interactive presentations.

fade away. This allows to focus on data sets that are

particularly interesting at this part of the presentation

and to hide objects or groups of objects when they are

not needed. The velocity of movement along the camera

path as well as the time delay for fading objects can be

adjusted individually.

For data sets containing geometric information that

have been enhanced using glyphs and tubes (i.e. Obser-

vations sites, streams, boreholes, etc. – see section 3.2)

we often include more than one representation. When

looking at the complete scene it is necessary that, for

instance, river networks can be seen right away and are

therefore represented using a rather large radius. When

the user zooms into the scene to see details, these thick

tubes would look rather chunky. Having multiple rep-

resentations of one data set allows to switch to another

representation consisting of tubes with a smaller radius

during the zoom process. The cross-fading possibilities

of Unity are again very helpful here, but the basic con-

cept also works for general frameworks.

Similar techniques can be applied to mesh represen-

tations, e.g. of surfaces. Meshes with fewer triangles can

still adequately capture the basic layout of surface from

further afar while refined meshes are needed for a closer

look at the data. In contrast to the use of multiple rep-

resentations detailed before to ensure an optimal visual

impression for geometric data, the governing factor for

surfaces is not one of visibility but performance. As all

data is loaded locally via a fast local network, adding

multiple resolutions of meshes to a scene only makes

sense for very large data sets. In our experience these

are rare, as the quality of the input data often does not

justify an exceedingly fine discretisation of the domain.

Interaction tools allow users to select 3D objects to

see additional information. This includes the placement

of clipping planes within data sets as well as the opening

of context menus to display additional information or

even photographs of the real-world location to get a

better impression of what kind of environment the data

set actually represents.

In addition, the frameworks we employ for presenta-

tion also allow adding “cosmetic” effects to the presen-

tations. These do not enlarge the actual amount of in-

formation but make the visualisation easier to compre-

hend for the audience by employing familiar metaphors

for certain aspects of the scene. Examples include the

display of a sky with clouds or sunshine, or a realistic

display of reflecting surfaces of water bodies by assigned

adequate shaders (Fig. 4b), etc.

4 Case Study

As an example for the proposed techniques we present

a visualisation project containing more than 70 data

sets acquired within the TERENO initiative [43] in

Germany. TERENO is concerned with the prediction

of possible impacts of climate change. Four areas in

Germany have been selected and are now heavily in-

strumented to allow extensive studies and simulations

for researchers from many different disciplines. Hydro-

geological analysis in central Germany is being con-

ducted in the catchment of the River Bode with a size of

3,100 km2. Within that area a number of intensive test

sites have been selected, such as the catchment of the

Rappbode reservoir system, used as a source of drink-

ing water and electricity generation. The data sets con-


Fig. 5: Overview over the

Rappbode catchment and

reservoir system. Details

on regions of interest (a)-

(c) can be found in fig-

ure 6, region (d) is dis-

cussed in figure 7. (Stream

information from ATKIS

DLM [1])

(a) (b) (c)

(d) (e) (f)

Fig. 6: Results of geometry mapping and lowpass filtering on unstructured surface meshes based on algorithms

presented in section 3.3.


Fig. 7: (left) Contrast-enhanced DEM containing artefacts due to reflections of the surface of the water body

outlined in blue. (centre) The resulting surface representation distorts the actual surface of the water body, the

coast line seems different and a number of “islands” are visible. (right) After correcting the surface based on sonar

data (see figure 8) the reservoir is depicted correctly.

Fig. 8: Example for removal of artefacts using mesh-on-mesh mapping: (left) The surface mesh generated from

the original DEM includes prominent artefacts. (centre) Bathymetry of the Rappbode reservoir with a resolution

of 5 m acquired via sonar. (right) Corrected mesh using sonar data and lowpass filter on mesh node elevation.

(Bathymetry provided by Landesbetrieb fur Hochwasserschutz und Wasserwirtschaft, Sachsen Anhalt).

cerned with the project range from raster data (DEM,

soil moisture maps, land use classification, etc.) cover-

ing the complete area to observation sites covering only

a few square metres or a small number of points (i.e.

sensors). A surface mesh of the whole region consisting

of 1.08 million triangles (max. edge length 90 m) has

been created based on the ASTER-DEM [37], acquired

via satellite and with a pixel size of 30×30 m. For inten-

sive test sites refined surface representations have been

created. The mesh for the catchment of the Rappbode

reservoir system has 930 000 triangles with a maximum

edge length of 30 m. At such a small scale the inaccu-

racies of data sets are easily noticeable. Figure 5 gives

an overview over the region. The default colour lookup

table (LUT) used by our framework (Fig. 1) has been

replaced with a customised LUT more suited for this

region. The river network in the north-western part ap-

pears more prominent than the rivers in south-eastern

part due to a larger radius of the tubes representing

the stream network, to ensure better visibility amid the

mountainous region. Observation sites are marked by

red dots, also enlarged to be visible at this scale. A num-

ber of regions of interest have been in selected figure 5 to

demonstrate the application of algorithms proposed in

section 3. Detailed illustrations of these regions can be

found in figure 6. While generally the simple mapping of

geometry on meshes gives good results (see Algorithm 2

in section 3.3), there are a number of problems visible

in figures 6a and 6b. Due to the large surface gradients

in the mountains and the irregular distribution of geo-


metric points forming the polylines, parts of the rivers

represented by those polylines are occluded by the sur-

face (Fig. 6a). This effect is even more noticeable when

the river course does not quite match the river valley ex-

tracted from the DEM (Fig. 6b). Therefore, additional

points have been inserted into the river geometry before

mapping at the intersections of geometric lines with the

edges of surface triangles. This fixes the problem in both

cases (Figs. 6d and 6e). The original stream geometry

data set contains 529 points. Including the additional

points results in a data set with 6505 points. Although

the number of points has increased by a magnitude,

the actual amount of data is negligible in comparison

to the ca. 470 000 surface mesh nodes. Even though the

river in figure 6e does still not match the valley, it is

clearly visible now and will not be overlooked during

assessment of the data. Fixing this remaining problem

of the river not fitting the valley is not straightforward,

however, as the other streams from this data set match

their respective valleys and it is not clear if the error is

originating in the DEM or the stream data set.

Since the edge length of triangle elements forming

the surface is equal or smaller than the resolution of the

underlying DEM, step artefacts are visible when map-

ping the surface mesh (Fig. 6c). Lowpass filtering using

the direct neighbours of each node almost completely

remove this effect and results in a much smoother sur-

face that allows a better assessment of topological fea-

tures such as ridges and valleys (Fig. 6f).

Figure 7 illustrates the difficulty of creating a sur-

face representation from a DEM in the presence of arte-

facts, in this case due to reflection on the water surface

of the reservoir. Figure 8 shows the effect of re-mappingparts of the surface based on sonar data. These data sets

have a resolution of three to five metres, respectively,

and are much more reliable than the satellite imagery.

In addition, a low pass filter has been applied to the

surface, removing step artefacts and, again, resulting

in a smoother surface.

Besides the DEM and stream data, this case study

includes a number of sensor positions at the in- and

outlet of each reservoir (see figure 5). These sensors

are constantly monitoring water levels, temperatures

at various depths, as well as the spectral absorption

for determining the amount of dissolved organic carbon

within the water bodies [29]. For instance, the diagram

in figure 1 shows time series data that is attached to

an observation site. The selected time interval depicts

the increase in dissolved organic carbon (measured via

spectral absorption coefficient at 256 nm (SAK254)) af-

ter rising water levels due to heavy rainfall. Correla-

tions such as this are important for the quality of drink-

ing water extracted from the reservoir. Simulations are

planned based on these and other data sets to make

predictions about the development of the water qual-

ity. Data sets on the interaction of streams with the

groundwater are measured for small sections of rivers

in this region and employed for separate visualisation

studies (see figure 4a).

5 Conclusions

We presented a number of visualisation strategies for

the concurrent visualisation of geoscientific and mod-

elling data using the OpenGeoSys Data Explorer frame-

work . The proposed system renders heterogeneous data

sets in 3D space and offers functionality for meshing,

mapping and filtering data sets. Visualisation supports

users prior to model generation or simulation when check-

ing input data sets for potential problems due to arte-

facts or inconsistencies and while re-analysing the data

after modification based on previously found issues. Af-

terwards simulation results can be added to that same

visualisation for checking plausibility of simulation re-

sults as well as presenting the progress or results of

this particular study to other scientists or stakehold-

ers. The 3D visualisation is implemented using VTK

and applying additional algorithms for mapping and en-

hancement of aspects within the data. For presentations

in a virtual reality environment an OpenSG framework

and the Unity engine are employed. Additional func-

tionality has been implemented in both environments

to allow navigation and interaction with specific data

sets using specialised input devices. Visual Effects are

added to those presentations for making them more ap-

pealing to the audience. Future enhancements of the

system are added in coordination with experts and in-

clude user guidance during the modelling process to

allow semi-automatic modification of data sets as well

as additional scripting for a more straightforward han-

dling of the Unity engine.

Acknowledgements The research has been supported by theHelmholtz Association with the programme “Earth and En-vironment” and the TERENO initiative (Terrestrial Envi-ronmental Observatories). The authors would also like tothank Karsten Rinke, Nico Trauth, Christian Schmidt andUte Wollschlager for providing the data sets presented in thecase study.

References

1. ATKIS DLM Geobasisdaten (2008). http://www.bkg.

bund.de2. Ahrens, J., Geveci, B., Law, C.: ParaView: An End-

User Tool for Large Data Visualization. In: C. Hansen,C. Johnson (eds.) In the Visualization Handbook. Else-vier (2005)


3. Alcaraz, S., Lane, R., et al.: 3d geological modelling us-ing new Leapfrog Geothermal software. In: Proc of 36thWorkshop on Geothermal Reservoir Engineering (2011)

4. Bremer, J.E., Harter, T.: Domestic wells have high prob-ability of pumping septic tank leachate. Hydrol EarthSyst Sci 16(8), 2453–2467 (2012)

5. DHI-WASY Software: FEFLOW6 User Manual - FiniteElement Subsurface Flow and Transport Simulation Sys-tem. http://www.feflow.info (2010)

6. ESRI: Shapefile Technical Description. http://www.esri.com/library/ (1998)

7. Farr, T.G., Rosen, P.A., Caro, E., et al.: The ShuttleRadar Topography Mission. Rev Geophys 45, RG2004(2007)

8. Forlines, C., Wittenburg, K.: Wakame: Sense Making ofMulti-Dimensional Spatial-Temporal Data. In: Proc. ofInt Conf on Advanced Visual Interfaces, pp. 33–40 (2010)

9. Goldstone, W.: Unity 3.x Game Development Essentials(2nd Edition). Packt Publishing (2011)

10. Goncalves, T.D., Fischer, T., Grabe, A., et al.: Ground-water flow model of the Pipiripau watershed, Federal Dis-trict of Brazil. Environ Earth Sci 69(2), 617–631 (2013)

11. Grabe, A., Rodinger, T., Rink, K., et al.: Numerical anal-ysis of the groundwater regime in the western Dead SeaEscarpment, Israel + West Bank. Environ Earth Sci69(2), 571–585 (2013)

12. Grundmann, J., Schutze, N., Schmitz, G.H., Al-Shaqsi,S.: Towards an integrated arid zone water managementusing simulation-based optimisation. Environ Earth Sci65(5), 1381–1394 (2012)

13. Helbig, C., Rink, K., Bauer, H.S., et al.: Concept andworkflow for 3D visualization of atmospheric data in avirtual reality environment for analytical approaches. En-viron Earth Sci (2014). (submitted)

14. Helbig, C., Rink, K., Marx, A., et al.: Visual integrationof diverse environmental data: A case study in CentralGermany. In: Proc. of Int Congress on EnvironmentalModelling and Software. Managing Resources of a Lim-ited Planet: Pathways and Visions under Uncertainty,Sixth Biennial Meeting, Leipzig, Germany, pp. 2350–2357(2012)

15. Hotzl, H., Moller, P., Rosenthal, E.: The Water of theJordan Valley. Springer (2009)

16. Huenges, E.: Geothermal Energy Systems: Exploration,Development, and Utilization. Wiley-VCH (2010)

17. Janicke, H., Bottinger, M., Mikolajewicz, U., Scheuer-mann, G.: Visual Exploration of Climate VariabilityChanges Using Wavelet Analysis. IEEE Trans VisualComput Graph 15(6), 1375–1382 (2009)

18. Jones, R.R., McCaffrey, K.J.W., Clegg, P., et al.: Integra-tion of regional to outcrop digital data: 3d visualizationof multi-scale geological models. Comput Geosci 35(1),4–18 (2009)

19. Kolditz, O., Bauer, S., Bilke, L., et al.: OpenGeoSys:An open source initiative for numerical simulation ofthermo-hydro-mechanical/chemical (THM/C) processesin porous media. Environ Earth Sci 67(2), 589–599(2012)

20. Law, M., Collins, A.: Getting to Know ArcGIS for Desk-top (3rd Edition). Esri Press (2013)

21. Lawrence Livermore National Laboratory: VisIt: an in-teractive parallel visualization and graphical analysistool. https://wci.llnl.gov/codes/visit/

22. Ling, M., Chen, J.: Environmental Visualization: Ap-plication to Site Characterization, Remedial Programs,and Litigation Support. In: Proc of EuroVis Workshop

on Visualisation of Environmental Data, pp. 37–41. Eu-roGraphics Digital Library (2013). DOI 10.2312/PE.EnvirVis.EnvirVis13.037-041

23. Marx, A., Mast, M., Knoche, R., Kunstmann, H.: Globalclimate change and regional impact on the water balance- Case study in the German alpine area. Wasserwirtschaft98(9), 12–16 (2008)

24. Neteler, M., Mitasova, H.: Open Source GIS: A GRASSGIS Approach (3rd Edition). Springer, New York (2008)

25. Potter, K., Wilson, A., Bremer, P.T., et al.: Visualiza-tion of uncertainty and ensemble data: Exploration of cli-mate modeling and weather forecast data with integratedViSUS-CDAT systems. J Phys Conf 180(1) (2009)

26. Reiners, D., Voss, G., Behr, J.: OpenSG: Basic Concepts.In: 1. OpenSG Symposium (2002)

27. Rink, K., Fischer, T., Selle, B., Kolditz, O.: A Data Ex-ploration Framework for Validation and Setup of Hydro-logical Models. Environ Earth Sci 69(2), 469–477 (2013)

28. Rink, K., Kalbacher, T., Kolditz, O.: Visual Data Ex-ploration for Hydrological Analysis. Environ Earth Sci65(5), 1395–1403 (2012)

29. Rinke, K., Kuehn, B., Bocaniov, S., et al.: Catchmentsas Sentinels: The Rappbode Reservoir Observatory (HarzMountains Germany). Environ Earth Sci 69(2), 523–536(2013)

30. Schanze, J.: Dealing with future change in flood risk man-agement. Journal of Flood Risk Management 5(1), 1–2(2012)

31. Scheibe, T.: Interactive models for ground water flow andsolute transport. Groundwater 42(1), 8–11 (2004)

32. Schroeder, W., Martin, K., Lorensen, B.: VisualizationToolkit: An Object-Oriented Approach to 3D Graphics(4th Edition). Kitware, Inc. (2006)

33. Schuchardt, K.L., Agarwal, D.A., Finsterle, S.A., et al.:Akuna integrated toolsets supporting advanced subsur-face flow and transport simulations for environmentalmanagement. In: Proc of Int Conf on ComputationalMethods in Water Resources (CMWR 2012), pp. 59–63(2012)

34. Schumann, H., Tominski, C.: Analytical, visual and in-teractive concepts for geo-visual analytics. J Vis LangComput 22(4), 257–267 (2011)

35. Stockmann, M., Hirsch, D., Lippmann-Pipke, J., Kupsch,H.: Geochemical study of different-aged mining dumpmaterials in the Freiberg mining district, Germany. En-viron Earth Sci 68(4), 1153–1168 (2013)

36. Sun, F., Shao, H., Wang, W., et al.: Groundwater dete-rioration in Nankou – a suburban area of Beijing: dataassessment and remediation scenarios. Environ Earth Sci67(6), 1573–1586 (2012)

37. Tachikawa, T., Kaku, M., Iwasaki, A., et al.: ASTERGlobal Digital Elevation Model Version 2 – Summaryof Validation Results. Tech. rep., NASA Jet PropulsionLaboratory, California Institute of Technology (2011)

38. Tenzer, H., Park, C.H., Kolditz, O., McDermott, C.I.:Application of the geomechanical facies approach andcomparison of exploration and evaluation methods usedat Soultz-sous-Forts (France) and Spa Urach (Germany)geothermal sites. Environ Earth Sci 61(4), 853–880(2010)

39. Tominski, C., Donges, J., Nocke, T.: Information Visual-ization in Climate Research. In: Proc of 15. Int Conf onInformation Visualisation, pp. 298–305 (2011)

40. Trauth, N., Schmidt, C., Maier, U., et al.: Coupled 3Dstream flow and hyporheic flow model under varyingstream and ambient groundwater flow conditions in apool-riffle system. Water Resour Res (2013). DOI10.1002/wrcr.20442


41. Walther, M., O.Delfs, J., Grundmann, J., et al.: Saltwa-ter intrusion modeling: Verification and application to anagricultural coastal arid region in Oman. J Comput ApplMath 236(18), 4798–4809 (2012)

42. Wycisk, P., Hubert, T., Gossel, W., Neumann, C.: High-resolution 3D spatial modelling of complex geologicalstructures for an environmental risk assessment of abun-dant mining and industrial megasites. Comput Geosci35(1), 165–182 (2009)

43. Zacharias, S., Bogena, H., Samaniego, L., et al.: A net-work of terrestrial environmental observatories in Ger-many. Vadose Zone J 10(3), 955–973 (2011)


visualisation strategies for environmental modelling datakarsten/data/rink2014vis.pdf ·...

Documents