InCorr: Interactive Data-Driven Correlation Panels for Digital Outcrop Analysis
Thomas Ortner, Andreas Walch, Rebecca Nowak, Robert Barnes, Thomas Höllt, Eduard Gröller
TTo appear in IEEE Transactions on Visualization and Computer Graphics
InCorr: Interactive Data-Driven Correlation Panelsfor Digital Outcrop Analysis
Thomas Ortner, Andreas Walch, Rebecca Nowak, Robert Barnes, Thomas H ¨ollt, and Eduard Gr ¨oller a b c
Fig. 1. System overview of InCorr: (a) Outcrop View, showing digital outcrop model annotated in 3D by a geologist and vertical 3Dlogging tool indicating layer thicknesses. (b) InCorrPanel, showing logs for two outcrops created directly from annotation data mimickingmanual illustrations. (c) GUI to assign rock types to rock layers
Abstract — Geological analysis of 3D Digital Outcrop Models (DOMs) for reconstruction of ancient habitable environments is a keyaspect of the upcoming ESA ExoMars 2022 Rosalind Franklin Rover and the NASA 2020 Rover Perseverance missions in seekingsigns of past life on Mars. Geologists measure and interpret 3D DOMs, create sedimentary logs and combine them in ‘correlationpanels’ to map the extents of key geological horizons, and build a stratigraphic model to understand their position in the ancientlandscape. Currently, the creation of correlation panels is completely manual and therefore time-consuming, and inflexible. With InCorrwe present a visualization solution that encompasses a 3D logging tool and an interactive data-driven correlation panel that evolveswith the stratigraphic analysis. For the creation of InCorr we closely cooperated with leading planetary geologists in the form of adesign study. We verify our results by recreating an existing correlation analysis with InCorr and validate our correlation panel against amanually created illustration. Further, we conducted a user-study with a wider circle of geologists. Our evaluation shows that InCorrefficiently supports the domain experts in tackling their research questions and that it has the potential to significantly impact howgeologists work with digital outcrop representations in general.
Index Terms —Geographic/Geospatial Visualization, Remote Sensing Geology, Digital Outcrop Analysis, Integration Spatial andNon-Spatial Data Visualization
NTRODUCTION
Geological analysis of image data collected by stereo camera systemson Mars rovers has proven to be a valuable tool in reconstructing an-cient environments on Mars. This is a key aspect of the upcomingESA ExoMars 2022
Rosalind Franklin
Rover and the NASA 2022Rover
Perseverance missions [5, 17]. A shared aim of both of thesemissions is to drill and sample rocks, which were deposited in ancientenvironments that scientists deem may have been habitable. This isdetermined largely by the geological characteristics. Within geology,the field of sedimentology is concerned with the analysis of texturesand internal fabrics of rocks formed by deposition and movement of • Thomas Ortner, Andreas Walch and Rebecca Nowak are with VRVisZentrum fr Virtual Reality und Visualisierung Forschungs-GmbH.E-mail: { ortner | walch | rnowak } @vrvis.at.• Robert Barnes is with Imperial College, London.E-mail: [email protected].• Thomas H¨ollt is with TU Delft.E-mail: [email protected]• Eduard Gr¨oller is with Technische Universit¨at Wien.E-mail: [email protected]. sand, silt, and clay sediment by wind, water, or ice. Burial and exposureto high heat, pressure and fluid circulation leads to the formation ofsedimentary rocks. Stratigraphy is a field concerned with document-ing and interpreting the vertical distribution of different textures andsedimentary structures in order to reconstruct the evolution of the en-vironments, which formed the sedimentary layers, or strata . Thereis strong scientific evidence, that there were rivers and lakes activeat the surface in the distant past [8]. Assuming analogous processeson Mars as on Earth, lake and river deposits are most promising forscientists to discover biosignatures. Therefore, stratigraphic analysisand correlation of observations between distant locations is essential.The main goal of geologists is to combine their observations tobuild a geological model, which encompasses the temporal evolutionof environmental processes of a region. They build such a model frommeticulously annotating and measuring textural features in a numberof outcrops , i.e. rock faces exposing strata as illustrated in Fig. 2, anddetermining their relative age-relationships. It is generally accepted thatyounger sediments are deposited on older sediments, therefore changesin rock characteristics are a record of its past. Correlation of theseobservations across large areas allows for regional evolutionary modelsto be built. Traditionally, geologists make measurements ‘in the field’1 a r X i v : . [ c s . G R ] J u l nit ContactBed ContactCross Bed b c SSWNNE
Sol 5Sol 2Sol 3Sol 4 N a Unit Bedset
Fig. 2. (a) Map of the Hanksville-Burpee Dinosaur Quarry campaign with its four outcrops along the canyon. (b) 3D triangulated mesh as digitaloutcrop model (DOM) of Sol2 with interpretation showing beds and units and their respective contacts. (c) cross bed measurements, which determinea layer’s deposition direction indicated by dip-and-strike disks using hands-on tools, such as a compass clinometer to measure surfaceorientation. The development of affordable remote sensing solutionshas prompted a sharp rise in geological analyses of digital outcropmodels (DOM) . DOMs present the geologists with 3D triangulated, andoften photographically textured, surfaces of rock outcrops. Applicationof these techniques to processed stereo-camera image data collected byrovers on Mars can be used to greatly enhance our understanding of theevolution of the planet and drive future robotic exploration missions.Robust and efficient interpretations and measurements can be col-lected from DOMs in PRo3D [1]. These data are the foundation forcreating a regional geological model. Outcrops can be understood assampling locations, partially exposing a succession of strata, whichin fact may extend over hundred thousands of square kilometers un-derneath a planet’s surface. To build a regional geological model,geologists look for occurrences of the same stratum in multiple out-crops. First, they characterize the succession of exposed strata froman outcrop by creating a geological log , as shown in Fig. 1a. Afterrepeating this for each outcrop, they combine all logs into a correlationpanel , and connect matching strata to form correlations , as indicated bythe red lines in Fig. 1b. Correlating strata across multiple outcrops thatcover a large area leads to a reliable characterization of the distributionof important geological units.
In a long lasting cooperation with planetary geologists, who are engagedin the geological survey of Mars as part of the scientific working groupsof ESA and NASA missions, we identified the following challengesthat come with remote geological analysis.At the moment, DOM visualization software is tailored towards the interpretation of outcrops, i.e. annotating and measuring strata. For fur-ther analysis, geologists typically rely on the export of measured valuesinto tools for statistical analysis and graph plotting, while they createcorrelation panels manually based on the obtained values. Geologistsoften use vector drawing software or they might draw the panels com-pletely by hand. In any case, the creation of correlation panels is verytime-consuming and therefore it is typically left to the end of the work-flow. However, only then potential weaknesses in the interpretationdata become apparent, such as, insufficient detail in the interpretationof a particular outcrop. This requires a tedious revisit of the inter-pretation stage and forces geologists to reorganize and largely redrawtheir panels. Further, the separation between digital tools enforces aseparation of interpreting and correlating, which would not be the case in the workflow of the traditional field work. Ultimately, this separationdisconnects correlation panel illustrations from the underlying anno-tations and measurements, which inhibits traceability, reproducibility,and reusability of analysis results throughout the geological sciencecommunity.
Based on these observations we came to the hypothesis that the work-flow of remote geological analysis would significantly benefit froman interactive and data-driven correlation panel, which achieves thefollowing goals:• G1 The panel should be completely data-driven and consequentlyevolve with the analysis with minimal effort.• G2 Allow geologists to relate visual representations in the corre-lation panel to the data they were created from.• G3 Such a panel should mimic manually illustrated correlationpanels including stylistic freedom, but without overwhelmingcustomization options.• G4 Using this panel needs to integrate well with the geologists’tool chain and workflow, otherwise it will not be used frequently.• G5 The panel should intuitively put 2D geological logs in contextto each other and the 3D outcrop interpretation.
As the primary contribution we present the visualization solution
InCorr , short for In teractive data-driven Corr elations. Based on thelong lasting cooperation with leading scientists in the field of planetarygeology, we conducted a design study to create a 3D geological loggingtool (Fig. 1a) and the
InCorrPanel , an interactive data-driven correla-tion panel (Fig. 1b). Both components are integrated into PRo3D [1],a tool for the geological interpretation of DOMs. We verify the ap-plicability of InCorr through a use case from a terrestrial campaignnear Hanksville, Utah, USA and we validate a correlation panel gener-ated with InCorr against a manually illustrated one based on the sameinterpretation data (Fig. 3). We further conducted a hands-on designvalidation to collect feedback from a broader range of geologists. An-other contribution is the introduction of interactive correlation panelsto the domain of remote geology analysis, which fosters traceability,reproducibility, and communicability of an otherwise static illustration.2 o appear in IEEE Transactions on Visualization and Computer Graphics
3D AnnotatedOutcrop Model Log Illustrated Manually a ebc d
Simplified Log Illustrated Manually
SSW NNE
Correlation Panel Illustrated Manually hf g hf g
Fig. 3. (left) annotated 3D outcrop model, followed by a sophisticated and simplified log representation, both illustrated manually with a vector drawingtool. (a) y-axis encodes true thickness of strata. (b) Rose diagram shows the distribution of dipping orientations (c) Individual styles and glyphs mayconvey rock characteristics. (d) Dashed lines convey uncertainty (e) x-axis encodes grain sizes logarithmically, directly relating to rock types. (right)Logs arranged in a manually illustrated correlation panel (f) with converging contacts. (g) Spatial distance between logs along geographic direction.(h) Fossil layer used as leveling horizon.
REVIOUS W ORK
The Petrel [24] software package for oil and gas exploration offers someoutcrop measurement capabilities and correlation panels. However, asthe software was originally intended for the analysis of seismic data inconnection with drill shafts, the logs present in these panels are createdfrom drilling wells and have a different visual encoding. Therefore,they are not suitable for performing outcrop-based correlation analysis.General purpose GIS or 3D visualization tools, such as ArcGIS [6]or Cloud Compare [4], are commonly found in geological publica-tions that include DOM analysis. These applications offer reliablemeasurement tools, but are neither targeted towards outcrop interpreta-tion applications nor do they support correlation analysis. Respectivepublications dealing with DOM interpretations typically describe aconcatenation of data transformations [22, 27]. A few specialized 3Doutcrop interpretation tools have recently emerged, including LIME [3],VRGS [11], or PRo3D [1, 29]. All three feature a tool set for creatingannotations and performing measurements on DOMs. Additionally,LIME and VRGS allow geologists to project manually illustrated logsonto the 3D surface. Nesbit et al. [19] use a 3D log in the context oftheir stratigraphic mapping. However, this logging is not integrated intoan interactive workflow and the measurements need to be translatedinto 2D logs manually.The visualization of geological phenomena has been an essential partof visualization research for decades, but mostly in the context of theanalysis of seismic data for oil and mining. Patel et al. [20] introducea tool to interpret 2D slices of seismic data from which they can pre-calculate horizon structures. H¨ollt et al. [12, 13] present an interactiveworkflow for interpreting the 3D data directly by incorporating welllogs retrieved from drilling. The steps of data retrieval, interpretation,well correlation and horizon extraction, and reservoir modeling [18]do align with the workflow of digital outcrop analysis, described inSect. 3, however the data, methods, and challenges differ significantly.Lidal et al. [16] focus on the communication of geological processesthrough visual stories and provide a sketch-based interface for thecreation thereof. To the best of our knowledge the field of digitaloutcrop analysis and the correlation of 3D outcrop interpretation datahas not been explored or addressed by the visualization community.
EOLOGICAL ANALYSIS OF D IGITAL O UTCROP M ODELS
The basis of remote virtual outcrop geology is the acquisition of 3D datafrom rock outcrops. In this section we will discuss the development ofoutcrop observations to regional geological models. We focus on the stages relevant to our work and introduce the geological concepts thatare important throughout this paper.
Outcrop interpretation (Sect. 3.1)produces contacts, which are the basis for logging (Sect. 3.2, Sect. 3.3).The result of the logging stage are geological logs, which are arrangedin a correlation panel to discover and create correlations (Sect. 3.4).Correlations are used to reconstruct geological surfaces, which arethen presented as 2D geological maps. The relevant geological termsintroduced in this section are summed up in Table 1.
Traditionally, geologists examine multiple outcrops in the field. At eachoutcrop they record textural characteristics, such as grain or crystalsize and shape, relative color, layer thickness, layering patterns, layerorientation, and boundary geometries. Relevant measurements, pho-tographs, notes, and outcrop sketches, as well as sketch logs or crosssections where necessary, are used to document the geology in the field,and analyzed out of the field for publication. The same process canbe applied to DOM analysis. After measuring the height and width ofan outcrop, geologists investigate contacts . A contact represents thedelineation where one layer of rock, i.e. stratum , ends and anotherone begins. These contacts can be numerous and are typically nested,so a stratum contains sub-strata, as shown by orange lines ( bedsets )between red lines ( units ) in Fig. 2b. Based on the different visiblerock characteristics, geologists meticulously trace contacts by drawingpolylines on the 3D surface along the discrete transitions between thedeposited strata. This allows them to characterize the architecture ofthe strata of an outcrop hierarchically into sub-structures of arbitrarydepth. Two contacts delimit a stratum that is homogeneous (to a certainextent). Geologists use different line thicknesses and colors to representthe magnitude of change between two adjacent strata. Geologists referto this process of identifying contacts and hierarchical grouping as geological interpretation . Cross beds are stratifications within a stratum, visible as thin whitelines in Fig. 2b, which are the preserved lee-faces of dunes or ripples,which migrated by wind or water action. The azimuth of the maxi-mum dip direction of these cross beds indicates the original transportdirection of the respective deposition medium [21]. Therefore, theinterpretation of cross beds is invaluable to determine the direction ofwind or water flowing in the geological past. In the field, geologistsuse a compass-clinometer. In PRo3D, they trace a cross bed by picking3D points on the DOM surface. Then a plane is fitted to these pointsvia total-least square regression [15]. The result is a so-called dip-and- trike measurement, where the dip is the direction of maximum negativeinclination of this plane, while the strike is orthogonal to the dippingdirection, as illustrated in Fig. 2c. In the context of correlation panels,geologists are primarily interested in the geographic direction of dips.Geographical directions are quantified as azimuth in degrees, where 0°,90°, 180°, and 270°point to north, east, south, and west respectively.The sizes of grains found in a stratum are a decisive factor, todetermine its rock type and its mode of deposition, that is either bywind (aeolian) or by water (fluvial). Grain sizes range from coarsesoil, such as cobble with 63-200 mm, to fine soil, such as clay with agrain size smaller than 0.002 mm. While in the field, geologists havemany tools at their disposal, measuring grain sizes in DOMs is limitedto grains visible on the textured mesh. For the exploration of Mars,scientists mostly have to rely on image-derived data, so they often needto infer grain sizes, rather than being able to measure them in 3D. A 3D interpretation that characterizes an outcrop may contain a plethoraof measurement values and annotations. For the sake of clarity, we willfocus on the ones that are essential for creating a geological log , as de-picted in Fig. 3. A log characterizes the sequence of strata as they weredeposited over time, starting with the oldest at the bottom and endingwith the most recently deposited at the top. Perfectly horizontal strataare rare; geological effects and the roles of deposition and erosion inlandscape evolution can produce irregular contacts. Therefore, the ge-ologists ‘draw’ a log over the interpretation, connecting the contacts intheir vertical, i.e. chronological sequence. The difference in elevationbetween two contacts determines the thickness of a stratum. Currentlyno 3D interpretation tool does support the direct semantic connectionof contacts to create a geological log. Instead, elevation values andnames of contacts are exported, and then geologists manually draw thecorresponding log, while cross-checking with distance measurementsin the 3D visualization. The notion of true thickness complicates thismatter significantly, which we discuss in Sect. 3.3. A geological logcharacterizes an outcrop by showing the type and succession of stratain an abstracted form. As illustrated by Fig. 3a, the y-axis of the logencodes the elevation of the contacts measured in the 3D view as wellas the thickness of the strata enclosed by the respective contacts. Thex-axis in the log (Fig. 3e) corresponds to the grain size on a logarithmicscale. Grain size relates to the rock type of a stratum, which is alsoencoded in the stratum’s color. To characterize the orientation of a stra-tum, geologists visualize the distribution of cross bed dipping-azimuthsin a rose diagram. In the example in Fig. 3b, the upper unit betweenthe yellow and the red contact ‘dips towards east/north/east’ basedon ten dip-and-strike measurements. Dashed lines are often used toconvey uncertainty, for instance, concerning a contact (horizontal line)or concerning the grain-size (vertical line) shown in Fig. 3d. Geologistsuse additional encodings such as curved lines for contacts of varyingelevation or glyphs to convey grain distributions (Fig. 3c).
For the sake of simplicity, we accepted that the difference in geographicelevation between two contacts results in the thickness of the enclosed
True Thickness Outcrop Surface
GeographicElevation D i p p i n g S t r a t u m ApparentThickness
Fig. 4. The apparent thickness of a stratum observed at an outcrop isoften misleading. The rock may be broken off at a slanted angle or thestratum itself may be tilted. A dip-and-strike measurement is essential todetermine the stratum orientation and compute its true thickness. Table 1. Summary of the most important geological terms. outcrop exposed rock face showingsedimentary structuresstratum / a layer of rock bounded by two contactscontact discrete transition between two stratadip inclination angle of a stratumdip-and-strike measurement to determine the orientationof a stratumtrue thickness thickness of a stratum with respect to its dipcross bed cross stratifications within a stratumunit, bedset specific strata, where a bedset is asub-stratum of a unitstratum. In nature that is often not the case, which is why geologistsdistinguish the measured or apparent thickness and the true thickness of a stratum. Our previous simplification is only valid if the measuredstrata lie in a horizontal plane and the outcrop surface is vertical. Fig. 4illustrates the discrepancies between geographic elevation, apparentthickness, and true thickness. In reality, the deposition of materialoriginally occurs horizontally, but strata may be tilted or even foldedover by a variety of geological or geomorphological phenomena. Hence,instead of using global elevation values over all strata and logs, eachapparent thickness needs to be corrected by the stratum’s dipping angle.However, in stratigraphy it is an accepted simplification to use oneangle per log, which still results in each log creating its own coordinatesystem. Throughout this paper, when we speak of thickness, we meantrue thickness, since the apparent thickness is of little relevance forgeological interpretation.
After having created multiple logs, geologists arrange them in a corre-lation panel, as shown in Fig. 3 on the right. The juxtaposition of logsallows geologists to identify similar strata across outcrops and to corre-late them, hence the name correlation panel. The notion of geologicalcorrelation is not related to the mathematical concept. Correlations arevisualized as colored connections between contacts. A dashed pattern isused to express uncertainty if two contacts belong to the same structure.The arrangement of logs from left to right is often determined by theirsuccession along the course of a geomorphological feature, such as acanyon or a crater rim. Rulers between the individual logs encode thedistances between them (Fig. 3g). The distances also convey a degreeof uncertainty and data quality, since inferring correlations betweenoutcrops across large distances is less reliable. Due to geological faultsor other geomorphological processes, different outcrops do not neces-sarily expose the same stratum at the same elevation or with the samethickness. Consequently, correlation lines are rarely horizontal, whichis why geologists often choose a distinct contact as a leveling horizonfor all outcrops. In the example shown in Fig. 3, a stratum rich of fossilshas been found in Sol1 and Sol4, indicated by dark gray rectangles atan elevation of 1m. This stratum is missing from the Sol2 and Sol3 logs,because it is not exposed at these locations and potentially buried. Still,geologists are able to infer its existence and its approximate elevation.
Image data was collected at the Hanksville-Burpee Dinosaur Quarry(HBDQ), near Hanksville, Utah, U.S.A. (110 ◦ ◦ units enclosed by unit contacts in red and bedsets enclosed by bed contactsin orange, as illustrated in Fig. 2. They arranged the individual logs ina manually illustrated panel and preliminary, derived correlations. The4 o appear in IEEE Transactions on Visualization and Computer Graphics c e da b Fig. 5. (a) Data model: tree of strata and their bounding contacts. (b)InCorrPanel: (c) secondary log shows a tree cut at depth 1, while theprimary log shows the leaf cut, (d) rose diagram aggregates cross bedorientations, (e) dashed border conveys uncertain rock type. resulting correlation panel is visible in Fig. 3. We use the DOMs, theinterpretations, the logs, and the correlation panel as a running examplethroughout this paper and as a use case in Sect. 6.
ESIGN P ROCESS AND D OMAIN A BSTRACTION
The results of InCorr are based on a decade-long collaboration withplanetary geologists, mainly in the context of research and develop-ment of PRo3D as an interpretation tool. In a workshop following theevaluation campaign described in Sect. 3.5, our collaborators statedthe need to semi-automatically generate correlation panels from in-terpretations and suggested a simplified visual encoding for logs, asshown in Fig. 3. We followed a participatory design approach [14]leading to a three-phase evolution of InCorr: In phase (1) , we wereconcerned with how to transform annotations into a log, researchinga hierarchical data structure and the transformations necessary. Thisresulted in an non-interactive log prototype matching the simplifiedvisual encoding. In phase (2) , we focused on understanding the domainbackground and tasks involved with correlation analysis and created aninteractive prototype. It is integrated with PRo3D featuring multiplelogs and correlations. In phase (3) , the necessity of a logging toolmeasuring true thickness became evident. Key aspects of this phasewere interaction and visualization design, and end-to-end workflowintegration. Each phase was accompanied by a week-long research stayand roughly quarterly meetings. In phase (3) we shortened intervalsand iterated on visualization and interaction prototypes sometimes on adaily basis, shaping InCorr through the continuous feedback providedby our collaborators.
With InCorr we address a set of tasks that bridges the gap betweenoutcrop interpretation and the creation of a geological model based onlogs and correlations:• T1 Create a geological log for each outcrop based on the annota-tions and measurements taken.• T2 Create correlations from geological logs as the basis for aregional geological model• T3 Edit and export the correlation panel to be manipulated inother tools for further analysis or disseminationWe abstract these tasks and their subtasks by following the multi-level task typology by Brehmer and Munzner [2]. We subdivide the creation of a geological log ( T1 ) into connect contacts ( T1a ), add rocktype (
T1b ), and add cross beds (
T1c ). In
T1a the geologists identify the relevant contacts they want to connect and then select them toform a log ( annotate ). For each stratum in the log they identify the grain size and select the respective rock type (
T1b ). Further, they identify and select cross bed measurements belonging to a stratumto summarize the orientation distribution within it (
T1c ). For T2 thegeologists arrange the logs from T1 to compare their characteristicsand find similarities ( T2a ). When found, they select one contact perlog they want to connect and create ( annotate ) a correlation (
T2b ).When the correlation analysis is completed, geologists ‘tidy up’ thecorrelation panel by vertically and horizontally arranging the logsand export it to be used in other tools of their workflow ( T3 ). To bridge the gap between outcrop interpretation and 3D logging, weneed to infer the hierarchical structure of strata from a set of contacts.Logging allows geologists to pick a 3D position on each contact, whilethe true height of these positions, i.e. elevation corrected by a dippingangle, determines the vertical sequence of the contacts. Each contacthas a rank assigned by the geologist, representing the magnitude ofchange between two rock layers. With this information, we can derivea tree of strata. Each stratum is bounded by an upper and lower contactdefining its minimum and maximum height, and thus its thickness.We achieve this by starting out with a fictive stratum with a heightinterval from negative to positive infinity. Our algorithm goes throughthe set of contacts sorted by their rank, finds the stratum with a heightrange matching the contact’s height and splits it into two sub-strata.Performing this step for each contact yields a tree of strata in a breadth-first fashion.Each stratum can be assigned a rock type and may contain geologicalfeatures, such as cross beds. To characterize the distribution of crossbed orientations within a stratum, geologists perform dip-and-strikemeasurements. Such measurements can then be assigned to leaf strata.This allows us to use simple tree traversals to aggregate measurementsfor arbitrary strata. As an example, we represent the log illustrationshown in Fig. 3 using our data model in Fig. 5a. The whole outcrop,represented as root stratum, contains units (sub-strata) defined by thered contacts. The upper two units contain sets defined by orangecontacts, which do not have descendant strata, but contain cross beds(white lines). Dipping-azimuth values of the cross beds are aggregatedat a unit level and visualized as rose diagrams. We define a correlationas a set of contacts connecting them between logs. Correlations areoften uncertain in parts, which means that the connection between twocontacts is either certain or uncertain, typically represented as solid ordashed line, respectively. N C ORR
InCorr consists of three components that we integrated with PRo3D:(1) the
InCorrPanel , a 2D interactive correlation panel, that offersgeologists an evolving summary of their geological analysis and that iseasy to keep in sync with annotations and measurements, (2) a loggingtool , that allows them to intuitively connect contacts in the OutcropView provided by PRo3D, and (3) a list view for assigning rock typesto strata in the InCorrPanel. In this section, we discuss the visualencodings in Sect. 5.1 and the interaction design in Sect. 5.2 that weuse for these components.
Geological logs are a visual abstraction of complex spatial relationshipsof 3D phenomena. Showing them in a correlation panel does providegeologists with a concise overview of large-scale geological analyses.Thus, static correlation panels already facilitate T2 by enabling a visualcomparison of outcrops without inspecting their spatial representa-tions. To exploit these properties, the design of the InCorrPanel needsto closely resemble manually illustrated logs and correlation panels,without overwhelming users with customization options ( G3 ).There is no universally agreed upon standard for creating geologicallogs. Every geologist has their own specific style. For instance, the5
1a Connect Contacts T1b Assign Rocktypes T1c Assign Orientations ab c d e f g h
Fig. 6. To create a 3D log, users (a) first pick a reference plane, then they (b) connect the contacts to form (c) a 3D log. Users then assign rock typesto the (d) empty log via the (f) rock types list, resulting in (e) a log encoding rock categories and grainsizes. To finish the log, they (g) assign crossbed measurements, which are aggregated and encoded as (h) rose diagrams. geological logs in our running example (shown in Fig. 3) can be con-sidered a rather sophisticated variant. Therefore, we did not conducta formal design space analysis, but let our collaborators decide on theappropriate visual representations. A key topic of design phase (2) was to derive which aspects need standardization and which aspectsneed to be left to the individual geologist. As there currently is no toolfor creating correlation panels directly from outcrop interpretations,our collaborators were eager to participate in these sessions. Hayeset al. [10], Van Lanen et al. [27], or Hampson et al. [9] present examplesof the variety of correlation panels published in geological journals. Inthe following we present our design decisions for InCorr starting withthe visual encodings for displaying a single log.
The visual representation of a single log in the InCorrPanel is illustratedin Fig. 5b. The x-axis encodes the grain size on a logarithmic scale,while the y-axis encodes the order and true thickness between contacts.To calculate the true thickness, a log requires a reference plane in 3D.For each contact we compute the height above the reference plane andmap it directly to a position on the y-axis. For each stratum, we draw afilled rectangle, ranging on the y-axis from the elevation of its lowercontact to the elevation of its upper contact. The stratum’s grain size,which relates to the associated rock type, is encoded redundantly bythe width of the rectangle and its color. Additionally, we draw thecontacts as horizontal lines at their elevation matching their thicknessand color in the Outcrop View ( G2 ). To indicate if geologists areuncertain with the rock type they assigned, we draw the right borderof the rectangle as a dashed line Fig. 5e. This is especially relevant inMartian use cases, where the tools for observation are mostly visualand the better part of the observed strata are of finer grains. We chosethe simplified rectangular log over the curved contact representationsand curved polygons for several reasons. One could derive the variationin elevation by picking multiple points per log, but (1) there is noinformation in a contact that would describe the variation of a stratumalong the x-axis; (2) it becomes difficult to read strata thicknessesdirectly from the log; and (3) the encoding is rather specific and rarelyused in manual illustrations. Ultimately, as we create the plot as anSVG image it can be directly edited in a vector image tool after export,to add custom illustration styles ( G3 , G4 ).When embedding an outcrop analysis into a bigger context, deephierarchies of strata may emerge. To manage such hierarchies, our col-laborators suggested to add a second, abstracted, log representation thatemphasizes the affiliation of sub-strata to their respective containingstrata. We denote this abstracted log as secondary log and show it to theleft of the primary log (Fig. 5c). The primary log always shows the full detail of strata (leaf nodes of the data model), whereas the secondarylog shows a single, user specified level (here level 0) of the hierarchyindicated in the data model (Fig. 5a). This two-column representationof logs resembles a specialized icicle plot [25], where all nodes exceptthe leaf nodes and the selected level are removed from the hierarchy.A similar double log representation can, for instance, be found in themanually illustrated correlation panel by Hampson et al. [9] ( G3 ).The selected depth in the secondary log also governs the granu-larity of how we aggregate measurement values. Supporting T2 , theInCorrPanel summarizes the distribution of dip-azimuths as rose dia-grams (Fig. 5d). We aggregate the orientations into 24 15 ◦ angularbins and encode the frequency into their area, as suggested by Sander-son and Peacock [23]. This allows the geologists to infer the majordipping-azimuths of a stratum without missing low frequency outliers.We further compute the mean angle (via polar coordinates) and encodeit as a red line. According to the geologists, this representation allowsthem to quickly judge if the distribution follows one general direction. In published correlation panels, the logs are carefully ordered to conveythe results in the best possible way. In the case of the HBDQ analy-sis, the logs are sorted along the geographic direction from SSW toNNE, as the result of a projection from longitude and latitude ontoa line, as illustrated in Fig. 2a. We did consider ordering the logs inthe InCorrPanel automatically along a fitted line or let the geologistspick a geographic direction. Ultimately, the log order is very contextsensitive to the location of the outcrops and domain experts will wantto order them manually. Each log is a concise summary of an outcropinterpretation, and a correlation panel supports geologists in identifyingstrata with similar characteristics. The InCorrPanel allows them tojuxtapose logs arbitrarily in an interactive fashion. This enables them toquickly compare rock types and cross bed orientations during the analy-sis in order to correlate contacts ( T2 ). This is similar to rearranging thedimensions of a parallel coordinates plot to investigate relationships.Further, we compute the distances between adjacent logs and displaythem as rulers between the log names, as seen in Fig. 7h.After determining the horizontal arrangement, we are left with thevertical positioning of each log in the InCorrPanel. In discussion withour collaborators, we realized that there is no single layout method forvertically positioning logs. Further, the most suitable layout is likelyto change throughout the analysis. We agreed on a two-fold approach,which provides standardization as well as flexibility ( G1 , G3 ). Westarted out rendering all logs, their contacts, and strata in a commoncoordinate system based on the geographic elevation of the individualelements. Since this approach does not accommodate for true thickness6 o appear in IEEE Transactions on Visualization and Computer Graphics we use the geographic elevation of a log’s reference plane as an anchorpoint as discussed in Sect. 5.1.1. When users find correlating stratathey often use the resulting surface as baseline for vertical alignment.In the best case, every log, i.e. every outcrop, exhibits a part of thesame rock stratum. Using this stratum as the baseline offsets each logvertically in such a way that all log nodes of this surface are at the samey-position in the log. In that case, the correlation lines associated withthe upper contact of the stratum become horizontal. In the example inFig. 3h, only two outcrops exhibit the baseline stratum. The dashedyellow lines in the correlation panel indicate that this surface is alsopresent at the locations of the other two outcrops, but not exposed. In this section we discuss the interactions necessary to achieve thedomain tasks in the sequence described in Sect. 4.1, i.e. creatinglogs ( T1 ), creating correlations ( T2 ), and editing the resulting panelfor export to other tools ( T3 ). Fig. 6 provides an overview of thesuccession of interactions for creating a log ( T1 ), starting with thecreation of a single log based on 3D contact lines from an outcropinterpretation ( T1a ). T1a Connect Contacts to Create Strata:
Before users can startconnecting contacts to form a log, they need to specify a reference planeto enable true thickness computation. They choose a contact whichthey suspect to have a suitable orientation to which we fit a plane using3D total-least squares regression. The fitted plane is then visualizedas a colored disc, where the color encodes the dipping angle of theplane (Fig. 6a). Geologists can try out different contacts and inspectthe respective planes or confirm the found plane. Then users pick apoint on each contact they want to add to their log. Due to the domainconstraints we discussed in Sect. 4.2, the selection of contacts neednot to happen in order. For each picked point we compute its heightabove the reference plane and sort the set of points by this criterion.This allows us to draw 3D line segments as a pairwise combinationof the elements of the sorted list (Fig. 6b). Further, a point picked fora contact can be modified or removed and the log polyline changesaccordingly, giving immediate visual feedback to the user. Users canedit existing logs in the same way to keep their analysis up-to-date withnew contacts ( G1 ). T1b Assign Rock Types to Strata:
The strata of a newly createdlog do not have rock types associated with them yet, indicated by theirwhite color (Fig. 6c). To assign a rock type to a stratum, users firstselect the respective stratum in the primary log and then they pick arock type from the rock types list (Fig. 6f). The color and the widthof the respective rectangle changes accordingly. To indicate how suregeologists are with their choice of rock type they can add an uncertaintystate shown as a dashed or solid right border encoding uncertainty orcertainty, respectively. A log fully annotated with rock types, includinguncertainty states, is shown in Fig. 6e ( G3 ). When selecting a stratumin the InCorrPanel, we also highlight the bordering contacts in theOutcrop View (Fig. 6g), allowing geologists to identify the respectivearea of the outcrop (G5). T1c Assign Cross Beds to Strata:
Analogously to the previoustask, users first select a stratum in the primary log in the InCorrPaneland the bounding contacts are highlighted in the Outcrop View. Thissupports the geologists in identifying the region where they want toselect cross bed measurements ( G5 ). After confirmation, the selectedcross beds are assigned to the selected stratum and a rose diagramappears next to the log. From this point on, selecting the stratum alsoselects the cross beds assigned to it. Changing this set by adding orremoving cross beds is also reflected in the rose diagram ( G1 ). Wediscussed approaches for the automatic selection of cross beds withour collaborators, but in the end resorted to letting them select themeasurements manually. After having created at least two logs, i.e.performing T1a twice, users can create correlations ( T2 ). T2a Find Similar Strata
Finding correlations, means that geolo-gists need to identify contacts that belong together, i.e. two contactsdelineate the transition between the same strata in different logs. Toachieve this, geologists need to be able to effectively compare outcropcharacteristics encoded into a log, comprising the rock type of a stratum, its thickness, and the orientation distribution of the cross beds within it.To support this task, the InCorrPanel allows users to arbitrarily changethe horizontal order of logs and compare any two logs via juxtaposition.
T2b Connect Contacts to Create Correlations
When suitable con-tacts are identified, the geologists select the respective lines, one per log.On confirmation, the selected contacts are then connected by curvedlines in the color of the selected contacts. Each connection betweentwo correlated logs is initially rendered in a dashed pattern to conveyuncertainty and can be switched to a solid line style. If the correlationanalysis is complete and the InCorrPanel contains all the findings thegeologists want to present for scientific dissemination, for instance asan essential part of a geological publication, they move on to T3 . T3 Level to Horizon and Export InCorrPanel:
To make logsmore comparable in a published correlation panel, geologists typicallylevel all logs to a common horizon ( G3 ). This means to verticallyalign all logs to a common baseline correlation. When drawn in thesame coordinate system, correlating contacts rarely occur on the sameheight, as exhibited by the red correlations in Fig. 3. Ideally thereis a distinctive rock layer present in every log, which results in acorrelation across the whole panel. Then this correlation can act as avertical baseline. To achieve this, geologists select a correlation in theInCorrPanel and confirm to vertically align all logs to it. This results instraight connections instead of curved ones, as illustrated by Fig. 7e.For final adaptions and editing, the content of the InCorrPanel can beexported as a scalable vector graphic readable by all common vectordrawing applications ( G4 ). InCorr is implemented in F
VALUATION
In this section we discuss the methods we used to evaluate InCorr.We verified InCorr by using it to recreate the correlation analysis ofthe HBDQ campaign. Then we validated the resulting InCorrPanel(Fig. 7a) against the manually created illustrations shown in Fig. 7b–d.Finally, we collected feedback from three geologists after performingtasks in the form of a design validation.
We used InCorr to create logs, one for each outcrop of the HBDQdataset named Sol2–5, as shown in Fig. 8. We used the logging toolto connect contacts from the interpretation data, which amountedto true thickness measurements. We assigned the rock types to thesame number of strata and selected strata to which we assigned atotal of cross bed measurements. We then added correlations,which resulted in the correlation panel presented in Fig. 7a. The wholeprocess took approximately hours. Carrying out this process in anon-assisted manner, using field data, involves manual measurement ofthe distances between contacts, conversion of apparent to true thickness,and manual plotting of rose diagrams. According to our collaborators,for the amount of data presented, this would take considerably longerthan 1.5 hours, and though the time it would take would vary betweenworkers, producing a correlation of equal precision would be on theorder of a day to several days of work.Our log representation matches the visual properties of the ‘sim-plified log illustrated manually’ (Fig. 7d) and so meets the minimalencoding capabilities, as proposed by our collaborators (Sect. 5.1.1).We extended this version by two data-driven features, the secondarylog and the rose diagrams. Our solution does not include the followingstyles present in the ‘log illustrated manually’ (Fig. 7c): (1) curved,converging, or tilted contacts, (2) gradients, pattern fills, and rounded7 Illustrated Manually Created with InCorr b c d fea gg ghh
Fig. 7. (a) Correlation panel created with InCorr based on outcrop interpretation data with (b,c,d) manual illustrations for comparison. (e) all logsare aligned to a common baseline correlation at the top. (f) no vertical exaggeration of strata, (g) no indication of buried regions, and no visibleconnection of non-adjacent logs. (h) showing spatial distance between logs. corners for strata, nor (3) glyphs to indicate additional rock properties.Besides depending on the individual geologist’s taste, most of thesefeatures do not directly encode information that is readily quantifiablein the data. We left these encodings to the manual workflow followingthe export of the InCorrPanel. Nevertheless, it is interesting to incorpo-rate styles that could be inferred from the data, such as curved or tiltedcontacts. We could not recreate the thicknesses of one stratum in Sol 2and one in Sol 5 (Fig. 7f). The reason is either vertically exaggerationto make them visible or the true thickness is inferred from informationnot present in the data. When we compare the ‘panel illustrated man-ually’ (Fig. 7b) with the InCorrPanel as a whole, the main differenceis that our illustration of the HBDQ campaign is flattened to the topcorrelation instead of the one connecting the two fossil layers. Thisis because in the ‘panel illustrated manually’ the location of the fossillayer is assumed for Sol 3 and Sol 4 although it is actually buried inthe physical outcrop and therefore the logs contain an empty region(Fig. 7g). We did not include this in InCorr, because it would requireadditional interactions and design iterations to make this data-drivenand geologists can rarely make such assumptions on Mars. We didnot implement the visual connection of contacts with a line betweennon-adjacent logs. This would require a layout algorithm that avoidscollisions with logs in between. At the same time, the resulting curvedline should encode the course of the correlation, which can only bespecified by the geologist. But, a correlation can be created betweenarbitrary logs, and the visual connection will show when the respectivelogs are adjacent.
We recruited three geologists to participate in our design validationand we will refer to them as P1, P2, and P3. P1 is a sedimentologistinvolved in design phases (1) and (2) , P2 is an engineering geologistfrom the field of tunnel constructions, and P3 is a planetary scientistwith a geological background, none of which are co-authors of thepaper. We conducted one to one interviews limited to one hour, startingwith background questions to their field and their expertise in DOMinterpretation and logging. Then we asked them to perform the tasks ofcreating logs T1 and creating correlations T2 on two outcrops of theHBDQ dataset, closing with a short questionnaire and open feedback.Prior to the interview the participants received a five minute tutorialvideo, briefly explaining the interactions for achieving T1 and T2 . Wedecided to collect qualitative feedback on the relevance of the problem,the quality and ease of our solution, and its potential impact. All participants agreed, that logging and the correlation analysisof DOMs is a time-consuming process: ‘Often it is faster to drawthe panel by hand on scales paper (opposed to using vector drawingsoftware)’ (P3) , and that InCorr can provide a significant speed up tologging and assigning cross beds: ‘Looking at this software I should beable to do it very quickly. About an hour instead of a week.’ (P1) .They were very positive about the interactions in the 3D OutcropView while creating a log and easily changing the succession of points: ‘Usually it is hard to get the scales correct’ (P3) , and ‘That works prettywell!’ (P1) . The linked highlighting was also a very welcome featurefor relating logs and their spatial counterparts, but also when assigningcross beds: ‘The highlighting of a stratum’s contacts is very intuitive.It feels visual and haptic.’, ‘This is perfect for correlating spatially overlarge distances.’ (P2). Each participant was missing different stylized features in the logsand correlations representation. Asked if they could encounter theInCorrPanel ‘as is’ in a geological publication they generally agreed,but they themselves would not use it directly without additional editing.Although P1 was content with post production ‘Really good that itis already in SVG, that’s important. Glyphs and styles don’t need tohappen in InCorr, I would export it to AI or Corel anyway’ . Still, hewished for more options on encoding different properties along thecorrelation lines. When two correlation lines end in one, it is interestingto show where the phase out happens. P2 wished for hatching patternsto fill strata and to be able to change the roundness of the rectanglesto encode weathering effects. He also pointed out that he is unfamiliarwith the quantification of the grain size in our rock types list. With ourcollaborators we agreed on displaying grain size categories as valuesof the Krumbein φ scale [30], but for broader acceptance we shouldalso show the metric values and support other categorizations. P3immediately addressed the absence of glyphs and curved contacts.When asked about the potential and implications of the tools pro-vided through InCorr, each participant expressed their excitement, butfor different reasons. P1 plans to use the logging tool personally fora publication requiring extensive true thickness analysis. P2 is eagerto try out logging and correlation analysis in lithological outcrops, i.e.tunnel faces, created during tunnel excavation. P3 focused on the im-plications of a data-driven approach to correlation analysis, since inher current research she needs to reproduce and compare results fromdata that is only present in figures. On the suggestion of bundling theanalysis and input data with a publication to be viewed with InCorr and8 o appear in IEEE Transactions on Visualization and Computer Graphics Sol 2 Sol 3Sol 5Sol 4
Fig. 8. Outcrops Sol2–5 of the HBDQ dataset, each with a 3D log createdwith InCorr and the resulting 2D log from the InCorrPanel.
PRo3D, she answered ‘Isn’t this the whole point of a publication? Tohave other people interact with your data. I hope that is the future ofpublishing’ . We will elaborate on these three topics in Sect. 7.
ISCUSSION AND F UTURE R ESEARCH
In this section we summarize the key aspects of InCorr, relate them tothe design goals we defined in Sect. 1.2, and present a critical appraisalof our solution supported by the results from Sect. 6, referring toparticipants of the design validation where necessary.We developed a 3D logging tool that allows geologists to creategeological logs with little effort, especially when compared to currentapproaches ( G1 ). All participants could create a 3D log in under aminute, while spending most of the time on considering which contactsto include. We had some usability issues when selecting contact linesor very thin strata in the InCorrPanel. To improve the selection processof thin lines, a list of potential candidates could be provided, combinedwith additional visual feedback, like a selection preview while hoveringthe lines or thin strata. Assigning cross beds to strata would greatlybenefit from better selection tools provided by PRo3D, as for instance aline brush. Further, with the knowledge gained throughout this work wecould try again to offer our collaborators an automatic pre-assignmentof cross beds within a stratum that they can then intuitively edit.Our data-driven approach of logging directly on the 3D contact linesties logs and correlations to the outcrop interpretation data ( G2 ). Allparticipants recognized this as a significant improvement over currentmethods, where a log is rather a collection of height measurementsthan a 3D polyline. It would be interesting to investigate how to enrichthe exported vector graphics file with meta data indicating its origin.According to the design feedback, the InCorrPanel contains a sufficientset of visual encodings and customization to pass as a correlation panelin a geological publication ( G3 ). In Sect. 6 we could demonstrate,that InCorr encompasses all features to conduct a correlation analysis.Nevertheless, G3 offers the largest room for improvement. InCorrPanelestablishes a baseline while some styling features are clearly left tovector drawing tools. It would be beneficial to explore the boundaryof this separation, adding features that are supporting the analysis orshould be reflected in the data.The integration with PRo3D and the export of the InCorrPanel toa vector format allows InCorr to successfully bridge the gap betweenoutcrop interpretation and creating regional geological models based oncorrelations ( G4 ). We also verified this by reconstructing the correlationanalysis of the HBDQ from existing outcrop interpretation data. On theother end, Fig. 5 shows a modified log representation exported fromthe InCorrPanel. The highlighting of the selected log and of a stratum’sbordering contacts allows users to establish a context between the 3Doutcrop interpretation and the 2D log representation ( G5 ). Geologistsare very aware of the spatial context between the outcrops and logs oftheir analysis, due to the fact that they may spend weeks on interpreting the outcrops. None of the participants was familiar with the HBDQ data,so the contact highlights helped them to quickly orient themselves in theinterpretation. We noticed that P3 subsequently tried to select strata inthe Outcrop View, i.e. the region on the surface between two outcrops.Geometry for strata, which one could infer from contacts and the log,would benefit InCorr on multiple levels. It would be the basis for theassisted assignment of cross beds as mentioned before. One coulddirectly assign rock types in the Outcrop View without switching to thepanel, and further it could tie the actual DOM geometry and texture tothe logging. P1 also entertained the idea of juxtaposing outcrops in 3D,aligning them along a geographic direction, analogous to the correlationpanel. Such a transformation would dissolve longitude and latitudecomponents of the data, while preserving elevation and orientation.InCorr was designed to support geologists in the geological analysisof Mars by translating static correlation panels to the interactive world.We could identify several other application areas for InCorr. As sug-gested by P2, InCorr can be used to log the succession of strata in solidrock, exposed in the form of tunnel face outcrops. These are createdand captured at regular intervals during tunnel excavation. In large-scale tunnel construction projects it is common practice to maintaina reconnaissance tunnel, smaller and a few hundred meters ahead ofthe main tunnel. Logging and correlation of the reconnaissance tunnelallows geologists to reconstruct the geological situation in the moun-tain and predict when the main tunnel will hit critical rock sections.Outcrop-based logging and correlation panels are not used in this fieldof geology and we will pursue this idea in an ongoing research project.In the context of sedimentary geology and exploration of Mars, we canadapt the InCorrPanel to display other measurement aggregations ofstrata than dip azimuths. Our data model and interactions support toadd, for instance, bedding thicknesses or grain sizes to strata, and showtheir distribution in histograms. Finally, stratigraphic succession playsan essential role in archeological excavations and InCorr could be avaluable addition to the work of Traxler et al. [26].Although correlation panels are a very domain-specific representa-tion, we can generalize them as follows: a log connects discrete featuresto form intervals. The order is implicitly given by a metric and theintervals are nested through the rank of their features. Users connecttwo features in different logs through a correlation depending on thesimilarity of the two intervals the feature is part of. As an example,one could use the InCorrPanel to compare the lives of different people.Each person’s life consists of events of different magnitude of change.These events are implicitly connected and sorted by their time of occur-rence and form a log. Then one could correlate the life-changing eventsacross multiple persons and investigate surrounding events or inspectsimilar phases of life and compare the events they begin and end with.From a geological perspective, it is obvious that correlation panels canbe used to represent time, because they already do. The elevation ofcontacts and strata marks their occurrence in time, while the rank ofa contact translates to the frequency of occurrence. For instance, thered contacts shown in Fig. 2 represent a change that may happen onceevery 100 to 1000 years, while the changes shown as orange contactscan happen once in a week or a month. ONCLUSION
With InCorr we present a design study of translating the static ge-ological illustration of correlation panels for outcrop analysis to aninteractive solution. It comprises a 3D logging tool and an interactivedata-driven correlation panel, the InCorrPanel. We integrated bothwith an existing outcrop interpretation software and provided a vectorgraphics export to integrate with the established workflow of geologists.We demonstrate InCorr’s functionality through a use case and validatedour results against a manually created illustration based on identicaloutcrop interpretations. Further, we collected qualitative feedback fromthree geologists. The evaluation indicates that InCorr is a powerful anduseful tool and can significantly improve geological analysis workflowsthrough data propagation, and reducing time and effort of illustratinglogs and correlation panels manually.9
EFERENCES [1] R. Barnes, S. Gupta, C. Traxler, T. Ortner, A. Bauer, G. Hesina, G. Paar,B. Huber, K. Juhart, L. Fritz, et al. Geological analysis of martian rover-derived digital outcrop models using the 3-d visualization tool, planetaryrobotics 3-d viewer pro3d.
Earth and Space Science , 5(7):285–307, 2018.[2] M. Brehmer and T. Munzner. A multi-level typology of abstract visualiza-tion tasks.
IEEE Transactions on Visualization and Computer Graphics ,19(12):2376–2385, Dec. 2013. doi: 10.1109/tvcg.2013.124[3] S. J. Buckley, K. Ringdal, N. Naumann, B. Dolva, T. H. Kurz, J. A. Howell,and T. J. Dewez. Lime: Software for 3-d visualization, interpretation, andcommunication of virtual geoscience models.
Geosphere , 15(1):222–235,2019.[4] CloudCompare. 3D point cloud and mesh processing software . https://cloudcompare.org . Accessed: 2020-04-28.[5] ESA. Searching for signs of life on mars. http://exploration.esa.int/mars/43608-life-on-mars/ . Accessed: 2020-04-28.[6] Esri Geospatial Cloud. ArcGIS - the mapping and analyticsplatform. . Accessed: 2020-04-28.[7] Y. Gao.
Contemporary Planetary Robotics: An Approach Toward Au-tonomous Systems . John Wiley & Sons, 2016.[8] J. Grotzinger, S. Gupta, M. Malin, D. Rubin, J. Schieber, K. Siebach,D. Sumner, K. Stack, A. Vasavada, R. Arvidson, et al. Deposition, ex-humation, and paleoclimate of an ancient lake deposit, gale crater, mars.
Science , 350(6257):aac7575, 2015.[9] G. J. Hampson, M. R. Gani, K. E. Sharman, N. Irfan, and B. Bracken.Along-strike and down-dip variations in shallow-marine sequence strati-graphic architecture: Upper cretaceous star point sandstone, wasatchplateau, central utah, u.s.a.
Journal of Sedimentary Research , 81(3):159–184, Mar. 2011. doi: 10.2110/jsr.2011.15[10] A. G. Hayes, J. P. Grotzinger, L. A. Edgar, S. W. Squyres, W. A. Watters,and J. Sohl-Dickstein. Reconstruction of eolian bed forms and pale-ocurrents from cross-bedded strata at victoria crater, meridiani planum,mars.
Journal of Geophysical Research , 116, Apr. 2011. doi: 10.1029/2010je003688[11] D. Hodgetts, R. Gawthorpe, P. Wilson, and F. Rarity. Integrating digital andtraditional field techniques using virtual reality geological studio (vrgs).In , pp. cp–27. European Association of Geoscientists & Engineers,2007.[12] T. H¨ollt, J. Beyer, F. Gschwantner, P. Muigg, H. Doleisch, G. Heinemann,and M. Hadwiger. Interactive seismic interpretation with piecewise globalenergy minimization. In , pp.59–66. IEEE, 2011.[13] T. H¨ollt, W. Freiler, F. Gschwantner, H. Doleisch, G. Heinemann, andM. Hadwiger. SeiVis: An interactive visual subsurface modeling ap-plication.
IEEE Transactions on Visualization and Computer Graphics ,18(12):2226–2235, 2012.[14] S. J¨anicke, P. Kaur, P. Kuzmicki, and J. Schmidt. Participatory visual-ization design as an approach to minimize the gap between research andapplication. In n.n., ed.,
Proceedings of the Workshop on the Gap betweenVisualization Research and Visualization Software (VisGap) , 2020. doi: 10.2312/visgap.20201108[15] R. R. Jones, M. A. Pearce, C. Jacquemyn, and F. E. Watson. Robustbest-fit planes from geospatial data.
Geosphere , 12(1):196–202, 2016.[16] E. M. Lidal, M. Natali, D. Patel, H. Hauser, and I. Viola. Geologicalstorytelling.
Computers & graphics , 37(5):445–459, 2013.[17] NASA. Mars 2020 mission overview. https://mars.nasa.gov/mars2020/mission/overview/ . Accessed: 2020-04-28.[18] M. Natali, E. M. Lidal, J. Parulek, I. Viola, and D. Patel. Modeling terrainsand subsurface geology. In
Eurographics (STARs) , pp. 155–173, 2013.[19] P. R. Nesbit, P. R. Durkin, C. H. Hugenholtz, S. M. Hubbard, andM. Kucharczyk. 3-d stratigraphic mapping using a digital outcrop modelderived from uav images and structure-from-motion photogrammetry.
Geosphere , 14(6):2469–2486, 2018.[20] D. Patel, C. Giertsen, J. Thurmond, J. Gjelberg, and E. Gr¨oller. Theseismic analyzer: Interpreting and illustrating 2d seismic data.
IEEETransactions on Visualization and Computer Graphics , 14(6):1571–1578,2008.[21] D. R. Prothero and F. Schwab.
An introduction to sedimentary rocks andstratigraphy . WH Freeman and Company, 1996.[22] H. Sahoo and N. D. Gani. Creating three-dimensional channel bodies in lidar-integrated outcrop characterization: A new approach for improvedstratigraphic analysis.
Geosphere , 11(3):777–785, 2015.[23] D. J. Sanderson and D. C. Peacock. Making rose diagrams fit-for-purpose.
Earth-Science Reviews , p. 103055, 2019.[24] Schlumberger Information Solutions. Petrel seismic to simula-tion software. . Accessed: 2020-04-28.[25] H.-J. Schulz. Treevis. net: A tree visualization reference.
IEEE ComputerGraphics and Applications , 31(6):11–15, 2011.[26] C. Traxler and W. Neubauer. The harris matrix composer - a new toolto manage archaeological stratigraphy. In
Arch¨aologie und Computer-Kulturelles Erbe und Neue Technologien-Workshop , vol. 13, pp. 3–5,2008.[27] X. M. van Lanen, D. Hodgetts, J. Redfern, and I. Fabuel-Perez. Ap-plications of digital outcrop models: two fluvial case studies from thetriassic wolfville fm., canada and oukaimeden sandstone fm., morocco.
Geological Journal , 44(6):742–760, 2009.[28] VRVis. Aardvark. Accessed: 2020-04-29.[29] VRVis. Pro3d. http://pro3d.space . Accessed: 2020-04-28.[30] A. J. C. Wilson. Stratigraphy and sedimentation by w. c. krumbein and l.l. sloss.
Acta Crystallographica , 17(8):1090–1090, Aug. 1964. doi: 10.1107/s0365110x64004170[31] S. Wlaschin.
Domain Modeling Made Functional: Tackle Software Com-plexity with Domain-Driven Design and F . Pragmatic Bookshelf, 2018.. Pragmatic Bookshelf, 2018.