Semantic Approach for Prospectivity Analysis of Mineral Deposits

Sławomir Wójcik

, Taha Osman

and Peter Zawada

International Geoscience Services (IGS) Ltd., BGS, Keyworth, Nottingham NG12 5GG, U.K.

School of Science and Technology, Nottingham Trent University, Nottingham NG11 8NS, U.K.

Keywords: Geodata, Semantic Web, Geospatial Reasoning, Mineral Prospectivity Analysis.

Abstract: Early mineral exploration activities motivates innovative research into cost-effective methods for

automating the process of mineral deposits’ prospectivity analysis. At the heart that process is the

development of a knowledge base that is not only capable of consuming geodata originating from multiple

sources with different representation format and data veracity, but also provides for the reasoning

capabilities required by the prospectivity analysis. In this paper, we present an integrative semantic-driven

approach that reconciles the representation format of sourced geodata using a unifying metadata model, and

encodes the prospectivity analysis of geological knowledge both at the schemata modelling level and

through more explicit reasoning rules operating on the semantically tagged geodata. The paper provides

valuable insights into the challenges of representation, inference, and query of geospatially-tagged

geological data and analyses our initial results into the prospectivity analysis of mineral deposits.

1 INTRODUCTION

World-wide expenditure on non-ferrous mineral

exploration (gold, copper, nickel and zinc) has

varied from 14 – 20 billion USD annually for 2011 –

2012 (SNL 2015). This significant level of

expenditure is part of a mining related value chain

that can have an important impact on national and

regional jurisdictions for creating wealth and

alleviating poverty. A vital component of early stage

exploration activities is the availability of multi-

source geodata comprising geology, geophysics,

geochemistry and remote sensing from which initial

prospectivity maps are assembled. Prospectivity

maps at this initial stage represent broad and

generalised conceptualizations of the geological

conditions that may indicate areas or commodities of

interest for more detailed follow-up exploration. The

availability of this geodata from public and private

sources such as national Geological Surveys is a

significant factor in attracting the mineral

exploration sector (MINEX). However, the geodata

is rarely seamless, is discontinuous and is in multiple

representation formats involving traditional methods

of collating and analysing these data in a lengthy and

labour-intensive process by specialists. Recently, the

dramatic increase in processing capacity of current

computer systems and increasing availability of

geodata in digitised format promoted investigating

more cost-effective, computerised prospectivity

analysis. Mineral prospectivity maps can then be

produced by an automated, iterative process that is

designed to reconcile the discrepancy in geodata

representational formats, correlate the multi-source

data, and reason upon it using geological rules in

order to infer and visualise potentially prospective

regions. This approach would radically shorten

delivery time by reducing the time to perform the

analysis using traditional methods and ultimately

provide the MINEX sector with early stage

indications of prospectivity. In collaboration with

Nottingham Trent University, the International

Geoscience Services Ltd (IGS) (IGS 2015) would

like to contribute to that paradigm change by

developing a system that is able to store, model and

query different types of geological data and perform

automatic and human-assisted automatic analysis of

these data to produce various reports and maps of

prospectivity (likelihood of a given mineral being

deposited in a given area).

In this paper, we argue that Semantic Web

technologies are currently well placed to assist in

addressing the challenges of mineral prospectivity

analysis, and present an integrative approach that

exploits the capabilities of semantic technologies to

solve the data reconciliation problems by deploying

180

Wójcik, S., Osman, T. and Zawada, P.

Semantic Approach for Prospectivity Analysis of Mineral Deposits.

In Proceedings of the 2nd International Conference on Geographical Information Systems Theory, Applications and Management (GISTAM 2016), pages 180-189

ISBN: 978-989-758-188-5

an ontology-based unifying metadata model, and

support the automation of the geospatial analysis by

encoding the relevant geological knowledge at the

ontology modelling and the semantic reasoning

levels. The paper also contributes to the

methodology of semantic-driven analysis in this

sector by highlighting various limitations hampering

the full automation of the prospectivity analysis

process such as free-form description of geological

attributes and the subjectivity in assessing some

structures.

The rest of this paper is organised as follows.

Section 2 reviews related work. The overall

architecture of our system is discussed in Section 3.

Section 4 provides insights into the workflow of the

spatial geodata management. Section 5 describes

how the domain knowledge was captured and its

translation into semantic ontologies. In section 6, we

detail how our proposed systems address the

challenges in automating mineral prospectivity

analysis. Section 7 evaluates the system

implementation. Section 8 concludes the paper and

presents our plans for further research.

2 RELATED WORK

Mineral exploration is one of the most important

topics in geology and arguably one that historically

motivated evolution of the science of geology.

While it is out of scope to describe the plethora of

prior art, there are some excellent efforts trying to

find synergy between computing principles and

methods and classical mineral exploration.

Most of currently used geology software suites

such as ArcGIS (Law and Collins 2013) or QGIS

(QGIS 2015), either have elements or modules

designed to simplify mineral prospectivity analysis

such as ones offered by GeoTools (Geosoft 2015), or

enable the processing of ancillary data for the same

purpose. Generally, one can assume that most

prospectivity analysis software packages are either

generalised cartography/GIS suites enriched with

geology-enabling modules, or large-scale/small-area

geodata management for mine development similar

to the system developed by MapTek Vulcan

(Maptek 2015). While almost all of these packages

can be used to help with mineral prospectivity

assessment of a given terrain, few offer

comprehensive, automated solutions and mostly rely

on deploying the geologists’ expertise in driving the

analysis the process using the software package as a

tool, thus contributing significantly to the investment

required at the early stages of mineral exploration.

The effort reported by Noack et al. in (Noack, et

al. 2012) is one of the few works attempting to use

advanced automatic techniques to guide the

prospectivity analysis process. Their approach uses

neural networks and statistical analysis to predict

presence of a terrain feature - existence of mineral

deposits or otherwise, as implemented in Beak

Advangeo software. This approach requires the user

to provide a set of training data that describes

describing terrain geology and measurements and

existing known mineral deposits or occurrences. For

most potential regions, such data is available,

however, most sourced geodata is incomplete and

therefore do not provide the necessary set of training

ground-truth to yield accurate analytical results.

Statistical analysis is also a black box, making it

difficult to trace and verify how certain decisions

were made.

In the recent years, fuzzy logic has been used in

geology context. Lusty in (Lusty, et al. 2009)

presents a fuzzy logic approach to assess gold

prospectivity of Irish geology and discusses controls

used to emphasise influence of different geological

features of the terrain. These controls present in their

approach parameterise the fuzzy logic analysis,

showing the need for proven targeting models that

are both flexible and transparent. Their discussion of

the results concludes the need to construct more

controllable and reliable methodologies for analysis,

where relation between the used criteria and analysis

result is more explicit and accurate.

Our investigation claims that Semantic Web

technologies can help to address some of these

limitations. The Semantic Web allows the modelling

of the taxonomy of the geological features as nodes

in a graph interconnected using object and data

relations, which describe the geological processes

that link those features as well as their interaction

with other elements in the target system such as non-

geothematic data, user profiles etc. Moreover,

semantically tagged data is inherently amenable to

reasoning that can be utilised to inform the

prospectivity analysis process based on pre-

compiled rules.

There is a significant body of work that focuses

on utilising semantic technologies to facilitate the

interpretation and sharing of geospatial data and

services. Zhang et al. in (Zhang, Zhao and Li 2010)

propose a framework for geospatial Semantic Web-

based spatial decision support system that provides

for heterogeneous ontology integration and web

services composition. Tian and Huang in (Tian and

Huang 2012) use purposely built semantic

ontologies to combine the Open Geospatial

Semantic Approach for Prospectivity Analysis of Mineral Deposits

181

Consortium (OGC) specifications with the Universal

Description, Discovery and Integration (UDDI)

standards in order to enhance the discovery of web

services compliant with OGS specifications and

promote utilising them for geospatial information

access. Janowicz et al. in (Janowicz, et al. 2010)

propose a user-transparent semantic enablement

layer for Spatial Data Infrastructure that promotes

the semantic interoperability of the OGC services

and facilitates reasoning to allow for their workflow-

based composition. These works offer a valuable

contribution towards creating frameworks enabling

the interoperation and composition of possibly

heterogeneous geospatial services and promoting

data exchange between them by means of ontology

alignments. However, the current MINEX sector

infrastructure provision, in terms of availability of

relevant geospatial services and ontology-aligned

geodata, suggests that the benefit from exploiting

these frameworks in the context of securing a

holistic cost-effective solution to mineral

prospectivity currently insignificant. It is therefore

necessary to build all the processes contributing to

our system architecture from the ground up.

3 SYSTEM ARCHITECTURE

Figure 1 below presents the overall architecture of

the system and illustrates the workflow between its

essential components, which begins with geodata

gathering. After identifying regions of interest,

potential data sources in the region are identified.

Data is sourced from third party public and private

organisations.

The majority of geodata are sourced from public

sector bodies such as national geological surveys

that historically have focussed on the production of

physical, paper based products. The integration of

these many maps, by different authors using

differing taxonomies, varying quality of digitisation

has therefore made the task of producing seamless

geological maps an important goal but one that has

not been achieved by many geological surveys.

Notable exceptions are however, the relatively small

scale map compilations of the Commission for the

Geological Map of the World and One Geology.

Consequently, an important stage prior to data

upload into our system is the assessment of

publically available in terms of scale, edition,

coverage, detail and digitization quality and where

necessary a data cleaning process is employed. This

is particularly important across adjacent map sheets

where line work and taxonomies used differ.

Figure 1: IGS Geodata system workflow.

To allow for further processing using Semantic

Web technologies, data has to be converted into the

appropriate format and uploaded to a spatially

enabled linked geodata store. Data conversion and

upload is a significant element of the process, in

which GIS database items are tagged with unique

identifiers and converted to data objects that are

assigned to the appropriate ontology class

(representing the geological taxonomy) in

accordance to the annotations in the source

geospatial database. The ontology model also

incorporates a set of necessary & sufficient

conditions that facilitate further classification of

basic input geological data by reasoning, for

instance, on the rock formations’ chemical and

physical properties. Next, the resultant data and the

associated polygon information are stored in a

spatially-enabled triple store, with geometries

represented by WKT (Well Known Text) strings.

Further interpretation of the data is facilitated by

a new approach to representing geological expertise.

While core concepts of geology and immutable

relations are encoded within ontologies,

prospectivity analysis required a new framework of

reference. To facilitate that, methods of geological

analysis were encoded as generic rules that guide the

prospectivity analysis process. These rules are

compiled as geospatial queries that can be fired

against the semantic triple store to evaluate the

prospectivity for a given natural resource in a

particular region.

To store our geospatially tagged geodata in

Data collection &

metadata generation

Data curation

& conversion

Sensory data

collection & curation

Data

Archiving

Semantic data

modelling

Report

generation

Integration of non-

geothematic data

Geospatial

Triplestore

Ontology

Encoded Geology

Knowled

e (SPARQL)

Visualisation &

delivery to GIS env.

Sextant semantic

geodata visualiser

Prospectivity

analysis

Fuzzy Logic

confidence analysis

GISTAM 2016 - 2nd International Conference on Geographical Information Systems Theory, Applications and Management

182

semantic format, we adopted the geospatially-

enabled triple-store Strabon (Kyzirakos et al., 2013).

It provides for storing linked geospatial data and

supports spatial datatypes enabling the serialization

of geometric objects in OGC standards WKT and

GML. Strabon is built by extending the well-known

RDF store Sesame and extends Sesame’s

components to manage thematic, spatial and

temporal data that is stored in the backend

RDBMS. Strabon supports the state of the art

semantic geospatial query languages stSPARQL and

GeoSPARQL and is integrated with the Sextant

(Bereta et al., 2013) tool that allows the seamless

visualisation of the complex geospatial query results.

The described workflow comprises the

integration of semantically tagged data, advanced

classification using model-embedded inference

conditions, and prospectivity analysis using

geospatial queries, thus enabling the departure from

the classical, project-based prospecting to an

iterative, repeatable, automated process.

4 MANAGEMENT OF

GEOSPATIAL DATA

The workflow of geodata management within our

system is illustrated in Figure 2. The data is sourced

from a multitude of suppliers with varied data

representational format and quality. Disorganised

nomenclature and use of out-dated database formats

is commonplace with missing data ranges, file

compression artefacts and noise from print

preparation techniques.

Data cleaning usually comprises of checking data

georeference and fixing geometrical errors common

to manually drawn polygons. Most of the datasets

procured suffer from some geometrical errors such

as polygon intersections without vertices. In some

severe cases, map georeference might be inaccurate,

missing or done in an obscure, locally used

projection. Sometimes even whole areas might not

conform to international standards, which is the case

with disputed borders of Venezuela, where two

neighbouring countries routinely include certain

areas within their territory.

In an overwhelming majority of cases maps

produced by different authors do not accurately

follow delineations and might even disagree about

entire border shape, as presented on Figure 3. While

correcting the above is a crucial step ensuring

adequate data representation, automation of the

process is complicated, especially where

delineations do not match at all. Correcting this

accuracy has to be done manually by a data engineer

or a geologist using a GIS editing tool.

Figure 2: Geodata sourcing and clearing workflow.

Figure 3: Discontinuities between neighbouring datasets in

the Guyana Shield region before and after cleaning.

The last step in data processing is the conversion

to the appropriate Semantic Web format. The

semantic technologies present an opportunity for

data reconciliation, enrichment and provides for

more sophisticated query mechanisms. The semantic

technology also enables integration of classical

geology data with non-geothematic data such as

cadastre data, economy-related maps and locality

descriptions through unifying metadata tagging.

The bulk of data processing is being achieved

automatically, but due to variability in data

representation and presence of freeform comments

and annotations in the original GIS representation,

some manual intervention is required to complete

the data conversion process. However, to enable the

process, basic transformations of given SQL

Semantic Approach for Prospectivity Analysis of Mineral Deposits

183

columns and records to appropriate taxonomies has

to be done manually by a person familiar with

geodata being processed. This is caused by the fact

that geospatial databases do not follow same or

similar structure, language or taxonomy and the

system requires an informed person to point out

where relevant data is located and how to translate it

into appropriate system-wide taxonomy.

Unfortunately, the working case shows that

among geodata classifications encountered in one of

the test areas only less than 10% of rock name

records were recurring among more than one

dataset, with the rest being dataset-specific. While it

was projected that it might be possible to automate

taxonomy conversion, this required supervision and

creation of dictionaries to translate between native

taxonomy and those designed for the project at hand.

This needed to be done on a case by case basis for

each dataset. It is worth noting, however, that

described data curation is limited to at most a few

man-hours per dataset and can be performed by a

fairly inexperienced geologist with little training.

5 DOMAIN ANALYSIS AND

ONTOLOGY ENGINEERING

This section describes the process of knowledge

modelling for our prospectivity analysis system, and

elaborates on the specific challenges to the MINEX

sector.

5.1 Capturing the Domain Knowledge

The domain knowledge relevant to prospectivity

analysis was compiled into a concept map that

follows intuitive conceptualisation (Osman et al.,

2013) of the proposed system integrating concepts

from the fields of geology, data processing and

visualisation. Figure 4 illustrates the segment of the

concept detaining the process of geodata analysis.

This process is split into two stages. The first is the

geodata modelling stage that is realised by the use of

ontologies and inference and uses knowledge that is

universal and applicable to any geodataset. It is

deployed using OWL ontologies to enable maximum

extensibility allowing the update of the system with

new geological concepts without invalidating

existing ones. Prospectivity analysis is implemented

at the second stage, where geospatial queries that

encode geological analytical knowledge are used to

evaluate the likelihood of mineral deposits existence

for a particular region.

Figure 4: Geodata analysis concept map.

5.2 Core Ontology Design

Our semantic modelling approach for prospectivity

analysis is implemented in two phases. The first

phase described in this section discusses the

modelling of the taxonomy structure of the MINEX

domain, while the next section details the

engineering of the necessary & sufficient conditions

driving the geological classification of new geodata

instances.

British Geological Survey with support from

Commission for the Management and Application of

Geoscience Information, IUGS and OGC has made

an excellent effort to provide a modern, complete

vocabulary of geological terms and concepts based

on XML language and called GeoSciML (Sen and

Duffy, 2005) that were compiled by Smyth and

Jondeau, members of the SEEGRID community,

into a semantic OWL Ontology (CGI Geoscience

Concept Definitions Task Group). Transparent

international standards are crucial to the

development of any innovative system and in this

case this work has been used as a basis for IGS

geodata modelling. We adopted the ontology as the

basis for modelling the taxonomy of lithological

concepts and properties in our ontology and

extended it to represent various subdomains of

geology present including various types of lithology,

geological structures, tectonics, geophysics etc.

Figure 5 below illustrates the geological

classification in our ontology where base rock type

classes are semantically annotated with semantic

object properties such as particle types and sizes,

chemical composition category or consolidation

degrees. This provides excellent opportunity to

create a unified classification of all rock properties

GISTAM 2016 - 2nd International Conference on Geographical Information Systems Theory, Applications and Management

184

and facilitate seamless data interpretation.

Figure 5: Excerpt from IGS ontology classes.

5.3 Classification of Geological Feature

by Composite Properties Inference

The Semantic Web ontology language (OWL)

allows the definition of a set of necessary and

sufficient conditions that assert class membership.

An individual fulfilling these conditions will be

automatically ‘inferred’ as belonging to the class.

We utilise this facility of OWL in our ontology to

compile a set of necessary and sufficient conditions

that define geological features by describing their

composite properties, which in turn will

automatically infer the appropriate geological

feature for newly sourced geodata instances.

Figure 6: Phyllonite rock definition encoded as object

relations (explicit above & inferred).

Figure 6 provides an example of encoding the

geological definition (class membership) for

phyllonite by a set of necessary and sufficient

conditions that explicitly denote its exclusively

pelitic constituent parts, mylonitic foliation fabric

type, fine or sand-size grains, crystalline particle

type etc.

5.4 Challenges in Semantic Geological

Modelling

We encountered two problems with data accuracy in

the sourced geodata sets. The first is the

contradictory information input in the dataset as a

result of a mistake or misconception, which usually

stems from assessments done by authors using

unclear criteria for differentiation, such as lack of

clear distinction between similar structures or types

of rock formations. A common example is the

practice of assigning properties as freeform

comments to features that should not have them, e.g.

assigning 'majorly mud-sized grain' property to a

block of muddy sandstone, which by definition is a

sand size grained rock, with a minority of mud-sized

grains. In the project we can't use such contradictory

statement and have to decide whether to keep

assigned grain size and change rock type or keep the

rock type while assigning different grain size. This

has been delegated to a person to make an informed

decision while performing data upload step

described in the previous section.

The second source of inaccuracy is the

discrepancy between basic geology definitions used

by different map authors, especially if affiliated to

different geology institutions. One of the major

efforts in ontology design was to redesign

classifications for the purpose of overriding

otherwise subjective values found in source data

with clear and universal definitions, while

preserving the internationally accepted nomenclature

as much as it was possible. Thus, properties such as

granularity, basic chemical compositions, genetic

categories and metamorphic grades have been

defined. These properties were encoded in our

ontology as class instances (individuals) to further

categorise the classes by certain attributes. For

instance, the grain size property aided in

categorising rock types into igneous, sedimentary

and clastic. This has been resolved by creating

comprehensive grain size scales, with equivalent

(aliased) subclasses to preserve traditional

nomenclature. However, we could not use class

instances (individuals) to denote the equivalent

geological definitions as OWL classification can

Semantic Approach for Prospectivity Analysis of Mineral Deposits

185

only be based on class definitions. Therefore, we

had to create several subcategories to allow for the

mapping between the overlapping geological

definitions, which somewhat bloated the ontology as

the subcategories classes, such as grain size, hosted

only one individual at any time.

Finally, it's worth noting that geological

knowledge is not exclusively contained within maps.

Reputable data sources publish their surveys in the

form of a map (often in GIS DB format)

accompanied by head geologist memoirs, report or

commentary, sometimes even embedded into the

database itself (see Figure 7). Some of those

comments carry invaluable information about

surveyor's findings at a given locality that can enrich

the geological database. Since automatic analysis of

freeform comments that deploy natural language

processing would be difficult and expensive to

implement, such data entries are manually

transformed into taxonomy items and properties by a

geologist at the time of data input.

Figure 7: Example of typical geology data (Venezuela)

with important info encoded as a freeform comment.

6 AUTOMATING MINERAL

PROSPECTIVITY ANALYSIS

Geology is often regarded by professionals to have

an element of art to it, and the consensus is that the

geologist should drive the prospectivity analysis

process. Our objective is to shorten the delivery time

and reduce the cost of the early stages of mineral

exploration by automating the lion-share of the

process of prospectivity analysis tasks, and only

deploying geological expertise at the one-off

semantic modelling phase and in minor supervisory

role related to data curation.

The last section discussed the semantic

modelling in our system and how we hard-wired

necessary & sufficient conditions into our ontology

that automatically classifies newly sourced geodata

instances into the appropriate geological categories.

This section describes the final phase of our

prospectivity analysis system, where we encode the

relevant geological knowledge as generic rules that

guide the prospectivity analysis process. These rules

are compiled as geospatial queries that can be fired

against the semantic triple store to evaluate the

prospectivity for a given natural resource in a

particular region.

The queries combine searching for geologically

interesting map features and spatial analysis of

geometries representing these features. A set of

queries retrieves polygons, which have parameters

indicating likelihood of existence of mineral

deposits at a given location, such as favourable rock

type, appropriate age or evidence of geological

processes necessary for mineralisation event. The

results of those queries are subjected to spatial

analysis, which transform retrieved geometries into

more appropriate format using operations of unions

and intersections as designed in the geological rules.

The geological rules were encoded as natural

language statements using intuitive geological

terminology. An example of encoding such a

statement can be seen on Figure 8. Mnemonic form

(in bold) of a geological rule used in the process of

gold prospectivity analysis is followed by its verbose

phrasing (in italics), explaining in detail what

geological features are being searched. An

stSPARQL expression of the same meaning is

presented below.

Figure 8: Example of typical geology data (Venezuela)

with important info encoded as a freeform comment.

It is noteworthy to mention that due to rock

formations following trends beneath the visible rock

outcrops that may not be evident from the data,

prospectivity of a certain buffer area around

geological features is affected. Even when one can

corroborate geophysical information to discover

those trends, geological processes are not always

limited to the volume of rock in question; contrarily,

mineralisation may occur away from the source

material. The range of this processes is very hard to

estimate and in certain cases has be performed

GISTAM 2016 - 2nd International Conference on Geographical Information Systems Theory, Applications and Management

186

empirically, but there are already methods of

establishing optimal spatial parameters in other

works. In a similar fashion, one has to recognise the

need for additional parameters - weight values for

establishing finer control of the influence of various

queries over the analysis or logical ones, to compile

a number of queries' results into one coherent result.

Thus, a number of adjustable parameters are present

in the system, increasing its flexibility, but also

requiring careful consideration at the analysis design

process.

One of the challenges encountered when

designing the system architecture was to decide

which elements of the analytical process should be

modelled into the ontology and which should be

implemented by means of explicit reasoning. The

decision is made on a case-by-case basis, with

arguments both for and against each of the

implementation methods. The main focus of the

design was to reduce the effort required to extend

and modify the system to incorporate additional

features such as new deposit models or use of data

of new types. Hence, wherever possible, we encoded

directly into the ontology model all the ‘universal

and immutable’ knowledge related to the

classification of newly sourced geodata instances

into geological categories, and only resorted to

explicit reasoning for the final stage of encoding the

rules evaluating the mineral prospectivity. Hence,

surficial and below-surficial features were hard-

wired into the semantic ontology, while the rules

encoded as geospatial queries focused on geological

processes of local range and objective approach to

geological analysis. This helped to decouple the

geodata processing from the analysis process thus

reducing the workload for designing and encoding

new rules and promoting their usability.

The identification and interpretation of

geological features on a map by a geologist is a

highly assimilative, cognitive and nuanced process.

It requires the experience, often collective of

integrating a sequence of features such as the spatial

distribution of strata at the surface, which can be

extrapolated to depth (3D visualization skills), age

relationships (relative and absolute) of adjacent and

intersecting features and many other components,

which are set within the constraints of a larger or

macro geological context. The above process is one

that is not easily replicated by current computing

methodologies. A good example is the definition of

a basin, which is a large geological depression, a

result of tectonic warping. Most of the common

features of basins - such as hipsometric depression,

normal younging of the strata and presence of

sedimentary material is found not only in basins but

also in other geological structures such as some

craters or some glacier valleys. To further

complicate the issue, detecting basins automatically

using stSPARQL geospatial queries would require

the queries to contain large amount of incidental

geological knowledge, specific to a given location.

For example, some basins can be detected by careful

spatial analysis of 3D model of topology, and rock

strata formation to distinguish them from impact

craters. However, some are filled with sediment due

to their age and don't show on the terrain model,

while other might have underwent geothermal

processes that disrupt original rock layers - both of

which are quite easy to spot by a geologist while

remaining hard to encode. Recognising that

removing false negatives and positives would be

laborious and require a geologist to perform a

supervisory role despite automation, the task of

recognising basins and other similar terrain features

has been delegated to a geologist during the data

input process (Figure 2), to manually add

appropriate taxonomy items.

7 SYSTEM EVALUATION

Geology knowledge is encoded into an ontology

data model and rule-driven prospectivity analysis

process. This required significant amount of

preparatory work by an experienced exploration

geologist in cooperation with a semantic technology

specialist to transform his knowledge into a

machine-readable form. The main advantage of the

system is capability to house an extraordinary

amount of geology knowledge, which is

automatically applied to a large set of geodata, from

which value-added analytical products can be

generated and delivered and updated as needed.

The above is accomplished without sacrificing

the transparency of the process, as explicitly

represented queries and ontologies are human-

readable and their outcomes can be backtracked. The

mutable parameters and separation between geology

model and prospectivity analysis allows for their

seamless modification and extension, which gives

our approach an advantage over statistical and

machine learning approaches, access to the

intricacies of the analytical process is difficult.

Data enrichment is evidenced by the increasing

number of relations in the system. For the test

dataset close to 8000 triples have been present at the

beginning of the process, while after applying the

geological model, that number increased to over

Semantic Approach for Prospectivity Analysis of Mineral Deposits

187

60000. From that data prospectivity maps for 30

different types of deposits can be produced, each

using a combination of 10-20 purpose-designed

stSPARQL queries. The result of combinations of

those queries is a set of polygons, delineating areas

of similar prospectivity rating. These polygons then

are stored within the geospatial database and

accessed by Sextant data visualiser seamlessly and

without the need of format conversion. Thus, one is

able to quickly produce a custom map compilation,

by compiling polygons straight from geospatial

database in desired combination.

Generating a prospectivity map in a visual form

is the goal of the analysis process. As shown on

Figure 9, delineation of different grades of

prospectivity for minerals over Google Maps

background enables approachable presentation,

which can be inspected even in image format. By

delivering it in an accessible form, effects of the

analysis can be included in a decision making

process, even without specialised knowledge or

tools.

While semantic technologies have been very

useful in modelling geology, some of the data types

proved to be difficult to describe, especially those

that in classical prospectivity analysis need to be

heavily modified and carefully examined by an

expert geologist, such as raster-based geophysical

measurements. Automation of geophysical analysis

is often hampered by the fact that very similar

patterns can be a result of a great number of

different subsurface features. With the exception of

dykes, recognising most subsurface rock bodies or

structures relies on geophysicist’s experience and

judgment and while automation of the delineation

process is currently implemented, it is still a

significant challenge to classify the delineated

structures. Since the automation of geophysical

analysis is a computationally intensive process using

advanced algorithms, it was impractical to

incorporate it fully and geophysics role has been

relegated to data validation and extrapolation. There

are plans to revisit this issue in the future.

One of the projected benefits of the project was

to be able to add more data without the need of re-

applying the analysis. This is only achieved for the

modelling stage, but not the querying stage. Because

of efficiency limitations of Sextant visualiser, very

complicated queries have to be run and their results

cached in advance. These caches need to be updated

each time new data is being added to the system,

which in practice happens infrequently, but poses an

additional difficulty. This inefficiency issue is not

present in Strabon triple store, so a different

visualiser might be able to perform all queries on

demand. The system is not required to provide real-

time response, and current processing times of under

a minute to perform modelling and 3 minutes per

deposit type are acceptable.

Figure 9: Polygons denoting prospectivity for orogenic

gold visualised over a map of northern French Guiana.

8 CONCLUSIONS AND FUTURE

WORK

Motivated by the need for more cost-effective

approach to prospectivity analysis, we presented

new semantic-driven integrative approach to

prospectivity analysis. Our approach initially

deploys an ontology-based unifying metadata model

to reconcile the discrepancy the representational

format of geodata that is sourced from multiple

private and public suppliers often with disorganised

nomenclature and non-digitised freeform text

describing geological features that are critical to the

analysis process. Our approach then uniquely utilises

semantic modelling to support the automation of the

prospectivity analysis by encoding the relevant

geological knowledge at the ontology modelling and

the semantic reasoning levels. At the semantic

modelling level, we hard-wired necessary &

sufficient conditions into our ontology to

automatically classify newly sourced geodata

instances into the appropriate geological categories,

and at the explicit reasoning level we encode the

relevant geological knowledge as generic rules that

guide the prospectivity analysis process. These rules

are compiled as geospatial queries that can be fired

against the semantic triple store to evaluate the

prospectivity for a given natural resource in a

particular region. We endeavour to strike the balance

between the elements of the analytical process

encoded at each level in order to decouple the

geodata processing activity from the prospectivity

GISTAM 2016 - 2nd International Conference on Geographical Information Systems Theory, Applications and Management

188

analysis process, thus reducing the workload for

designing and encoding new prospectivity rules and

promoting their seamless extensibility.

The reported work in this paper also contributes

to the methodology of utilising semantic

technologies for mineral prospectivity analysis by

investigating the practical constraints hindering the

complete automation of the prospectivity analysis

process. Such limitations include the misleading

assignment of properties as freeform comments to

features in the sources geodata, the complexity in

modelling geophysical measurements, and the

limitation of the visualisation tool in caching the

geospatial query results.

Our plans for future research involve the curation

and processing of sensory raster data that comprises

geophysical measurements, various types of imaging

and LIDAR data. We are optimistic this will further

improve the accuracy of our prospectivity analysis

model. We also intend to investigate the use of fuzzy

logic to model the certainty in the perceived

accuracy of the prospectivity analysis as a function

of quality and completeness of the sourced geodata.

ACKNOWLEDGEMENTS

This research was partially supported by Innovate

UK through a Knowledge Transfer Partnership

funding (KTP009221).

REFERENCES

Bereta, K., Nikolaou, C., Karpathiotakis, M., Kyzirakos,

K. and Koubarakis, M., 2013. SexTant: Visualizing

Time-Evolving Linked Geospatial Data. In:

International Semantic Web Conference (Posters &

Demos), pp. 177-180.

CGI Geoscience Concept Definitions Task Group, OWL

Ontology for Lithology [online]. . Available at:

https://www.seegrid.csiro.au/wiki/CGIModel/Concept

DefinitionsTG#OWL_Ontology_for_Lithology

[Accessed 12,09 2015].

Geosoft, 2015. ArcGIS Extensions, Mapping Software for

ArcGIS [online]. Available at: http://www.geosoft.

com/products/arcgis-extensions/overview [Accessed

12/09 2015].

IGS, 2015. International Geoscience Services (IGS) Ltd

[online]. Available at: http://www.igsint.com/

[Accessed 12/09 2015].

Janowicz, K., Schade, S., Bröring, A., Keßler, C., Maué,

P. and Stasch, C., 2010. Semantic Enablement for

Spatial Data Infrastructures. Transactions in GIS, 14

(2), 111-129.

Kyzirakos, K., Karpathiotakis, M., Bereta, K., Garbis, G.,

Nikolaou, C., Smeros, P., Giannakopoulou, S.,

Dogani, K. and Koubarakis, M., 2013. The

spatiotemporal RDF store strabon. In: The

spatiotemporal RDF store strabon. Advances in Spatial

and Temporal Databases. Springer, 2013, pp. 496-

500.

Law, M., and Collins, A., 2013. Getting to know ArcGIS

for desktop. Esri Press.

Lusty, P., Gunn, A., McDonnell, P., Chacksfield, B.,

Cooper, M. and Earls, G., 2009. Gold potential of the

Dalradian rocks of north-west Northern Ireland:

prospectivity analysis using Tellus data. Applied Earth

Science: Transactions of the Institutions of Mining and

Metallurgy: Section B, 118 (3-4), 162-177.

Maptek, 2015. Mine Planning Solutions [online]. .

Available at: http://www.maptek.com/solutions/

mine_planning.html [Accessed 12/09 2015].

Noack, S., Barth, A., Irkhin, A., Bennewitz, E. and

Schmidt, F., 2012. Spatial modeling of natural

phenomena and events with Artificial Neural

Networks and GIS. International Journal of Applied

Geospatial Research (IJAGR), 3 (1), 1-20.

Osman, T., Rmaswamy, S., Mahmoud, S. and Saeed, M.,

2013. Utilising semantic technologies for decision

support in dementia care. In: Computer Modelling and

Simulation (UKSim), 2013 UKSim 15th International

Conference on, IEEE, pp. 628-633.

QGIS, 2015. Open Source Geographic Information System

[online]. . Available at: http://www.qgis.org/

[Accessed 12/09 2015].

Sen, M., and Duffy, T., 2005. GeoSciML: development of

a generic geoscience markup language. Computers &

Geosciences, 31 (9), 1095-1103.

SNL, 2015. Expert Mining Research [online]. . Available

at:

http://www.snl.com/Sectors/MetalsMining/ExpertMini

ngResearch.aspx [Accessed 12/09 2105].

Tian, Y., and Huang, M., 2012. Enhance discovery and

retrieval of geospatial data using SOA and Semantic

Web technologies. Expert Systems with Applications;

Expert Syst.Appl., 39 (16), 12522-12535.

Zhang, C.R., Zhao, T. and Li, W., 2010. The framework of

a geospatial semantic web-based spatial decision

support system for Digital Earth. International Journal

of Digital Earth; Int.J.Digit.Earth, 3 (2), 111-134.

Semantic Approach for Prospectivity Analysis of Mineral Deposits

189