Www.csiro.au Observations and sampling: common patterns Simon Cox CSIRO Exploration and Mining 7...

www.csiro.au

Observations and sampling: common patterns

Simon Cox

CSIRO Exploration and Mining

7 March 2007

SEEGrid Observations, 2006-11-30 2 of 22

Science relies on observations

Evidence & validation

Involves sampling

A cross-domain terminology and information-model


What is “an Observation”

Observation act involves a procedure applied at a specific time and place (Fowler & Odell, 1997ish)

Result of an observation is an estimate of some property value

The property is associated with the observation domain or feature of interest

The location of the procedure may not be the location of interest for spatial analysis of results


Observed property

Sensible phenomenon or property-type

Length, mass, temperature, shape

location, event-time

colour, chemical concentration

count/frequency, presence

species or kind

Expressed using a reference system or scale

Scale may also be ordinal or categorical

May require a complex structure

“Sensible”, but not necessarily physical …


Feature-of-interest

The observed property is associated with something

Location does not have properties, the substance or object at a location does

The property must be logically consistent with the feature-type, as defined in the application domain

E.g. rock-density, pixel-colour, city-population, ocean-surface-temperature

The proximate feature-of-interest may merely sample the more meaningful domain-feature

Rock-specimen samples an ore-body

Well samples an aquifer

Sounding samples an ocean/atmosphere column

Cross-section samples a rock-unit

… i.e. the Observation-target


«FeatureType»Observ ation

+ quality: DQ_Element [0..1]+ responsible: CI_ResponsibleParty [0..1]+ result: Any

«FeatureType»Event

+ eventParameter: TypedValue [0..*]+ time: TM_Object

«DataType»TypedValue

+ property: ScopedName+ value: Any

«Union»Procedure

+ procedureType: ProcedureSystem+ procedureUse: ProcedureEvent

AnyIdentifiableObject

«FeatureType»AnyIdentifiableFeature

AnyDefinition

«ObjectType»Phenomenon

+followingEvent 0..*+precedingEvent 0..*

+generatedObservation

0..*

+procedure 1

+observedProperty1{Definition must be of aphenomenon that is a propertyof the featureOfInterest}

+propertyValueProvider

0..*

+featureOfInterest

1

A common pattern: the observation model

An Observation is an Event whose result is an estimate of the value of some Property of the Feature-of-interest, obtained using a specified Procedure


When is this viewpoint interesting?

Primarily if the data-acquisition metadata is of concern


Procedures

Instruments & Sensors

Respond to a stimulus from local physics or chemistry

Intention may concern local or remote source

Sample may be in situ or re-located


Procedures are usually process chains

Procedure often includes data processing, to transform “raw” data to semantically meaningful values

Voltage orientation

count radiance NDVI

Position + orientation scene-location

Mercury meniscus level temperature

Shape/colour/behaviour species assignment

This requires consideration of “sensor”-models and calibrations


Advanced procedures

Modelling, simulation, classification are procedures

“raw” data == modeling constraints (sensor-outputs, process-inputs)

“processed” data == simulation results (outputs)

“interpreted” data == classification results (outputs)

SensorML provides a model and syntax for describing process-chains








«Union»Procedure




AnyDefinition




0..*

+procedure 1



0..*

+featureOfInterest

1

Application to a domain

feature of interest Feature-type taken from a domain-model

observed property Member of feature-of-interest-type

procedure Suitable for property-type


Conceptual object model: features

Digital object corresponding with identifiable, typed, object in the real world

mountain, road, specimen, event, tract, catchment, wetland, farm, bore, reach, property, license-area, station

Feature-type is characterised by a specific set of properties

Specimen

ID (name)

description

mass

processing details

sampling location

sampling time

related observation

material

…


Geology domain model - feature type catalogue

Borehole collar location shape collar diameter length operator logs related observations …

Fault shape surface trace displacement age …

Ore-body commodity deposit type host formation shape resource estimate …

Conceptual classification

Multiple geometries

Geologic Unit classification shape sampling frame age dominant

lithology …

License area issuer holder interestedParty shape(t) right(t) …


Water resources feature type catalogue

Aquifer

Storage

Stream

Well

Entitlement

Observation

…


Meteorology feature type catalogue

Front

Jetstream

Tropical cyclone

Lightning strike

Pressure field

Rainfall distribution

…

Bottom two are a different kind of feature


Spatial function: coverage

(x1,y1)

(x2,y2)

Variation of a property across the domain of interest

For each element in a spatio-temporal domain, a value from the range can be determined

Used to analyse patterns and anomalies, i.e. to detect features (e.g. storms, fronts, jetstreams)

Discrete or continuous domain

Domain is often a grid

Time-series are coverages over time


Features vs Coverages

Feature

object-centric

heterogeneous collection of properties

“summary-view”

Coverage

property-centric

variation of homogeneous property

patterns & anomalies

Both needed; transformations required


“Cross-sections” through collections

Specimen Au (ppm) Cu-a (%) Cu-b (%) As (ppm) Sb (ppm)

ABC-123 1.23 3.45 4.23 0.5 0.34 A Row gives properties of one feature

A Column = variation of a single property across a domain (i.e. set of locations)


Some feature types only exist to support observations

class Figure: samplingBase


«FeatureType»Surv eyProcedure

«FeatureType»SamplingFeature

«estimatedProperty»+ property: TypedValue [0..*]

constraints{each relatedObservation.observedProperty must match a self.property.propertyType}{relatedObservation.featureOfInterest=self}

«DataType»SamplingFeatureRelation

+ role: GenericName

«FeatureType»AnyFeature


+ propertyType: PropertyType+ value: Any

Substitutability of SamplingFeature f orAny Feature is implied by the class stereoty pe <<FeatureTy pe>>

propertyValueProvider

0..*

featureOfInterest

1

relatedObservation

0..*

surveyDetails 0..1

Intention

sampledFeature

1..*

relatedSamplingFeature0..*

source

0..*

target


Extensive sampling feature types

Station

+ elevation: DirectPosition [0..1]+ position: GM_Point

SamplingFeature

+ responsible: CI_ResponsibleParty [0..1]

Trav erse

Flightline

Profile

+ begin: GM_Point+ end: GM_Point+ length: RelativeMeasure [0..1]

Shape3D

SurfaceOfInterest

+ area: RelativeMeasure [0..1]

Interv al

Shape2D

SolidOfInterest

+ volume: RelativeMeasure [0..1]

Shape1D

SamplingFeatureCollection

constraints{count(member)>=1}

Swath

Section

Surv eyProcedure

Sounding

LidarCloud

Specimen

+ currentLocation: Location [0..1]+ mass: Measure+ material: CV_Coverage

+shape 1+shape 1+shape 1

+member 0..*

+surveyDetails

0..1


Assignment of property values

For each property of a feature, the value is either

i. asserted

name, owner, price, boundary (cadastral feature types)

ii. estimated

colour, mass, shape (natural feature types)

i.e. error in the value is of interest


Variable property values

Some property values are not constant

colour of a Scene or Swath varies with position

shape of a Glacier varies with time

temperature at a Station varies with time

rock density varies along a Borehole

Variable values may be described as a Coverage over some axis of the feature


RockSample-A :Specimen

DensityItA :Observ ation

Density :Phenomenon

Densitometry :Observ ationProcedure

2610 kg/T :Measure

2006-11-23 :TM_Instant

Leederv ille, WA :Location

RockSample-B :Specimen

DensityItB :Observ ation

2580 kg/T :Measure


West Leederv ille, WA :Location

+time+result

+procedure+observedProperty

+featureOfInterest

+sampl ingLocation

+density

+sampl ingLocation

+time


+featureOfInterest

+result

+density

ProbeItA :Observ ation

Material :Phenomenon

Microprobe :Observ ationProcedure

MineralDistribution :CV_Cov erage

2006-11-24/2006-11-26 :TM_Period



+observedProperty +procedure

+result+time

+material

+featureOfInterest

+sampl ingLocation RockSample-A :Specimen

2610 kg/T :Measure


+density



Density :Phenomenon


2610 kg/T :Measure



+featureOfInterest


+result

+density

+time


2610 kg/T :Measure


RockSample-B :Specimen

2580 kg/T :Measure

West Leederv ille, WA :Location

+density

+sampl ingLocation

+density

+sampl ingLocation

ProbeItA :Observ ation


Microprobe :Observ ationProcedure


2006-11-24/2006-11-26 :TM_Period



Density :Phenomenon


2610 kg/T :Measure




+result+time

+featureOfInterest

+material

+featureOfInterest


+result

+density

+time

+sampl ingLocation



2610 kg/T :Measure


+material

+density

+sampl ingLocation

Observations, features and coverages

Feature summary

Property-valueevidence

Multiple observations one feature, different properties:feature summary evidence

A property-valuemay be a coverage

Same property onmultiple samplesis a another kindof coverage

Multiple observations different features, one property:coverage evidence


Development and validation

O&M conceptual model and XML encoding

Developed in the context of

XMML Geochemistry/Assay data

OGC Sensor Web Enablement – environmental and remote sensing

Subsequently applied in

Water resources/water quality (WQDP, AWDIP, WRON)

Oceans & Atmospheres (UK CLRC, UK Met Office)

Natural resources (NRML)

Taxonomic data (TDWG)

Geology field data (GeoSciML)

I could have put dozens of logos down here


Status

OGC Best Practice paper, r4 – 2006

OGC RFC ends 2007-03-08 (tomorrow!)

OGC Adopted Specification – mid/late 2007?

ISO specification – 2008-9?


Conclusions

Different viewpoints of same information for different purposes

Summary vs. analysis

Some values are determined by observation

Sometimes the description of the estimation process is necessary

Transformation between views important

Management of observation evidence can be integrated

(Bryan Lawrence issues)

For rich data processing, rich data models are needed

Explicit or implicit

Data models (types, features) are important constraints on service specification

www.csiro.au

Thank You

CSIRO Exploration and Mining

Name Simon Cox

Title Research Scientist

Phone +61 8 6436 8639

Email [email protected]

Web www.seegrid.csiro.au

Contact CSIRO

Phone 1300 363 400

+61 3 9545 2176

Email [email protected]

Web www.csiro.au


Features, Coverages & Observations (1)

Observations and Features

An observation provides evidence for estimation of a property value for the feature-of-interest

Features and Coverages (1)

The value of a property that varies on a feature defines a coverage whose domain is the feature

Observations and Coverages (1)

An observation of a property sampled at different times/positions on a feature-of-interest estimates a discrete coverage whose domain is the feature-of-interest

feature-of-interest is one big feature – property value varies within it


Features, Coverages & Observations (2)

Observations and Features

An observation provides evidence for estimation of a property value for the feature-of-interest

Features and Coverages (2)

The values of the same property from a set of features constitutes a discrete coverage over a domain defined by the set of features

Observations and Coverages (2)

A set of observations of the same property on different features provides an estimate of the range-values of a discrete coverage whose domain is defined by the set of features-of-interest

feature-of-interest is lots of little features – property value constant on each one


premises:

O&M is the high-level information model

SOS is the primary information-access interface

SOS can serve:

an Observation (Feature)

getObservation == “getFeature” (WFS/Obs) operation

a feature of interest (Feature)

getFeatureOfInterest == getFeature (WFS) operation

or Observation/result (often a time-series == discrete Coverage)

getResult == “getCoverage” (WCS) operation

or Sensor == Observation/procedure (SensorML document)

describeSensor == “getFeature” (WFS) or “getRecord” (CSW) operation

Sensor service

optional – probably required for dynamic sensor use-cases


SOS vs WFS, WCS, CS/W?

WFS/Obs

getFeature, type=Observation

WCS

getCoverage

getCoverage(result)

Sensor Registry

getRecord

SOS

getObservation

getResult

describeSensor

getFeatureOfInterest

WFSgetFeature

SOS interface is effectively a composition of (specialised) WFS+WCS+CS/W operations

e.g. SOS::getResult == “convenience” interface for WCS


ISO 19101, 19109 General Feature Model

Properties include

attributes

associations between objects

value may be object with identity

operations

Metaclass diagram


class Fig 03 - CV_Cov erage subclasses

CV_GeometryValuePair{n}

+ geometry: CV_DomainObject+ value: Record

«Abstract»CV_ValueObject

{n}

+ geometry: CV_DomainObject+ interpolationParameters[0..1]: Record

+ interpolate(DirectPosition) : Record

Discrete Cov erages::CV_DiscreteCov erage{n}

+ locate(DirectPosition) : Set<CV_GeometryValuePair>

«Abstract»CV_ContinuousCoverage

{n}

+ interpolationParametersType[0..1]: Record+ interpolationType: CV_InterpolationMethod

+ locate(DirectPosition) : Set<CV_ValueObject>+ locateRegion(GM_Object) : Set<CV_ValueObject>

«CodeList»CV_InterpolationMethod

{n}

+ barycentric: + bicubic: + bi l inear: + biquadratic: + cubic: + l inear: + lostarea: + nearestneighbor: + quadratic:

«Abstract»CV_Coverage

{n}

+ commonPointRule: CV_CommonPointRule+ domainExtent[1..*]: EX_Extent+ rangeType: RecordType

+ evaluate(DirectPosition, Sequence<CharacterString>) : Record+ evaluateInverse(Record) : Set<CV_DomainObject>+ find(DirectPosition, Integer) : Sequence<CV_GeometryValuePair>+ l ist() : Set<CV_GeometryValuePair>+ select(GM_Object, TM_Period) : Set<CV_GeometryValuePair>

geometry implements the association Domain in Figure 2value implements the association Range in Figure 2

+extension

0..*Control

+controlValue

1..*

+col lection 0..*

CoverageFunction

+element 1..*

+col lection 0..*

CoverageFunction

+element 1..*

ISO 19123 Coverage model


«DataType»CV_GeometryValuePair


CV_Coverage

CV_DiscreteCov erage

«DataType»CV_PointValuePair

+ geometry: GM_Point

CV_DiscretePointCov erage

+element 1..*

+collection 0..*

+collection 0..*

+element 1..*



CV_Coverage


«DataType»CV_PointValuePair

+ geometry: GM_Point

CV_DiscretePointCov erage CV_DiscreteTimeInstantCov erage

«DataType»CV_TimeInstantValuePair

+ geometry: TM_Instant

+element 1..*

+collection 0..*

+collection 0..*

+element 1..* +element 1..*

+collection 0..*

Discrete coverage model



CV_Coverage


+element 1..*

+collection 0..*


Value estimation process: observation

An Observation is a kind of “Event Feature type”, whose result is a value estimate,

and whose other properties provide metadata concerning the estimation process








«Union»Procedure




AnyDefinition




0..*

+procedure 1



0..*

+featureOfInterest

1

Observation model – Value-capture-centric view

An Observation is an Event whose result is an estimate of the value of some Property of the Feature-of-interest, obtained using a specified Procedure


“Cross-sections” through collections

Specimen Au (ppm) Cu-a (%) Cu-b (%) As (ppm) Sb (ppm)

ABC-123 1.23 3.45 4.23 0.5 0.34 A Row gives properties of one feature

A Column = variation of a single property across a domain (i.e. set of features)

A Cell describes the value of a single property on a feature, often obtained by observation or measurement


Feature of interest

may be any feature type from any domain-model …

observations provide values for properties whose values are not asserted

i.e. the application-domain supplies the feature types


SamplingFeature

Specimen


SamplingFeature

Specimen


Observation

Measurement

+ result: RelativeMeasure

Observation

Cov erageObserv ation

+ result: CV_DiscreteCoverage

Mass :Phenomenon


+observedProperty


+featureOfInterest

+observedProperty


+featureOfInterest

Observations support property assignment

These must match if the observation is coherent with the feature property

Some properties have interesting types …


Observations and coverages

If the property value is not constant across the feature-of-interest

varies by location, in time

the corresponding observation result is a coverage

individual samples must be tied to the location within the domain, so result is set of e.g.

time-value

position-value

(stationID-value ?)

Time-series observations are a particularly common use-case

Www.csiro.au Observations and sampling: common patterns Simon Cox CSIRO Exploration and Mining 7...

Documents

Transcript of Www.csiro.au Observations and sampling: common patterns Simon Cox CSIRO Exploration and Mining 7...