Describing change in the real world: from observations to events

Gilberto CamaraKarine Reis FerreiraAntonio Miguel MonteiroINPE – National Institute for Space Research

AGILE Conference 2012, Avignon (France)

Useful References

AU Frank, “One step up the abstraction ladder: combining algebras – from functional pieces to a whole”, COSIT 1999

RH Guting et al., “A foundation for representing and querying moving objects”, ACM Transactions on Database Systems, 2000

M Worboys, “Event-oriented approaches to geographic phenomena”, IJGIS, 2003

A Galton & R Mizoguchi, “The Water Falls but the Waterfall does not Fall: New Perspectives on Objects, Processes and Events”, Applied Ontology, 2009.

W Kuhn, “A Functional Ontology of Observation and Measurement”, GeoS 2009.

Terrestrial

Airborne

Near-Space

LEO/MEO Commercial Satellites and Manned Spacecraft

Far-Space

L1/HEO/GEO TDRSS & CommercialSatellites

Dep

loyab

le

Perm

an

en

t

Forecasts & Predictions

Aircraft/Balloon Event Tracking and Campaigns

User Community

Vantage Points

Capabilities

Welcome to the Age of Data-intensive GIScience!

Data-intensive GIS = principles and applications of geoinformatics for handling very large data sets

Which data is out there?How to organize big spatial data?

How to get the data I need?

Challenges for data-intensive GIScience

How to model big data?How to access and use big data?

Data-intensive GIS is not “more maps”Spatio-temporal data that captures changeWe need new theories and methods

Objects and events

The coast of Japan is an object

The 2011 Tohoku tsunami was an event

Processes and events

Flying is a process - Virgin flight VX 112 (LAX-IAD) on 26 Apr 2012 is an event

When did the Aral Sea shrank to 10% of its original size?

Aral Sea (an object) – disaster (an event)

objects exist, events occur

Mount Etna is an objectEtna’s 2002 eruption was an event

A view on processes and events

Objects EventsMatter Processes

Space Time

Count

Mass

water or lake? football or game?

(Worboys & Galton)

A pragmatic view on objects and events

Objects EventsMatter Processes

Space TimeObservable

Abstract

water or lake? football or game?

Object (GPS buoy) + event (tsunami)

Data types for moving objects (Guting) mpoint: instant → point  mregion: instant → region

Frank, Kuhn, Guting – algebras are better than 1st order logic for modelling geo-things

Data types for moving objects (Guting)

flight (id: string, from: string, to: string, route: mpoint)

weather (id: string, kind: string, area: mregion)

Detecting flood (gauges in Netherlands)

Source: Llaves and Renschler, AGILE 2012

Event processing architecture

Source: ENVISION project (http://www.envision-project.eu/)

source: USGS

Events are categories (Frank, Galton)

identity : id · a = a

composition : a, b, c, c = a.b ∀ ∀ ∃

associativity : a · (b · c) = (a · b) · c

How can we design an algebra for spatiotemporal data that represents change?

Observations allow us to sense external reality

Observations allow us to sense external reality

An observation is a measure of a value in a location in space and a position in time

Building blocks: Basic Types

type BASE = {Int, Real, String, Boolean}

operations: // lots of them…

Building blocks: Geometry (OGC)

type GEOM = {Point, LineString, Polygon, MultiPoint, MultiLineString, MultiPolygon}

operations: equals, touches, disjoint, crosses, within, overlaps, contains, intersects: GEOM x GEOM → Bool

Building blocks: Time (ISO 19108)

type TIME = {Instant, Period}

operations: equals, before, after, begins, ends, during, contains, overlaps, meets, overlappedBy, metBy, begunBy, endedBy: TIME x TIME → Boolean

Observation data type

type Obs [T: TIME, G: GEOMETRY, B: BASE] operations:

new: T x G x B → Obs value: Obs → Bgeom: Obs → Gtime: Obs → T

From observations to events

Why do we need interpolators?

How long do you take from Frankfurt to Beaune?

Why do we need interpolators?

We cannot sample every location at every moment – we need to estimate in space-time

Sensors: water monitoring

Brazilian CerradoWells observation 50 points 50 semimonthly time series(11 Oct 2003 – 06 March2007)

Rodrigo Manzione, Gilberto Câmara, Martin Knotters

MAY JUNE JULY

AUGUST SEPTEMBER

Estimates of water table depth for an area in Brazilian Cerrado

Manzione, Câmara, Knotters

Three types of interpolators

IntValueInTime [T: TIME, B: BASIC]estimate: {Obs} x T → B

IntSpaceInTime [T: TIME, G: GEOM]estimate: {Obs} x T → G

IntInSpaceTime [T: TIME, G: GEOM, B: BASIC]estimate: {Obs} x (T,G) → B

type STgen [T: TIME, G: GEOM, B: BASE] operations:

getObs: ST → {Obs}begins, ends: ST → Tboundary: ST → Gafter, before: ST x T → STduring: ST x Period → ST

What do ST types have in common?

Time Series

Continuous variation of a property value over time(water table depth sensors)

Time Series

Type TimeSeries [T: TIME, B: BASE] uses ST

operations: new: {Obs [T,S,B]} x

IntValueInTime [T,B] → TimeSeries

value: TimeSeries x T → B

Moving objects

MOVING OBJECTS Objects whose position and extent change continuously

Moving objects

individual entity that varies its location (and its extent) over time

Moving Object data type

type MovingObject [T: TIME, G: GEOM] uses ST

operations:

new: {Obs [T,G,B]} x IntSpaceInTime [T,G] → MovingObject

value: MovingObject x T → G

Moving Object data type

distance: MovingObject x MovingObject → TimeSeriesdistance (mo1, mo2) {

ObsSet oset for t = mo1.begin(); t <= mo1.end(); t.next() Point p1 = mo1.value (t)

Point p2 = mo2.value (t)

o1 = new Obs (t, dist (p1, p2))

oset.add (o1)

ts = new TimeSeries (oset)return ts}

How many walruses reached Baffin island?

source: USGS

Coverage: T → G → B

Multi-temporal collection of values in space.

Two-dimensional grids whose values change

Samples from fixed or moving geosensors.

source: USGS

type Coverage [T: TIME, G1: GEOM, G2: GEOM, B: BASE] uses ST

operations: new: {Obs [T, G1, B} x IntInSpaceTime[T, G1, B] x G2 → Coverage value: Coverage x G1 x T → B

Functions on coverages

getWaterArea (Coverage cov, Time t)area = 0forall g inside cov.boundary() if cov.value (g,t) == "water” area = area + greturn area}

From a coverage to a time series

From a coverage to a time series

timeSeries: Coverage x S → TimeSeriestimeSeries (c1, loc)

ObsSet oset for t = c1.begin(); t <= c1.end(); t.next()

Real v = c1.value (loc, t)

o1 = new Obs (t, loc, val)

oset.add ( o1 )

ts = new TimeSeries ( oset )return ts}

When did the large flood occur in Angra? When precipitation was > 10mm/hour for 5 hours

Coverage set (hourly precipitation grid)

Event (precipitation > 10 mm/hour for 5 hrs)

The event data type

An event is an individual episode with a beginning and end, which define its character as a whole.

An event does not exist by itself. Its occurrence is defined as a particular condition of one spatiotemporal type.

The event data type

Type Event [T1: TIME, T2: TIME] uses ST

operations: new: {ST x (T1, T2) → Event compose: Event x Event → Eventintersect: Event x Event → Event

Exploração intensiva

Floresta

Perda >90% do dossel

Corte raso

Perda >50% do dossel

time

Event 1

Event composition Forest loss > 20%

Floresta

Loss > 90%

Clear cut

Loss > 50%

Event 2

Event 3

Event 4

When did the large flood occur in Angra?

When did the large flood occur in Angra?

Coverage prec = getData (weather forecast)flood = new Event()from t0 = prec.begin(); t0 <= prec.end(); t.next() if getRain (prec, t0, t0 + 24) > 100 strong = new Event (prec, t0, t0 + 24) flood.compose (strong)

When did the Aral Sea shrank to 10% of its original size?

getWaterArea (Coverage cov, Time t)area = 0forall g inside cov.boundary() if cov.value (g,t) == "water” area = area + greturn area}

When did the Aral Sea shrank to 10% of its original size?

aralSea = new Coverage (images)

findDisaster (aralSea) {t0 = aralSea.begin()areaOrig = getWaterArea (aralSea,t0) for t = aralSea.begin(); t <= aralSea.end(); t.next() if getWaterArea (aralSea,t) < 0.1* areaOrig disaster = new Event (aralSea, t, t.aralSea.end()) breakreturn disaster}

From observations to events

TerraLib: spatio-temporal database as a basis for innovation

Visualization (TerraView)

Spatio-temporalDatabase (TerraLib)

Modelling (TerraME)

Data Mining(GeoDMA)Statistics (aRT)

GIS technology for big data

Algebras for spatio-temporal data are a powerful way of representing change