Database Reference
In-Depth Information
mining approaches that are able to monitor features in continuous data
fields, especially with the continual expansion of spatio-temporal climate
datasets.
5. Relationship Mining
Predicting changes in the earth's system requires a comprehensive un-
derstanding of the complex feedback interactions among its underlying
processes. Relationship mining aims at discovering information about
the interactions between sensor attributes to provide an insight into the
structure of the underlying phenomena. Instead of discovering anoma-
lous or change events in all or some of the multiple variables, the primary
objective here is to get a better understanding of the relationships be-
tween the variables in order to determine the structural properties of
multivariate data. As an example, increased sea surface temperatures
are known to affect the amount of precipitation received during the wet
season in South America, eventually making the vegetation more suscep-
tible to fire [17, 25]. In a recent work by Chen et al. [17], relationships
between Sea Surface Temperature (SST) Anomalies in the Atlantic and
Pacific Ocean, and Fire Season Severity (FSS) in South America were
studied.
Through economic and policy actions, human behavior also often en-
ters into these feedback structures. For example, deforestation for palm-
oil plantation in peat-land regions of Indonesia is often followed by fires,
in close spatial and temporal proximity [60]; such patterns are attributed
to complex physical processes whereby deforestation leads to soil degra-
dation, which reduces the moisture in the soil, making the peat reserves
underneath more susceptible to fire leading to massive emissions of car-
bon [48]. As another example, increase in the average night-time win-
ter temperature attributed to global warming in the higher latitudes of
North America has led to an increase in pine beetle infestation in those
regions. However attributing the effect of these infestations on forest
fire frequency is under dispute as there have been studies supporting the
effect [32] and otherwise [66] (see
Figure 15.6
)
.
As another example, relationships between events occurring at multi-
ple locations can also be expressed as spatio-temporal patterns, evolving
in space and time. Spatio-temporal sequential pattern mining aims at
finding patterns of events that occur in close proximity of space and
time. For example, Huang et al. [35] proposed a model for discovering
sequential chains of events by extending the spatial co-location frame-
work to spatio-temporal databases. Recently, Mohan et al. [53, 54]
proposed a directed acyclic graph based approach to detect sequences of
