Database Reference
In-Depth Information
3. Data-centric Challenges
Earth science datasets pose several unique challenges: some are due
to their inherent spatial-temporal nature and others are specific to the
domain. The following paragraphs throw some light on the key charac-
teristics and challenges of sensor datasets used in earth science research.
Spatial-temporal data.
Many earth science datasets have been
interpolated to gridded time series for ecosystem and environmental vari-
ables, i.e., each series represents an individual co-registered cell in a
latitude-longitude grid that covers the entire surface (or a region) of the
Earth.
Uncertainty and incompleteness.
Earth science datasets are
frequently plagued with noise/uncertainty and incompleteness due to
sensor interference and instrument malfunctions. This issue is partic-
ularly acute in the case of remotely sensed land surface data, where
atmospheric (clouds and other aerosols) and surface (snow and ice) in-
terference are constantly encountered. This motivates the need for de-
velopment of algorithms that are robust to presence of uncertainty and
incompleteness in data.
Temporal variability.
Ecosystem observations tend to have a
high degree of temporal variation. For example, vegetation data such as
greenness usually changes naturally on multi-year scale, but infrequent
and local events such as forest fires and logging can induce short-time
events in naturally occurring spatio-temporal processes. These events
need to be distinguished from other more regularly occurring events
such as the seasonal cycle and recurring rain seasons. Handling such
naturally occurring temporal variations is necessary to avoid detection
of spurious patterns.
Spatial heterogeneity.
This refers to variability of the observed
processes over space and is illustrated by natural boundaries of wild-
fires or deforestation due to topographical constraints, growth of cities
along a spatial gradient, or preferential land conversion for agricultural
intensification near resources such as lakes and cities. Further, data het-
erogeneity drives the need for developing local or regional models, each
corresponding to a homogeneous group of locations, into the data mining
framework.
