E. Asian Basins Workshop

1
Linking Dynamic Temporal Processes And Spatial
Domains
East Asian Basins February 2001 Casey McLaughlin
University of Kansas, USA
2
Extracting global effects from data on local
scales

• Field studies are inherently site-specific.
• Can we obtain global understanding from mosaics
  of local processes ?
• Cluster analysis can organize habitats by
  function.

3
Managing Granularity of Models Data

• How much resolution is necessary?
• Averaging to coarser scales is easy, but what
  about sub-pixel characterization?
• Do components scale linearly?

4
Linking dynamic mosaics of local biogeochemical
flux budgets to large scale regional and ....
5
Coastal Development
The Yellow River Example Historic records
provide a context of long-term sediment fluxes
and geomorphic development within which to
evaluate short-term, high resolution remote
sensing images and the acceleration of
anthropogenic effects.
6

• The Typology Approach to Globalization of
  Function
• Develop global database at a scale (30)
  appropriate to the parent data and global models
• Include sub-grid-scale parameterization
  statistics on spatio-temporal variability,
  alternative time slices
• Use similarity analysis to extrapolate function
  measures and to test for effectiveness of proxy
  variables (clustering LoiczView)
• Encourage community collaboration to develop
  local-regional higher resolution analogs,
  extensions, and tests (eg East Asian Basins
  Workshop)

7
The LOICZ domain Grid Cells Coastal (30,
shoreline defined),Terrestrial (1o
inland),Oceanic I (1o seaward, or shelf)
The CoML domainOceanic I, Oceanic II, and
Oceanic III (all the rest)
8
Global Cell Structure
9
0.5 Degree Cells
10
Effective spatial resolution can be enhanced by
inclusion of statistics or summaries from higher
resolution data sets
Coastal cells can be populated with complexity
statistics derived from GIS analysis of digital
shorelines length, tortuosity, number of
islands, land area, etc.
Coastal and oceanic cells contain 2 bathymetry
statistics mean, s.d., range, areas within
selected depth classes, etc. Land cells are
similarly treated based on one-km DEMs
11
Complex Process Models
12
An interactive WWW database link permits
selection of variables by type, by geographic
region, and by cell type for viewing, downloading
and augmentation, clustering and visualization.
13
Geospatial Clustering (LOICZVIEW) is a Tool for

• User-friendly, robust cluster analysis of
  georeferenced data
• Visualization of results, with comparison
  features and GIS-compatibility
• Nested and cross-scale applications (using both
  internal and external dataset characteristics)
• Community building and linking of distributed
  databases
• Developing the power of the internet for
  long-range collaboration on major, spatially
  distributed issues

14
What is this thing called
LoiczView
?
Developed by
B. A. Maxwell
http//www.
palantir
.
swarthmore
.
edu
/
maxwell
/
loicz
/
1. A program for similarity analysis of high
-
dimensionality ( lots of
variables) data sets using k
-
means clustering techniques (conceptual
analog PCA and
dendrogram
techniques).
2. Clusters are determined on the basis of the
data vectors in
n
-
dimensional space.
3. Operator has control of data inputs, cell
classes for analysis
number of clusters, and distance measure.
4. Designed to be robust with
sub-optimal
data sets, scale
-
independent.
5. Has built
-
in
Geo-spatial
and similarity visualization capabilities.
6. Going into final beta
-
test phase.
15
Clustering of means and standard deviations
permits assessment of habitat and variability.
Sea surface temperature, precipitation, and
runoff were clustered into 5 classes using a
k-means clustering algorithm
Cluster of Intra-Annual Std Deviations
Cluster of Annualized Values
Low Precip, Low SST, Low Runoff High Runoff Low
Precip, Med SST, Low Runoff Med Precip, Low SST,
Low Runoff High SST, Low Runoff
High Runoff Med Runoff, High SST Med Precip, Low
Runoff Low SST, Low Runoff Low Precip, Low Runoff
16
Critical aspects of temporal variability
seasonal and interannual can be captured by
climatology statistics
Total annual precipitation (CRU, 1961-1990)
Mean
Std. Dev

• Areas with similar average totals show major
  differences in seasonality.
• Max, Min, Median and Range statistics can be
  similarly used.
• Other statistics can provide interannual
  variability indices.
• The example also illustrates the power of
  latitude as a proxy variable.

Low....HighNo Data
17
Inland effects continent-scale impacts on the
local CZ
Classed river basin flow/cell
Classed runoff/cell
Local effects vs. coastal projection of
continental forcing most of the world CZ is
locally controlled!
18
Expert typology
Calibration of clustering by expert judgment
Alternative 2
Alternative 1
19
Simplification and Aggregation Across Spatial
Domains

• Can we achieve reliable predictions for variables
  of interest?
• Can these simplified relations be generalized or
  are they site/domain specific?

20
Balancing Objectives
Scientific Enlightenment and Predictive
Accuracy Ie. Identify proxies for comparisons
Acknowledgements Apologies
21
Contributers
University of Kansas, Lawrence, KS Casey J.
McLaughlin (CJM_at_UKANS.EDU) Dr. Robert
Buddemeier Jeremy Bartley
Moss Landing Laboratories, Monterey Bay. CA Dr.
Richard Zimmerman.