Overview of Spatial Modeling

Report
Overview
• What is Spatial Modeling?
• Why do we care?
What’s the problem?
• The issues we need to solve are:
– Getting larger spatially
– Involving more complex data
– Involving more data
– Require special algorithms
– Require meeting the needs of, and
communicating with, much larger groups of
people
• These issues cannot be solved with
traditional GIS analysis
What’s the solution?
• ArcGIS has limited ability to:
– Manage complex datasets
– Process large datasets
– Create custom models
– Run batch processes
• Have to use ArcGIS appropriately, find
other solutions to tough problems
–R
– BlueSpray
– Others…
Marine Spatial Planning
•
•
•
•
Over 100 raster layers
Millions of model runs
Years of work by teams of people
Multiple modeling packages
– Maxent
– Marxan
– ArcGIS
STAC
Spatial Data Can be Big!
• MODIS:
– Entire earth at 250 meters resolution twice a
day
• Landsat:
– Entire earth at 15/30 meters twice a month
for 26 years
• DayMet: Daily Climate Predictions
• LiDAR point clouds
Breaking it down
• “Type” of spatial data:
– Points
– Polygon
– Polyline
– Rasters
• Attributes/Measures:
– Continuous, categorical measures
– Dates
– Descriptive text
• Remotely sensed vs. Field data
Putting it Together
• Almost all spatial data has:
– Measures: occurrences, height, etc.
– Spatial coordinates
– Temporal information
• Can also have:
– 2D, 3D, “4D”, or N-dimensions
– Relational and/or hierarchical structure
How the data is stored
•
•
•
•
•
Large files (to be avoided)
Large sets of files
Relational databases
Distributed networks
Hierarchical storage
Spatial Modeling
• Spatial Model:
– Abstraction of something spatial
– Typically on, or near, the earth’s surface
• Spatial Processing:
– Converting spatial data for a specific use
• Spatial Analysis:
– Analysis that uses spatial data
• Spatial Simulation:
– Models something that has or could occur
spatially and temporally
Goals of Modeling
• Verifiable against the real world
• Robust; repeatable and insensitive to
parameter variance
• Methods are transparent to modelers
and stake holders
• Simple to understand
• Applicable to a real-world situation
• Real world is within uncertainty bounds
of the prediction
General Modeling Methods
• Density:
– Points (occurrences) -> Density surface
• Interpolation:
– Points with measured values -> Continuous
Surface
• Correlation:
– Points with measured values & continuous
covariant -> Continuous surface
• Simulations:
– Very general
• Others…
Density
• Find a density, abundance,
concentration, of discrete occurrences
• Examples:
– Plants and animals
– Disease
– Crime
en.academic.ru
Density Methods
• Minimum Convex Polygon
• Kernel Density Estimates (KDE)
Wikipedia
Interpolation
• Creates a raster with values for each
pixel based on the proximity of sample
points
• Examples:
– Climate layers from weather stations
– Biomass from tree diameters (DBH)
– Soil maps from pits
– DEMs from points
• Must have:
– Autocorrelation
Interpolation: Methods
•
•
•
•
•
•
•
•
•
•
Kriging
Nearest-Neighbor
Bilinear
Spline
Bezier Surface
Natural Neighbor
Delaunay Triangulation
Inverse Distance Weighting (IDW)
Kernel Smoothing
Others…
Correlation
• Variable being predicted is dependent on
other variables (N-dimensional space)
• Examples:
– Habitat Suitability / Species Distributions
– Fire potential
– Land use change
– Disease risk
– Productivity
Correlation or Dependence
•
•
•
•
•
•
•
•
•
•
•
Systems of differential equations
Common Statistical Functions
Kernel functions
Bayesian Inference
Regression
Index Models
Trees
Neural Nets
“Graphical” techniques
Machine Learning Methods
Combinations of the above
Non-Linear Correlation
Several sets of (x, y) points, with the Pearson correlation coefficient
of x and y for each set.
Wikipedia
Simulations
• Use computer software to create a
“simulation” for a general phenomenon
• Examples:
– Climate simulations
– Population models
– Disaster scenarios
– Fire models
– Shipping
Simulations
• Cellular automaton
• Agent-Based
Typical Spatial Models
•
•
•
•
•
•
•
•
•
Flood Planes
Potential Habitat/Species Distribution
Soil Erosion
Ice Extents
Climate Models
Oil Spill Extents
Bark Beetle Infestation
Geologic Layers
Flight Control Software
Atmospheric humidity on June 17, 1993, NASA
Model Characteristics
•
•
•
•
•
•
•
Stochastic or Deterministic
Transparent or “Black Box”
Simple or Complex
Rigorous or Lax
Applied or Theoretical
Internal or “External” Evaluation
Parametric or Non-Parametric
Software
• Correlation
–
–
–
–
–
–
–
–
–
–
–
ArcGIS
R (GLM, GAM…)
Maxent
HyperNiche (NPMR)
BlueSpray
ENVI/IDL
Marxan
WinBugs (Bayes)
BioClim
GARP
Open Modeler
• Interpolation
– ArcGIS
–R
• Simulations
–
–
–
–
Simple: ArcGIS
OpenSource?
Logo?
NetLogo?
• Build your own!
– Java
– C++
– Python!
More Detailed Process
•
•
•
•
•
•
•
•
Define the problem
Investigate the topic
Gather, process, and analyze the data
Investigate and select methods
Find, evaluate, and select the software
Build, parameterize, and run the models
Evaluate the model and results
Along the way, document:
– Assumptions
– Uncertainties
– Problems others have seen
Occam’s Razor
• “other things being equal, a simpler
explanation is better than a more
complex one”
Additional Slides
Others
•
•
•
•
Spatial Networks
Finite Element Analysis
Hydrology Simulations
Disaster Simulations
What is…
•
•
•
•
•
•
A shapefile of zip code regions?
A text file of points of bird observations?
The PRISM Data?
GoogleEarth?
“Final Lab” from 270?
World of Warcraft?

similar documents