Habitat and Connectivity Modeler Toolboxes for ArcGIS

1
Habitat and Connectivity Modeler Toolboxes for
ArcGIS
Nicholas School Geospatial Analysis Program

• Benjamin D. Best, Dean L. Urban, Patrick N.
  Halpin, Song S. Qian

download www.env.duke.edu/geospatial
Habitat Modeler
Introduction
Connectivity Modeler
Regression Techniques
Goals
pts_obs
tin
ln_edgeslc
Statistical
Random
pts_rand
dir_plots
rstr_viable
Plots
Points
pt_nodes

• What
• Model habitat with multivariate regression
• Model connectivity with graph theory
• How
• ESRI ModelBuilder scientific workflow
• Interface to R statistics application
• Utilize Python NetworkX module

poly_patches
Sample to
tbl_env
Network
Table
Create
Least Cost
poly_patchsm
ln_edges
Multivariate
rstr_landcov
Network
ltgt Lakes
Path
rstr_glm
Regression,
GLM
rstr_dem
txt_network
rstr_cost
txt_networklc
Graphical summaries of predictors and response
for the three most common habitat modeling
techniques. For the figure source and general
review of these techniques, see Guisan and
Zimmermann, 2000. Predictive habitat
distribution models in ecology. Ecological
Modelling135.
rstr_glmroc
rstr_aspect
pt_centroids
Network
rstr_tci
Centrality
Example Data
Metrics
poly_patchesn
Prep Patches and Cost Surface
Generalized Linear Model (GLM)
aspect
demdigitalelevationmodel

• form Logit(y) ß0 ß1x1 ß2x2 ßmxm
• prediction (Inv.Logit(response)) 0,1 y 1 /
  (1 exp( -S ßx))
• example Logit(presence) ß0 ß1dem ß2tci
• categorical predictors, e.g. landcover, handled
  as factor
• stepAIC used for model selection of best
  predictors

landcover
tcitopographic convergence index
Binary habitat from the Habitat Modeler can be
converted to distinct patches with the
RegionGroup ArcGIS function. A cumulative cost
surface, using the CostDistance ArcGIS function,
provides the second required input for creating a
network.
Example environmental data from Grandfather
Mountain, NC. The topographic convergence index
(TCI) is derived from a DEM and describes local
wetness conditions. The landcover data is
categorical. These are just a sample of multiple
environmental rasters that could be used for
habitat modeling.
Classification and Regression Tree (CART)
Create Network with TIN
Random Points
ArcMap
ArcScene
ESRI ArcCatalog listing of script tools found and
example models on left of the two toolboxes
Connectivity Modeler and Habitat Modeler. To
the right is ESRI Model Builder which allows for
drag-and-drop addition of data inputs (circular)
for wiring to processes (square).
For each script tool in Model Builder, a form
allows for argument inputs (like top left). This
tool Random Points uses the R library spatstat,
which is capable of many more point patterns than
uniform random. Candidate points can be
restricted to an optional grid mask, read with
Rgdal library. To the bottom left, observed
points (green) are compared to the generated
random points (red), draped over the DEM layer.
Requirements

• Commercial
• ArcGIS 9.0
• ArcInfo
• Spatial Analyst
• 3D Analyst
• Free/Open-Source
• Download these toolboxes www.env.duke.edu/geospat
  ial
• Python 2.3.5 (www.python.org)
• modules NetworkX (networkx.lanl.gov)
• R 2.0.1 (www.r-project.org)
• libraries mass, rpart, mgcv, maptools, foreign,
  Rgdal, spatstat
• R COM connector

Generalized Additive Model (GAM)
Here we present a novel method for the creation
of a graph-theoretic network by the creation of a
triangulated irregular network (TIN) from the
cumulative cost surface and polygon of patches.
From the TIN, we can extract the patch centroid
nodes, edges, exterior nodes, patch polygons and
create a network text file (weighted edge
adjacency list).
Sample to Table
Least-Cost Paths
This minor tool merges the sampling of
environmental rasters by observed / presence / 1
and random / absence / 0 points into a single
table.
A rich suite of graph-theoretic algorithms is
available. In particular, the Djikstra algorithm
for calculating all patch-to-patch least-cost
paths is computationally extremely efficient over
the ArcGIS CostPath function which must be run
for each pairwise combination of patches.
Python the programming glue
Statistical Plots
The easy robust summary statistics and diagnostic
plots provided by R are exploited with this tool
to generate density histograms comparing the
presence to random values per variable, and a
pair-wise plot with histograms, scatterplots and
trend lines.
Binary Habitat with ROC Curves
Network Metrics

• Density Histograms

• Pairs Plot

Closeness
Betweenness
Degree
Acknowledgements
Individuals Scott Loarie, Ben Poulter Funding
If the optional binary map path is provided, one
is produced based on the binary threshold. If no
threshold is specified, then the optimum is used
based on the Receiver Operating Characteristic
(ROC) curve, which represents the models
prediction performance.
Graph-theoretic centrality metrics, per Brandes
(2000. Faster Evaluation of Shortest-Path Based
Centrality Indices. CiteSeer).