Data synergy and spatial modeling of leafy spurge

1
Data synergy and spatial modeling of leafy spurge

• Catherine S. Crosier
• Thomas J. Stohlgren

2
Why Data Synergy

• Species lists lacking or incomplete
• What is it?
• Where is it?
• Ecological studies small extent, short term
• Make the most of existing data

Colorado
Vegetation Survey Data
Species Lists
GIS Data
3
Problem

• Spread since introduction to Massachusetts in
  1827
• Economic impacts in the Northern Great Plains
• Noxious weed list in 35 states
• Reduces native species biomass and richness
• Decreases biomass of key forage species

4
Competitive Ability

• Grazing avoidance / tolerance mechanisms
• Root system
• Reproductive capability and growth

5
Objectives

• Illustrate benefits of data synergy
• Predict individual species occurrence over a
  broad spatial extent maintaining a fine spatial
  resolution by using multiple datasets

6
(No Transcript)
7
Vegetation Plot Data

• Researchers
• Field research
• Clumpy distribution from different study
  locations
• All species

8
Weed Mapping Data

• Weed managers
• GPS data
• Specific areas of interest
• Specific noxious weeds

9
Quarter Quad Data

• State of Colorado
• Survey of County Weed Managers
• Covers all quarter quads
• Only specific noxious weeds

10
Datasets

• Absence data
• Approximates 30m x 30m
• 1504 plots
• Presence data
• Any data type lt30m X 30m
• 1180 locations

11
Datasets

• Environmental variables
• Euclidean distance to water
• Average annual precipitation
• Human factors
• Population density per zip code
• Euclidean distance to roads
• Remotely sensed imagery
• Landsat TM derived products
• Normalized Difference Vegetation Index
• Tasseled cap 1 4

• Topographic
• Elevation
• Slope
• Absolute aspect
• Soil fractions
• Percent sand
• Percent silt
• Percent clay

12
(No Transcript)
13
(No Transcript)
14
(No Transcript)
15
Statistics

• Binary data gt GLM with stepwise selection
• Capture large scale trend
• Spatially autocorrelated residuals
• (Morans I 0.53, plt0.001)
• Variogram
• Kriging
• Capture fine scale variability
• Mimics dispersal

16
Model Evaluation

• Percent deviance explained
• Receiver Operating Characteristics (ROC)
• Independent of threshold
• True positives fraction/ false negatives fraction
• Mann-Whitney U-test (Wilcoxon rank sum)
• Preliminary truth data quarter quad survey
• E. esula values zonal statistics

Chance
Sensitivity
Model Prediction
1-Specificity
17
Coefficients D20.67
18
(No Transcript)
19
(No Transcript)
20
Mann-Whitney results
21
Conclusions

• Deviance explained is high
• Quarter quad data may not be good evaluation
  dataset
• Hyperspectral imagery might improve model
• Modeling technique has promise-
• Requires data synergy

22
Acknowledgements

• Computer assistance Jim Graham, Greg Newman and
  Rick Shory, and partners at NASA Goddard Space
  Flight Center
• Data partners George Beck, Biota of North
  America Program (John Kartez), Bureau of Land
  Management, Colorado Natural Heritage Program,
  Colorado State Parks, Susie Geer, Sarah Goslee,
  Tom Hobbs, Merrill Kaufman, Larimer and Otero
  County Weed Managers, Dan Manier, Amanda Myers,
  Niwot Ridge and Shortgrass Steppe Long Term
  Ecological Research Stations, Eric Petterson, San
  Luis Valley GIS Authority, State of Colorado Weed
  Mapping Program, Jason Sutton, The Nature
  Conservancy, U.S. Fish and Wildlife Service, U.S.
  Forest Service FSVeg, and the U.S. Geological
  Survey.
• Funding USGS and Agriculture Experiment Station