An Early Detection and Rapid Assessment Network for Plant Invasions - PowerPoint PPT Presentation

Loading...

PPT – An Early Detection and Rapid Assessment Network for Plant Invasions PowerPoint presentation | free to download - id: 1274ed-NmJhN



Loading


The Adobe Flash plugin is needed to view this content

Get the plugin now

View by Category
About This Presentation
Title:

An Early Detection and Rapid Assessment Network for Plant Invasions

Description:

An Early Detection and Rapid Assessment Network for Plant Invasions – PowerPoint PPT presentation

Number of Views:88
Avg rating:3.0/5.0
Slides: 54
Provided by: navat
Category:

less

Write a Comment
User Comments (0)
Transcript and Presenter's Notes

Title: An Early Detection and Rapid Assessment Network for Plant Invasions


1
An Early Detection and Rapid Assessment Network
for Plant Invasions
2
  • Predictive Modeling of Species Distributions and
    Early Detection
  • A Regional Approach to Early Detection and Rapid
    Response IPANE
  • An Information Network for Early Detection and
    Rapid Response Local to National Levels

3
Predictive Modeling of Species Distributions Appl
ications to Invasives
4
Proteas from South Africa
5
http//protea.worldonline.co.za/default.htm
6
Protea Atlasing
Protea altasing in the field
Field forms
mapping
7
Protea Atlas sample sites across the Cape
Floristic Region 60,000 sites, 250,000
species-site records, globally one of the most
intensively sampled group of species for any
biogeographic region.
8
Cape Floristic (Biogeographic) Region
90,000 km2 9000 plant species 70 found nowhere
else 1400 threatened or
endangered species
9
Objective to develop Bayesian hierarchical
regression models to explain and predict species
distribution patterns
10
  • We use the Bayesian perspective to model
    predictions
  • Bayesian inference answers the question
    How probable are my hypotheses, given my data?
    (i.e. P(Hdata)).
  • Versus the classical, frequentist perspective
    that answers the question
    How probable are my data give my
    (null) hypothesis? (i.e. P(dataH)).

Rev. T. Bayes
Sir R.A. Fisher
11
  • How does the Bayesian perspective work
  • We may have some prior expectations about the
    outcome of an experiment, the parameters of the
    model, or the confidence that an hypothesis is
    correct before any data are examined prior
    probability.
  • These probabilities are modified in light of the
    data available providing posterior probability
    or confidence in the hypothesis.
  • Use Bayes theorem
  • posterior probability that the Ho is correct,
    given the data
  • probability that data would be observed, given
    the Ho X
  • the prior probability that Ho was correct,
    before any data were collected
  • p(Hodata) p(dataHo) p(Ho)

12
Simple conceptual example predict the
distribution of spotted owls in the NW with some
degree of confidence. We know that owls are
rare in the forested landscape prior experience
may indicate that the probability of finding them
in any particular forest is .05 But, by
including data on their ecology can we have
increased confidence in detecting their presence?
For example we may hypothesize that they nest in
snags. We can evaluate this by gathering
observations p/a of owls in forests w/ and w/o
snags. Example forest survey data
13
  • Set up as a Bayesian model
  • Posterior prob. of owls, given snags prob. of
    snags, given owls prior prob. of owls

For above example, the probability of finding
owls in forests given snags
  • Versus usual frequentest perspective
  • There is a high probability that owls are not
    distributed at random among forests (w/ and w/o
    snags) plt0.0001, or that there is a strong
    relationship between no snags and no owls.

14
Presence/absent of species in grid-cells across
the landscape explained by
  • a suite of environmental variables (23)
  • plus species attribute variables (5)
  • spatial effects

15
Kogelberg-Hawequas sub-region Sample locations,
with a 1x1 minute grid overlain (41 x 107 km
1554 grid cells) versus 37,000 cells for full
CFR
16
The hierarchical structure of the model
17
(No Transcript)
18
(No Transcript)
19
(No Transcript)
20
(No Transcript)
21
(No Transcript)
22
Full Model Protea cynaroides
10 significant environmental covariates (Betas)
23
Full Model Sorocephalus imbricatus
0 significant environmental covariates (Betas)
24
Data Thinning Protea cynaroides
25
Data Thinning Aulax cancellata
26
Sister-Species Ghosting
M. arboreus M. argenteus
27
Hakea sericea (silky needlebush), introduced from
Australia projected distribution and
uncertainty in distribution.
28
Species distribution predictions of the full Cape
Floristic Region
29
(No Transcript)
30
(No Transcript)
31
Predicting species distributions for New England
invasives
32
(No Transcript)
33
(No Transcript)
34
  • Explanatory data layers
  • Field collected information
    habitat/community type, canopy closure,
    slope aspect, soil moisture regime
  • Climate variables Mean annual precipitation,
    Mean annual snow fall, Mean annual min max
    temperatures, Annual record extreme min max
    temperature, Mean annual temperature, Mean length
    of frost free period, Mean number of growing
    degree days, Mean annual number of heating
    degree days 2 km grid scale.
  • Topographic variables DEM - average elevation
    for 2km grid cell, DEM
    range, high and low values for 2km
    grid cell.
  • Road influences Distance to Road (for each
    IPANE plot point), Length of Roads within
    each 2km grid cell.
  • Landuse landcover percent land cover of
    each of 9 classes resampled from 30m
    resolution to 2 km.

35
IPANE data are used for predictive modeling.
Here is the predicted spread of Celastrus
orbiculatus (from dark areas to light areas) in
New England.
The uncertainty (variance) is also measured in
the predictive model (darker areas are most
uncertain).
Celastrus orbiculatus Oriental bittersweet
Example of modeled invasive distribution at the
regional scale Oriental bittersweet
36
Local to landscape patterns
37
C. orbiculatus 251 out of 603 - 41.6
38
Is there an effects of land-use or land-use
change on the distribution of invasive species in
the New England landscape?
39
Land-use change in a local landscape
40
Quantifying land-use change in the landscape over
the past 70 years from aerial photography in
Connecticut
2000
1970
1951
1934
41
Procedure
  • Process historical aerial photographs create
    geo-corrected mosaic images of 5 different time
    periods.
  • Digitize land use features - visual
    interpretation of aerial mosaics using a suite of
    LULC categories.
  • Create a stratified random sampling scheme -
    Generate 50 random points for each change
    category.
  • Sampling - based upon the IPANE plot protocols

42
Digitizing Example
43
LULC Change Categories
1. Forest - No Change 2. Cultivated Fields - No
change 3. Pasture/Meadow - No change 4.
Residential/Commercial - No change 5. Abandoned
fields to Forest 6. Cultivated fields to
Forest 7. Pasture/Meadow to Forest 8.
Cultivated Fields to Abandoned to Forest 9.
Pasture/Meadow to Abandoned to Forest 10.
Cultivated fields to Abandoned fields 11.
Pasture/Meadow to Abandoned fields 12. Forest
to Residential/Commercial 13. Cultivated fields
to Residential/Commercial 14. Pasture to
Residential/Commercial 15. Abandoned fields to
Residential/Commercial 16. Abandoned fields to
Forest to Residential/Commercial
No Change
Agricultural Fields Reverted to Forest
Abandoned Fields as of 2003
Conversion to Residential/ Commercial
44
Final Plot Points
603 in Total 507 Random, 96 Opportunistic
45
Presence of One or More Invasive Species
All Plots 363 / 603 or 60.2
Random Plots 288 / 507 or 55.4
46
C. orbiculatus 251 out of 603 - 41.6
Random 186 / 507 36.7
1000 3 100 - 999 53 20 - 99 100
47
B. thunbergii 199 out of 603 - 33.0
Random 140 / 507 27.6
1000 2 100 - 999 31 20 - 99 47
48
Presence/Absence Individual Species by LULC
Groups (From Random Plots N 507)
49
Hierarchy of Invasion ( Abundance of all species
combined)
1st Abandoned Fields as of 2003 (p
.005) 2nd Agricultural Fields Reverted to
Forest (p lt .0001) 3rd Conversion to
Residential / Commercial (p lt .0001) 4th
Categories with No Change
50
Regression models
Invasive species abundance f(environmental
explanatory variables)
Explanatory variables Community type (24
IPANE categories), Habitat class (4), Canopy
closure, Slope aspect, Soil moisture, LULC
(12 or 5), Distance to Road, Road density,
Distance to nearest building, Building density,
Edge Distance (and type), Elevation,
Geology, Soil type and Soil drainage class.
51
Explaining spatial patterns in Oriental
bittersweet abundance using regression models
Community type Aspen/birch, old fields,
forest/field edge, roadsides. - Oak/pine
forests. LULC type conversion to
residental/commercial, abandoned fields,
fields reverting to forest. - agricultural
fields, no change, especially forest. Canopy
closure - moderate to dense canopies Edge
distance - Minor contribution soils, building
distance, prior vegetation cover
52
Explaining spatial patterns in Japanese barberry
abundance using regression models
Canopy closure - for densest canopy
closure Community type upland red maple,
northern hardwoods, birch/aspen, old
fields, open fields - agricultural fields,
coniferous forests, oak/pine forests. LULC
field reverting to forests - agricultural
fields, forests Edge distance -
53
Mid-Spring ADARA Possible Remote Sensing
Search Tool for Invasives?
About PowerShow.com