Transcript IALE 2003 presentation, Darwin, AU

Linking FIA Data and Satellite Imagery to Build
a Habitat model for the Marbled Murrelet
Martin G. Raphael
Pacific Northwest Research Station
Funding contributed by:
Region 6, USFS, PNW Research Station
USDI Fish and Wildlife Service
Northwest Forest Plan of 1994
• Conservation plan for older forests and
•
Physiographic
provinces
(57 mill. ac.,
46 mill. ac
forest)
•
species on 57 mill. ac. of federal land
Effectiveness Monitoring modules for older
forest, n. spotted owl, marbled murrelet,
watershed condition
Key questions for monitoring older forest:
– How much, how is it changing, how
might it change in the future?
– Is the Plan providing for its conservation
and management?
USA
A primary objective of the
Northwest Forest Plan was to
achieve:
“maintenance and/or restoration of habitat
conditions for the Northern Spotted Owl and
the Marbled Murrelet that will provide for
viability of each species -- for the owl, well
distributed along its current range on federal
lands, and for the murrelet so far as nesting
habitat is concerned”
--FEMAT 1993:iv
Objectives
• Estimate amount and distribution of murrelet
nesting habitat in WA, OR, CA
• Estimate change over time – from start of
plan to now
• Make estimates over all lands within the
murrelet range in WA, OR, CA
• Use existing sources for environmental
variables (e.g., IMAP, PRISM)
Needs for regional vegetation information
• Methods that integrate plot and remotely sensed data to provide info.:
– Consistent over large, multi-ownership regions (“all lands”)
– Spatially explicit (mapped)
– Detailed attributes of forest composition and structure
•
– Support integrated landscape analyses of multiple forest values
Latest challenge: provide trend information that is spatial
– Monitoring older forest for Northwest Forest Plan
Effectiveness Monitoring for Late-Successional
and Old-Growth Forest (LSOG)
•
Objective: develop tools and data to assess change in older forest
– Gradient nearest neighbor (GNN) imputation (maps of detailed forest
attributes)
•
– Change detection from Landsat time series (LandTrendr) (trends)
Approach: minimize sources of error in models, map real change
– Corroborate with sample-based estimates
•
Monitoring report every 5 years
– 10-year report (Moeur et al. 2005)
– In progress: 15-year report
– 1996 to 2006 (Wash. and Oreg.), 1994 to 2007 (Calif.)
* Moeur, M., et al. 2005. Northwest Forest Plan–The first 10 years (1994-2003): status
and trend of late-successional and old-growth forest. Gen. Tech. Rep. PNW-GTR-646.
Gradient Nearest Neighbor Imputation (GNN)
k=1
Accuracy assessment (‘obsessive transparency’)
•
Local- (plot-) scale accuracy via cross-validation:
– Confusion matrices, kappa statistics, root
mean square errors, scatterplots, etc.
•
Landscape- to regional-scale accuracy:
1
5
2
3
4
6
7*
8
10
11
12
9
local
(1-ha
plot) scale
13
– Area distributions in map vs. plot sample
– Range of variation in map vs. plot sample
– Riemann et al. (2010) diagnostics
landscape- or
watershedscale
– Bootstrap variance estimators for kNN
(Magnussen et al. 2010)
•
Spatial depictions of uncertainty:
– Variation among k nearest neighbors
– Distance to nearest neighbor(s)
(sampling sufficiency)
•
‘Look-and-feel’ issues
Oregon
regional
scale
LSOG change from GNN
‘bookend’ maps,
1994/6 to 2006/7
LSOG
change
(% of
forest)
•
GNN models and change at 30-m
pixel scale
•
Recommend summarizing to
coarser scales
•
Example: 10-km hexagons
Change in habitat suitability
NWFP Effectiveness Monitoring
Northern
spotted owl
Marbled murrelet
•
Maxent (machine learning) models based
on forest structure and composition
attributes from GNN, trained with nest
location data
•
Subtract models to get change
Modeling Marbled Murrelet Nesting Habitat:
Estimating Nesting Habitat Suitability
Natural History
 Fish-eating
seabird
Distributed
along West coast
S to Monterey Bay
 Nests on limbs
of big conifers
 Nests within 20
to 50 km from
shore
Model Form
• Presence/available
• Presence = set of murrelet nests plus equal
number of “occupied” sites
• Available = entire landscape within study
region that is “capable” of being habitat
– We masked out barren lands, non-forested
areas
Variable selection
• Team developed initial list from available data
based on experience, literature
• Variables must cover range
– Used GNN for forest attributes
– PRISM for climate variables
– DEM for slope, aspect
• Ran correlations, dropped one if r > 0.9
– Kept variable with better support in
literature
Murrelet Sites
Model
area
WA
Nests
Occupied
54
54
OR
65
65
CA
52
52
Total
171
171
Examples of GNN data
Platforms per Tree (all species)
We used these data to derive a new variable
from the GNN data
Higher-suitability habitat (thousands of acres)
1600
1400
Baseline estimate
Fed. Reserved
1200
Fed. Nonreserved
1000
800
Nonfederal
600
400
200
0
Washington
Oregon
California
Higher suitability habitat (millions
of acres)
Change in habitat from 1994/96 to 2006/07
Baseline
(1994/96)
3.0
Maxent
bookend
2.5
LandTrendr
2.0
1.5
1.0
0.5
0.0
Washington
Oregon
California
Loss of suitable habitat from
baseline to 2006/07
Baseline
Fire
Harvest
Other
Total
Percent
- Thousands of acres Fed.
2,163.1
reserved
51.6
8.8
3.7
64.2
3.0
Fed.
262.7
Nonreserved
5.3
6.5
0.8
12.6
4.8
Non-fed
0.9
394.3
18.7
413.9
29.8
1,386.6
Murrelet population size
in relation to amount of
nesting habitat
Murrelet numbers are declining
Is amount and trend of nesting
habitat a primary driver of population
trend?
• Spatial distribution of murrelets is well-
predicted by spatial distribution of habitat
• Habitat trend is not as clear, but suggest a
possible correlation
• If marine conditions are the driver, we’d
expect similar trends among other related
birds and we don’t see such trends
Sources of uncertainty in overall monitoring results
•
Multiple estimates, lots of moving parts with different limitations
– Map- and plot-based estimates can’t be compared statistically
– Look for corroboration
– Complexity and uncertainty pose challenges for users
•
Error in model-based estimates
– Error in plots, spatial predictors; model specification; etc.
– Limitation of Landsat for mapping LSOG recruitment
– Time period is short (10-13 years), and data will improve
•
Uncertainty associated with murrelet habitat definition:
– Habitat attributes can be affected by one or a few trees
– Disturbance can create habitat gain, habitat loss, or no change
Products from NWFP monitoring study
•
GNN models and diagnostics available for download
– 2006/7 and 1994/96 vegetation maps and accuracy assessments
•
15-year reports (PNW GTRs) published or in press:
– LSOG, northern spotted owl, marbled murrelet, watershed condition