Modeling and mapping vegetation composition and structure at broad spatial scales Janet L. Ohmann, Research Forest Ecologist Ecosystem Processes Program, PNW Station USDA Forest Service, Corvallis, OR [e-mail]

1
Modeling and mapping vegetation composition and
structure at broad spatial scalesJanet L.
Ohmann, Research Forest Ecologist Ecosystem
Processes Program, PNW StationUSDA Forest
Service, Corvallis, ORjohmann_at_fs.fed.us

• Vegetation mapping linking regional plot and
  spatial data
• Current vegetation biodiversity in coastal Oregon
  (CLAMS)
• Regional patterns of forest fuels in three
  ecoregions (GNNfire)
• Regional ecology of dead wood (DecAID)

2
Predictive Vegetation Mapping in Coastal Oregon
Funded by PNW (CLAMS, NWFP, WCI, FIA)

• Objectives
• Develop analytical tools to integrate field plot,
  remotely sensed, and mapped environmental data to
  map forest structure and composition
• Quantify spectral, environmental, and disturbance
  factors associated with regional gradients of
  tree species and structure

3
CLAMS needs for a vegetation map

• Initial conditions (1996) for landscape
  simulations
• Response models for wildlife, aquatic, timber
• Big picture vegetation conditions, biodiversity
• Vegetation data that are spatially complete,
  regional in scope, rich in detail
• tree list for each pixel
• species and structures
• fine-scale pattern

Coastal Landscape Analysis and Modeling Study
4
Plot configurations
CVS
(n823)
FIA
Old Growth
(112 963 pixels)
Vegetation data Basal area by species and
diameter class
5
Environmental and Disturbance Gradients in
Coastal Oregon
6
Overview Gradient Nearest Neighbor Method
Tree list for each pixel
Remote sensing
Climate
Geology
Imputation
Topography
Ownership
Direct gradient analysis
Plot locations
Plot data
Spatial prediction
Spatial data (GIS)
Statistical model
7
Factors Associated with Vegetation Gradients
(canonical correspondence analysis)
Variable Subset Relative contribution to explained variation
Ownership 2.2
Topography 4.5
Geology 1.8
Climate 8.0
Landsat TM 15.2
Location 5.2

• Forest structure associated with Landsat
• Species gradients associated with climate
• All variables improved predictions over Landsat
  alone

8
Dominant Gradients in Coastal Oregon
Frequent summer fog, maritime climate
Older stands, large trees, public lands
Younger stands, small trees, private lands
Summer moisture stress, less maritime
CCA axis 2 species composition
CCA axis 1 stand structure
9
(No Transcript)
10
GNN vegetation map ...somewhere SW of Eugene, 1996
11
How good are the GNN vegetation maps?

• Evaluated maps in several ways, primarily
  cross-validation
• Excellent at regional level range of
  variability, landscape proportions
• Reasonable portrayal of fine-scale heterogeneity
• Site-level accuracy similar or better than other
  Landsat maps
• Appropriate for planning and policy analysis at
  broad scales, not local management decisions.

Basal area (m2/ha)
Broadleaf proportion
Quadratic mean diam. (cm)
12
Advantages of GNN maps for ecological analysis,
simulation modeling, integrated assessment

• Rich in detail 25-m resolution, tree list for
  each pixel.
• Analysis flexibility wide range of continuous
  vegetation variables and classifications can be
  derived and mapped.
• Ecologically based (use of environmental data)
• For more information
• www.fsl.orst.edu/clams/gnn
• Ohmann, J.L. Gregory, M.J. 2002. Predictive
  mapping of forest composition and structure with
  direct gradient analysis and nearest neighbor
  imputation in coastal Oregon, USA. Canadian
  Journal of Forest Research 32725-741.

13
Current Vegetation Biodiversity in Coastal Oregon

• New regional assessment opportunities using GNN
  maps
• Coarse filter but rich in detail
• plant community types
• successional stages
• forest structure elements
• individual (common) species

Plant communities
Structural elements
Individual species
14
Biodiversity Along a Continuum of Management
Emphasis
Mid-Coast Region
Ecological goals predominate
Timber goals predominate
15
Species Composition Linked to Environmental
Gradients
Pacific silver fir / noble fir
Sitka spruce
Maritime
Low
High
Valley
Foothill oak woodlands
Western hemlock
Climate
Elevation
Potential Vegetation Types
Dry w.hemlock / mixed evergreen
16
Vegetation Types and Management Objectives

• About 1/3 of each forest type managed for
  ecological goals EXCEPT...
• Foothill oak woodlands 94 on private lands, few
  reserves, threatened by nonforest development.

Timber goals predominate
Ecological goals predominate
17
Forest Age and Structure

• Associated with management history, land ownership

Old forests, closed canopies, public lands
Young forests, open canopies, private lands
Young
Old
18
Very young (0-25 cm, lt70 cover)
Young to middle-aged (25-50 cm, gt70 cover)
Mature (gt50 cm, not old growth)
Old growth (OGHI gt75)
of basin
of basin
of basin
of basin
52 of forest 66 pvt. 37 eco-goals
29 of forest 80 pvt. 24 eco-goals
17 of forest 70 public 72 eco-goals
2 of forest 78 public 79 eco-goals
19
Legacy Trees
Lack of legacies under intensive management
Forest management w/ legacies
Natural legacies after wildfire
20
Legacy TreesLarge Dead Wood

• Snags strongly affected by forest management
  most abundant in older forests on public lands,
  diminished in young managed forest
• Down wood associated with site productivity,
  long-term history, more evenly distributed across
  forest ages (longevity)
• Tillamook Burn legacy

m3/ha gt50 cm
21
Legacy Trees Broadleaf Trees

• Coastal, riparian, foothill, disturbed habitats
• Reduced by intensive forest management favoring
  conifers
• Most abundant on nonindustrial private lands

broadleaf basal area
22
Key Findings Vegetation Biodiversity in Coastal
Oregon

• In semi-natural forested landscapes, all
  ownerships contribute to biodiversity.
• Some biodiversity elements (tree species and
  communities) persist with disturbance, but
  conservation must consider regional environmental
  gradients.
• Vegetation types are unevenly protected (oak
  woodlands).
• Older forests small part of landscape but being
  addressed. Diverse young forests also rare but
  receiving less attention. Legacy tree habitat
  uncertain future.
• New ways of assessing biodiversity at regional
  scale are possible.

23
Mapping Regional Vegetation and Fuels
(GNNfire)co-PIs Mike Wimberly (UGa), Jeremy
Fried (PNW FIA)Funded by Joint Fire Sciences
Program

• Objective map vegetation and fuels using
  Gradient Nearest Neighbor method (GNN)
• Research questions
• How well can we spatially model fuel components
  with GNN?
• How well does GNN work in different ecosystems?
• Applications
• Assessing fire risk
• Planning fuel management
• Modeling fire behavior and effects

Temperate steppe ecoregion
Marine ecosystem
Mediterranean ecosystem
24
Fuel variables (examples) modeled with GNN
25
Linkages between GNN vegetation maps and
vegetation dynamics models, fuel classification
programs, fire behavior models, and fire effects
models
Satellite
Satellite
Imagery
Imagery
Inventory
Gradient Nearest
-
Inventory
Gradient Nearest
-
Plots
Neighbor Method
Plots
Neighbor Method
GIS
GIS
Data
Data
Landscape
Landscape
Vegetation Map
Vegetation Map
Fire Behavior
Fire Behavior
Models (FARSITE)
Models (FARSITE)
Vegetation Simulators
Fuel maps
Vegetation Simulators
FCC
Fuel
FCC System
(FVS
-
FFE)
(FVS
-
FFE)
System
Maps
Fire Effects Models
Fire Effects Models
(FOFEM, CONSUME)
(FOFEM, CONSUME)
Predicted Future
Predicted Future
Landscapes
Landscapes
FCC Fuel Characteristic Classification
26
Regional Ecology of Dead Wood in Oregon and
Washington

• Issues wildlife habitat and biodiversity,
  productivity and sustainability, fuels and fire
  risk, carbon stores
• Studies
• Understanding regional variation and
  environmental and disturbance controls (with K.
  Waddell, PNW-FIA)
• Describing variation within wildlife habitats for
  DecAID Advisor (with DecAID Science Team)

27
Inventory Plots Used in Dead Wood Analyses
Wildlife habitat type
No. plots
4,637
Westside conifer-hardwood
514
Westside white oak-Douglas-fir
SWO mixed conifer-hdwd
1,024
Montane mixed-conifer
2,485
Subalpine parkland
139
Eastside mixed-conifer
4,752
Eastside ponderosa pine
2,194
Lodgepole pine
757
Western juniper
365
Total
16,867
28
Variation across Wildlife Habitat Types

• Regional variation was most strongly associated
  environmental / productivity gradients (wildlife
  habitat types)

WLCH Westside conifer-hardwood WODF Westside
white oak-Douglas-fir SWOMC SWO mixed
conifer-hdwd MMC Montane mixed-conifer PARK
Subalpine parkland EMC Eastside
mixed-conifer EPPWO Eastside ponderosa pine LP
Lodgepole pine WJ Western juniper
29
Successional Trends in Dead Wood

• Dead wood abundance generally increases with
  successional development
• More pronounced with snags than down wood

Early lt25 cm Mid 25-50 cm Late gt50 cm
30
Effects of Harvesting Disturbance

• Snags more abundant in unharvested
• Down wood similar in harvested and unharvested
  (except MMC)
• Snags track recent disturbance while down wood
  tracks environmental gradients

WLCH Westside conifer-hardwood WODF Westside
white oak-Douglas-fir SWOMC SWO mixed
conifer-hdwd MMC Montane mixed-conifer PARK
Subalpine parkland EMC Eastside
mixed-conifer EPPWO Eastside ponderosa pine LP
Lodgepole pine WJ Western juniper
31
Summarizing Regional Inventory Data for the
DecAID Decayed Wood Advisor

• Statistical synthesis of wildlife and inventory
  data to guide management decisions
• 9 wildlife habitat types and 3 structural
  conditions
• Quantify dead wood populations on plots for
• Unharvested plots proxy for natural conditions
• All plots current landscape conditions, regional
  context

DecAID Science Team K.Mellen (R6) K.Waddell,
B.Marcot, T.Dreisbach (PNW), S.Livingston
(USFWS), B.Willhite and B.Hostetler (R6), C.Ogden
(NPS)
32
Non-Normal Snag Distributions in westside-lowland
conifer-hardwood forest, Oregon Coast, smaller
trees
n67 unharvested plots, snags gt50
n67 unharvested plots, snags gt25
All (n307) plots, snags gt25
All (n307) plots, snags gt50
33
Tolerance Intervals

• Distribution-free, using an ordering statistic
• Tolerance limit value associated with kth
  observation (quantile), based on binomial
  distribution and sample size
• Gamma (certainty level) 90, betas of 30, 50,
  80
• Example for Oregon coast, smaller trees, snags
  gt50 cm (n42) 90 certain that 50 of the
  forest area has lt 13.1 snags/ha.

45.1 (max.)
1.3 (min.)
5.2 (30 limit)
13.1 (50 limit)
28.8 (80 limit)
34
Example density of snags gt25 cm in Westside
Lowland Conifer-Hardwood Forest, western Oregon
Cascades, Large Trees structural condition
Wildlife study data
n 247 inventory plots
n 24 Spies OG plots 80-195 yr
n 53 Spies OG plots gt200 yr
Plot locations (white unharvested)
35
Key DecAID Findings

• Inventory and wildlife data are generally
  consistent in terms of dead wood abundances.
• Amounts are generally higher than current
  management guidelines.
• Disagreement on how to translate statistical
  summaries into management recommendations.
• Caveats and limitations...
• Sampling gaps (dead wood, ownerships) and
  irregularities
• Lack information on spatial pattern (within or
  between plots)
• Disturbance data unharvested not equal to
  natural or historic range of variability,
  especially eastside
• Scaling up plot data to guide management at
  broader scales

36
For more information

• DecAID website, in final stages of peer review
  and beta testing
• Ohmann, J.L. Waddell, K.L. 2002. Regional
  patterns of dead wood in forested habitats of
  Oregon and Washington. In Laudenslayer, et al.
  Proceedings of the symposium on the ecology and
  management of dead wood in western forests. Gen.
  Tech. Rep. PSW-GTR-181, p. 535-560.
• Mellen, et al. 2002. DecAID a decaying wood
  advisory model for Oregon and Washington. In
  Laudenslayer, et al. Proceedings of the symposium
  on the ecology and management of dead wood in
  western forests. Gen. Tech. Rep. PSW-GTR-181
  527-533.