39
Mid Scale Data Data to Model Ecosystem Service Outputs for Agency Planning Jimmy Kagan OSU | PSU | UO

Mid Scale Data - UF/IFAS OCI | Home 11 Thursda… · plot in gradient space (4) impute nearest neighbor’s value to pixel Methods: GNN gradient space geographic space CCA Axis 2

  • Upload
    others

  • View
    2

  • Download
    0

Embed Size (px)

Citation preview

Page 1: Mid Scale Data - UF/IFAS OCI | Home 11 Thursda… · plot in gradient space (4) impute nearest neighbor’s value to pixel Methods: GNN gradient space geographic space CCA Axis 2

Mid Scale Data

Data to Model Ecosystem Service Outputs for Agency 

Planning

Jimmy Kagan

OSU | PSU | UO

Page 2: Mid Scale Data - UF/IFAS OCI | Home 11 Thursda… · plot in gradient space (4) impute nearest neighbor’s value to pixel Methods: GNN gradient space geographic space CCA Axis 2

Presentation Goals & Outline

1. What is Mid‐Scale Data and how does it compare to National and Local scale info?

2. How is it generated?

3. What does it get you and why is it needed for ecosystem service outputs? 

4. How is it used in agency planning?

Page 3: Mid Scale Data - UF/IFAS OCI | Home 11 Thursda… · plot in gradient space (4) impute nearest neighbor’s value to pixel Methods: GNN gradient space geographic space CCA Axis 2

What is Mid‐Scale Data?

• Defined by– Spatial Extent– Spatial Resolution– Attribute Resolution

Page 4: Mid Scale Data - UF/IFAS OCI | Home 11 Thursda… · plot in gradient space (4) impute nearest neighbor’s value to pixel Methods: GNN gradient space geographic space CCA Axis 2

Spatial Extent

• Local  (e.g., map of birds at PDX)• Regional (e.g., GNN maps of the PNW)• Nationwide (e.g., LANDFIRE EVT, USGS GAPvegetation)

Page 5: Mid Scale Data - UF/IFAS OCI | Home 11 Thursda… · plot in gradient space (4) impute nearest neighbor’s value to pixel Methods: GNN gradient space geographic space CCA Axis 2

Spatial Resolution

• Fine‐scale data = 1m or finer spatial resolution (NAIP or LIDAR)

• Mid‐scale resolution = 30m spatial resolution (Landsat)

• Broad‐scale resolution >= 250m spatial resolution (Modis)

Page 6: Mid Scale Data - UF/IFAS OCI | Home 11 Thursda… · plot in gradient space (4) impute nearest neighbor’s value to pixel Methods: GNN gradient space geographic space CCA Axis 2

Attribute resolution

• Single‐species maps• Community types (e.g., Ecological Systems)• Multivariate information on vegetation community composition and structure– NN Imputation = a multivariate prediction– Maintenance of covariance structure (Henderson et al. 2009)

• Confidence of predictions

Page 7: Mid Scale Data - UF/IFAS OCI | Home 11 Thursda… · plot in gradient space (4) impute nearest neighbor’s value to pixel Methods: GNN gradient space geographic space CCA Axis 2

Uses for estimating ecosystem services

• When data depth is needed• Flexible vegetation summaries.• When accurate areal representation over large areas is of primary importance. (e.g., ‘how much of watershed ‘x’ is covered with closed‐canopy forests?)

Page 8: Mid Scale Data - UF/IFAS OCI | Home 11 Thursda… · plot in gradient space (4) impute nearest neighbor’s value to pixel Methods: GNN gradient space geographic space CCA Axis 2
Page 9: Mid Scale Data - UF/IFAS OCI | Home 11 Thursda… · plot in gradient space (4) impute nearest neighbor’s value to pixel Methods: GNN gradient space geographic space CCA Axis 2

2008 USGS Gap Landcover in Oregon

Page 10: Mid Scale Data - UF/IFAS OCI | Home 11 Thursda… · plot in gradient space (4) impute nearest neighbor’s value to pixel Methods: GNN gradient space geographic space CCA Axis 2

Example: Estimating habitat

Page 11: Mid Scale Data - UF/IFAS OCI | Home 11 Thursda… · plot in gradient space (4) impute nearest neighbor’s value to pixel Methods: GNN gradient space geographic space CCA Axis 2

What’s missing in GAP?

• Vegetation Structure– useful for estimating habitat value for wildlife or carbon or wood product availability

• Site attributes (down wood, snags, litter)– Useful for evaluating fire risk, wildlife habitat or biomass

• Species Composition of the Vegetation (system)– Useful for summarizing ecological condition 

• (e.g., invasive species)

Page 12: Mid Scale Data - UF/IFAS OCI | Home 11 Thursda… · plot in gradient space (4) impute nearest neighbor’s value to pixel Methods: GNN gradient space geographic space CCA Axis 2

(2) calculate

axis scores of pixel from

mapped data layersstudyarea

(3) find nearest-

neighbor plot in

gradient space

(4) impute nearest

neighbor’s value to

pixel

Methods: GNNgradient space geographic space

CCAAxis 2

(e.g., Climate)

CCAAxis 1

(e.g., elevation, Y)

(1)conductgradient

analysis ofplot data

Using nationally available FIA & AIM data to generate midscale data

Page 13: Mid Scale Data - UF/IFAS OCI | Home 11 Thursda… · plot in gradient space (4) impute nearest neighbor’s value to pixel Methods: GNN gradient space geographic space CCA Axis 2

Methods: Random Forest (RFNN)• One Classification Tree:

|MTM100 < 21.1095

DEM < 679.825

MTM300 < 13.874

SLPPCT < 29.3858

YFIRE < 3.15519

CANOPY < 42.085930725365 3415

4323

8793 48475767

Page 14: Mid Scale Data - UF/IFAS OCI | Home 11 Thursda… · plot in gradient space (4) impute nearest neighbor’s value to pixel Methods: GNN gradient space geographic space CCA Axis 2

Methods: Random Forest (RFNN)

• A whole forest of classification trees!

• Each tree model is built from a random subset of explanatory variables and input data.

• When the model is applied to mapped data, each tree ‘votes’ on which Plot best represents a pixel should be.

|TM100 < 22.9069

TM100 < 19.0223 FOG < 0.5

TM100 < 33.82934108 5120

5977 86398622

|ANNHDD < 2766.21

SLPPCT < 10.3216 STDTM100 < 16.1739

ANNHDD < 3469.43STDTM100 < 46.7235

9148 5675 3517 4192 4607 5832

|TM200 < 22.9549

STRATUS < 201.108

TM200 < 35.1356

4156

5559 82698694

|MTM300 < 28.9494

MTM300 < 13.8683 IDSURVEY < 0.5

MTM300 < 43.80133922 4672

6770 89475136

|DECMINT < 48.1696

MTM700 < 27.8564

DECMINT < -214.708DECMINT < -276.567R5700 < 145.622

R5700 < 185.313

4280 36687896 4737

6104

8506 5086

|TM100 < 22.9069

TM100 < 19.0223 DISTNF:b

4108 5120

5833 8480

Page 15: Mid Scale Data - UF/IFAS OCI | Home 11 Thursda… · plot in gradient space (4) impute nearest neighbor’s value to pixel Methods: GNN gradient space geographic space CCA Axis 2

Mid‐Scale Data vs National Data for Species Habitat Mapping

Northern Spotted Owl Habitat on Mt. Hood.

Greater Sage Grouse Habitat in Southeast Oregon.

Page 16: Mid Scale Data - UF/IFAS OCI | Home 11 Thursda… · plot in gradient space (4) impute nearest neighbor’s value to pixel Methods: GNN gradient space geographic space CCA Axis 2
Page 17: Mid Scale Data - UF/IFAS OCI | Home 11 Thursda… · plot in gradient space (4) impute nearest neighbor’s value to pixel Methods: GNN gradient space geographic space CCA Axis 2
Page 18: Mid Scale Data - UF/IFAS OCI | Home 11 Thursda… · plot in gradient space (4) impute nearest neighbor’s value to pixel Methods: GNN gradient space geographic space CCA Axis 2
Page 19: Mid Scale Data - UF/IFAS OCI | Home 11 Thursda… · plot in gradient space (4) impute nearest neighbor’s value to pixel Methods: GNN gradient space geographic space CCA Axis 2
Page 20: Mid Scale Data - UF/IFAS OCI | Home 11 Thursda… · plot in gradient space (4) impute nearest neighbor’s value to pixel Methods: GNN gradient space geographic space CCA Axis 2

Forest Canopy CoverAlso available in national scale data, but more accurate with     mid scales information

Page 21: Mid Scale Data - UF/IFAS OCI | Home 11 Thursda… · plot in gradient space (4) impute nearest neighbor’s value to pixel Methods: GNN gradient space geographic space CCA Axis 2

Tree DiameterAvailable only in mid (or fine) scale data and needed to model ecosystem service outputs

Page 22: Mid Scale Data - UF/IFAS OCI | Home 11 Thursda… · plot in gradient space (4) impute nearest neighbor’s value to pixel Methods: GNN gradient space geographic space CCA Axis 2

Why Is Mid Scale Data Needed to Model Ecosystem Service Outputs 

for Land Management Agency Planning?

Because it is the data needed to make State and Transition Models work

Page 23: Mid Scale Data - UF/IFAS OCI | Home 11 Thursda… · plot in gradient space (4) impute nearest neighbor’s value to pixel Methods: GNN gradient space geographic space CCA Axis 2

State‐and‐Transition Models (STMs)• Simulate changes in vegetation over time due to succession, 

disturbance, and management activities• Are a modeling form of Box and Arrow Diagrams• Agency models historically were developed in the Vegetation Dynamics 

Development Tool (VDDT) which are non‐spatial simulations.• These simulations run in the Path Landscape Model• A spatial version (ST‐Sim) is available from APEX (apexrms.com)

Page 24: Mid Scale Data - UF/IFAS OCI | Home 11 Thursda… · plot in gradient space (4) impute nearest neighbor’s value to pixel Methods: GNN gradient space geographic space CCA Axis 2

Herbaceous    Open shrub    Closed shrub    Woodland

Perennial bunchgrasses

Mixed grass

Exotic annual grasses

SUCCESSIONDEG

RADA

TION

DISTURBANCE

State‐and‐Transition Models

Western Juniper / mountain big sagebrush

Herbaceous layer:

Page 25: Mid Scale Data - UF/IFAS OCI | Home 11 Thursda… · plot in gradient space (4) impute nearest neighbor’s value to pixel Methods: GNN gradient space geographic space CCA Axis 2

Thinning

Low‐severity fire

High‐severity fire

State and Transitions describe changes over time, which can be converted to ecosystem outputs 

Planting

Regeneration

Page 26: Mid Scale Data - UF/IFAS OCI | Home 11 Thursda… · plot in gradient space (4) impute nearest neighbor’s value to pixel Methods: GNN gradient space geographic space CCA Axis 2

Use Mid Scale Data to Run Models and Hook to Outputs

Fuels

Wildlife habitat

Treatment finances 

Output Intrepretations

Watershed

Evaluate management scenario

State and Transition Models

Wildfire‐fuel hazards

Terrestrial habitat

Aquatic habitat

Economic potential

Ecosystem Services

Page 27: Mid Scale Data - UF/IFAS OCI | Home 11 Thursda… · plot in gradient space (4) impute nearest neighbor’s value to pixel Methods: GNN gradient space geographic space CCA Axis 2

Merchantable Volume, Sawtimber, Biomass

Deschutes National Forest Output Example

Page 28: Mid Scale Data - UF/IFAS OCI | Home 11 Thursda… · plot in gradient space (4) impute nearest neighbor’s value to pixel Methods: GNN gradient space geographic space CCA Axis 2

Wallowa‐Whitman project area –Current Vegetation

Page 29: Mid Scale Data - UF/IFAS OCI | Home 11 Thursda… · plot in gradient space (4) impute nearest neighbor’s value to pixel Methods: GNN gradient space geographic space CCA Axis 2

VegetationNo Treatments30 Years

Page 30: Mid Scale Data - UF/IFAS OCI | Home 11 Thursda… · plot in gradient space (4) impute nearest neighbor’s value to pixel Methods: GNN gradient space geographic space CCA Axis 2

VegetationNo Management 150 years

Page 31: Mid Scale Data - UF/IFAS OCI | Home 11 Thursda… · plot in gradient space (4) impute nearest neighbor’s value to pixel Methods: GNN gradient space geographic space CCA Axis 2

Alternative 2 Treatments first decade

Page 32: Mid Scale Data - UF/IFAS OCI | Home 11 Thursda… · plot in gradient space (4) impute nearest neighbor’s value to pixel Methods: GNN gradient space geographic space CCA Axis 2

Vegetation, Alt. 2, 30 Years

Page 33: Mid Scale Data - UF/IFAS OCI | Home 11 Thursda… · plot in gradient space (4) impute nearest neighbor’s value to pixel Methods: GNN gradient space geographic space CCA Axis 2

Vegetation, Alt 2, 150 years

Page 34: Mid Scale Data - UF/IFAS OCI | Home 11 Thursda… · plot in gradient space (4) impute nearest neighbor’s value to pixel Methods: GNN gradient space geographic space CCA Axis 2

MC2 Dynamic Global Vegetation Model

Climate‐informedState‐and‐Transition Models

Integrating Climate Change & Land Management Models

Climate Information

Oak woodlands

Early Mid

Late Conifer

Chaparral

Ponderosa Pine

Early Mid

LateOld‐

growth

Douglas‐fir

Early Mid

LateOld‐

growth

Page 35: Mid Scale Data - UF/IFAS OCI | Home 11 Thursda… · plot in gradient space (4) impute nearest neighbor’s value to pixel Methods: GNN gradient space geographic space CCA Axis 2

Midscale data is not all TerrestrialNWI is relatively high spatial resolution with uncertain accuracy….yet can

Page 36: Mid Scale Data - UF/IFAS OCI | Home 11 Thursda… · plot in gradient space (4) impute nearest neighbor’s value to pixel Methods: GNN gradient space geographic space CCA Axis 2

Channel Type Classification

Based on Montgomery and Buffington 1997

Page 37: Mid Scale Data - UF/IFAS OCI | Home 11 Thursda… · plot in gradient space (4) impute nearest neighbor’s value to pixel Methods: GNN gradient space geographic space CCA Axis 2

And Can Include Species

Page 38: Mid Scale Data - UF/IFAS OCI | Home 11 Thursda… · plot in gradient space (4) impute nearest neighbor’s value to pixel Methods: GNN gradient space geographic space CCA Axis 2

Mid‐scale data can improve species mapsCalifornia mountain kingsnake Lampropeltis zonata

Predicted Habitat quality within occupied watersheds

Page 39: Mid Scale Data - UF/IFAS OCI | Home 11 Thursda… · plot in gradient space (4) impute nearest neighbor’s value to pixel Methods: GNN gradient space geographic space CCA Axis 2

Conclusions• Mid Scale data can provide information to improve modeling of ecosystem service outputs

• Mid Scale data provides critical inputs into state and transition models which predict many ecosystem service outputs from management decisions

• Mid Scale data is created from nationally available datasets with proven methods

• Mid Scale data is available in some areas of the US, and cab be developed across the country if federal agencies worked together and decided to make it happen

Jimmy Kagan, INR‐Portland, [email protected]@oregonstate.edu,  503‐725‐9955