Developing a National Data Surface of FIA Plot Data

Kenneth Brewer -- USFS, Remote Sensing Research Program Leader John Coulston -- USFS, Southern Research Station Sean Healey -- USFS, Rocky Mountain Research Station Emilie Henderson -- Oregon State University, Institute for Natural Resources Gretchen Moisen -- USFS, , Rocky Mountain Research Station Gretchen Nicholas -- USFS, Program Manager, PNW Research Station Janet Ohmann -- USFS, PNW Research Station Barry Ty Wilson -- USFS, NRS Forest Inventory and Analysis

What

• Build a 30-m resolution raster data surface describing forest composition and structure using nearest neighbor imputation from FIA plot data (NN data surface), across all forests of the lower 48 states.

Why?• Statistically valid mapped estimates of forest

structure and composition have increasing utility for conservation and natural resource management.– Problem solving occurs at strategic level, State

assessments, EIS requirements for cumulative impacts analysis.

– Issues like fire, disease, wildlife, and conservation plans require information on spatial distribution of forest structure and composition.

• Abundant regional projects demonstrate the need. • Coordinated national effort is needed.

McRoberts et al., 2002– 1999 to 2000

Regional Projects

CLAMS – 1996

GNNFire – 2000

GNNFire – 2000

GNNFire – 2000

PNW ReGAP, IMAP – 2000

NaFIS - 2000

CMonster - Coming

NaFIS - 2000

NaFIS 2000

NaFIS 2000 NaFIS 2000

ILAP – 2006

NWFP Effectiveness Monitoring, 15 yr – 2006/7

NWFP Effectiveness Monitoring, 20 yr – Coming

Wilson et al., 2012 - 250m – 2001 to2006

Franco-Lopez et al., 2001 -- 1988

Hudak et al. 2008 - 2003

Homer, C., Dewitz, J., Fry, J., Coan, M., Hossain, N., Larson, C., Herold, N., McKerrow, A., VanDriel, J.N., and Wickham, J. 2007. Completion of the 2001 National Land Cover Database for the Conterminous United States. Photogrammetric Engineering and Remote Sensing, Vol. 73, No. 4, pp 337-341.

Lister et al., 2004 – 1999 to 2001

Meng 2006 - 2005

What is nearest neighbor imputation?

• NN imputation is a method for building raster data surfaces from forest inventory data (Eskelson 2009).

• A method for filling in missing values in a dataset, from known values in that same dataset.

• The family of imputation methods includes: – k-NN (Tomppo 1991)– MSN (Moeur and Stage 1995)– GNN (Ohmann and Gregory 2002)– RFNN (Crookston and Finley 2008)

studyarea

(2) Place new pixel

withinfeature space

(3) find nearest-neighbor plot within feature

space

(4) impute nearest

neighbor’s value to

pixel

What is nearest neighbor imputation?

feature space geographic space

Elevation

Rainfall

(1)Place plots

within feature space

Raster Data: Topography Climate Soils Landsat

Plot Data

– Maintains multivariate covariance (Ohmann et al. 2011*, Wilson et al., 2012) .

– Statistical distribution of predicted values mimics that of input values.• Basal area: (Wilson et al., 2012).

Why nearest neighbor imputation?

* Manifested in realistic predictions of species composition, and joint species distributions

• Validity EstablishedTomppo, EO. (1991) Satellite image-based national forest inventory of Finland. International Archives

of Photogrametry and Remote Sensing 28.7-1 :419-424. -- Cited by 179

Moeur, M and AR Stage (1995). Most Similar Neighbor: An Improved Sampling Inference Procedure for Natural Resource Planning. Forest Science 41(2): 337-359. -- Cited by 205

Ohmann, JL and MJ Gregory (2002). Predictive mapping of forest composition and structure with direct gradient analysis and nearest- neighbor imputation in coastal Oregon, U.S.A. Canadian Journal of Forest Research 32(4): 725-741. -- Cited by 250

McRoberts, RE., and EO Tomppo. (2007) Remote sensing support for national forest inventories. Remote Sensing of Environment 110.4 : 412-419. -- Cited by 85

Ohmann, JL, MJ Gregory, EB Henderson and HM Roberts (2011). Mapping gradients of community composition with nearest-neighbor imputation: extending plot data for landscape analysis. Journal of Vegetation Science 22(4): 660-676. -- Cited by 17

Wilson, BT, AJ Lister and RI Riemann (2012). A nearest-neighbor imputation approach to mapping tree species over large areas using forest inventory plots and moderate resolution raster data. Forest Ecology and Management 271(0): 182-198.: -- came out in May, Cited by 1

Is more study needed?

Is more study needed?• Feasibility Demonstrated

– Regional-scale projects (e.g., Pierce et al., 2009, Ohmann et al. 2011).

– National pilot project (Nationwide Forest Imputation Study, Grossmann et al., 2009).

– Related nationwide projects:• National Land Cover Dataset’s Canopy

layer update (Coulston et al. 2012).• 250-m resolution nationwide NN data

surface (Wilson et al., 2012).• Landfire’s Tree-list layer (Drury and

Hernyk 2011)• North American Forest Dynamics

project (NAFD, Goward et al., 2008).

Wilson, BT, AJ Lister and RI Riemann (2012). A nearest-neighbor imputation approach to mapping tree species over large areas using forest inventory plots and moderate resolution raster data. Forest Ecology and Management 271(0): 182-198.

Does this proposal duplicate other efforts?

• No existing projects have all desired traits:

–All forests for 48 coterminous United States–Methods tuned to:• Maintain compatibility with FIA plot sample.• Handle multiple variables simultaneously.

– 30-m spatial resolution; • However, current project headed by Ty Wilson at

250-m pixel resolution is a good pilot project for imputation at the national scale.

Features of proposal

• Phased; build on 250-m project that links to forest atlas (Wilson et al. 2012), and the NLCD Canopy Cover update that is moving into its production phase (Coulston et al. 2012). Move in stages through one time period, region by region, at 30-m resolution, always maintaining national consistency.

• National team to resolve issues of:– Consistency with FIA estimates from plots.– Coherent maps through time.– Characterization of statistical properties of data (establish

standards, specify scales).

Workflow

• Input data management – plot and image management (FIA,RSAC).

• Model building and NN Data Surface Creation– Four methods currently being used. Will choose.

• Accuracy Assessment and Metadata Creation.• Distribution and User Support.– Collaboration between FIA, RSAC, and the National

Atlas (http://www.nationalatlas.gov/) that currently distributes NLCD’s canopy cover layer will serve as a model.

National Project

ManagerScience Team Production

Team

Oversight Team

• Mimic successful administrative structures.• Build upon the successes of existing projects.• Strengthen existing collaboration with RSAC as

partner.

Our Strategy

Next steps

• Agree to build proposal for funding from IRB.

• Hire contractor or assign staff to build proposal for IRB.

• Funding request would include provisions for resolution of existing technical issues, building a plan for phased implementation, and first steps of implementation.

ReferencesCoulston, JW, GG Moisen, BT Wilson, MV Finco, WB Cohen, et al. (2012). Modeling Percent Tree Canopy Cover: A Pilot Study.

Photogrammetric engineering and remote sensing 78(7): 715-727.Crookston, NL and AO Finley (2008). yaImpute: An R package for kNN imputation. Journal of Statistical Software 23(10): -.Drury, SA and JM Herynk (2011). The national tree-list layer, US Department of Agriculture, Forest Service, Rocky Mountain

Research Station.Eskelson, BNI, H Temesgen, V Lemay, TM Barrett, NL Crookston, et al. (2009). The roles of nearest neighbor methods in imputing

missing data in forest inventory and monitoring databases. Scandinavian Journal of Forest Research 24(3): 235-246.Franco-Lopez, H, AR Ek and ME Bauer (2001). Estimation and mapping of forest stand density, volume, and cover type using the k-

nearest neighbors method. Remote Sensing of Environment 77(3): 251-274.Goward, S, J Masek, W Cohen, G Moisen, G Collatz, et al. (2008). Forest disturbance and North American carbon flux. Eos 89(11).Grossmann, E, J Ohmann, M Gregory and H May (2009). Nationwide Forest Imputation Study (NaFIS) - Western Team final report.

http://www.fsl.orst.edu/lemma/export/pubs/grossmann_etal_2009_nafis_report.pdf.Hudak, AT, NL Crookston, JS Evans, DE Hall and MJ Falkowski (2008). Nearest neighbor imputation of species-level, plot-scale forest

structure attributes from LiDAR data. Remote Sensing of Environment 112(5): 2232-2245.Lister, A, M Hoppus and RL Czaplewski (2004). K-nearest neighbor imputation of forest inventory variables in New Hampshire. Tenth

Forest Service Remote Sensing Applications Conference, Salt Lake City.McRoberts, RE, MD Nelson and DG Wendt (2002). Stratified estimation of forest area using satellite imagery, inventory data, and

the k-Nearest Neighbors technique. Remote Sensing of Environment 82(2–3): 457-468.Meng, Q, C Cieszewski, M Madden and B Borders (2007). K Nearest Neighbor Method for Forest Inventory Using Remote Sensing

Data. GIScience & Remote Sensing 44(2): 149-165.Moeur, M, JL Ohmann, RE Kennedy, WB Cohen, MJ Gregory, et al. (2011). Northwest Forest Plan-the first 15 years (1994-2008):

status and trends of late-successional and old-growth forests. Moeur, M and AR Stage (1995). Most Similar Neighbor: An Improved Sampling Inference Procedure for Natural Resource Planning.

Forest Science 41(2): 337-359.Ohmann, JL and MJ Gregory (2002). Predictive mapping of forest composition and structure with direct gradient analysis and

nearest- neighbor imputation in coastal Oregon, U.S.A. Canadian Journal of Forest Research 32(4): 725-741.Ohmann, JL, MJ Gregory, EB Henderson and HM Roberts (2011). Mapping gradients of community composition with nearest-

neighbour imputation: extending plot data for landscape analysis. Journal of Vegetation Science 22(4): 660-676.