Spatial monitoring of older forest for the Northwest Forest Plan

25
Spatial monitoring of older forest for the Northwest Forest Plan Janet Ohmann 1 , Matt Gregory 2 , Heather Roberts 2 , Robert Kennedy 2 , Warren Cohen 1 , Zhiqiang Yang 2 , Melinda Moeur 3 , and Maria Fiorella 4 1 Vegetation Monitoring and Remote Sensing Team (VMaRS) Resource Monioring and Assessment Program (RMA) PNW Research Station, USFS, Corvallis, OR 2 Department of Forest Ecosystems and Society Oregon State University, Corvallis, OR 3 Region 6, USFS, Portland, OR; 4 BLM, Portland, OR Funding contributed by: Region 6, USFS PNW Research Station (WWETAC and ECOP)

description

Spatial monitoring of older forest for the Northwest Forest Plan. Janet Ohmann 1 , Matt Gregory 2 , Heather Roberts 2 , Robert Kennedy 2 , Warren Cohen 1 , Zhiqiang Yang 2 , Melinda Moeur 3 , and Maria Fiorella 4 1 Vegetation Monitoring and Remote Sensing Team (VMaRS) - PowerPoint PPT Presentation

Transcript of Spatial monitoring of older forest for the Northwest Forest Plan

Page 1: Spatial monitoring of  older  forest  for the Northwest Forest Plan

Spatial monitoring of older forest for the Northwest Forest Plan

Janet Ohmann1, Matt Gregory2, Heather Roberts2, Robert Kennedy2, Warren Cohen1, Zhiqiang Yang2, Melinda Moeur3,

and Maria Fiorella4

1 Vegetation Monitoring and Remote Sensing Team (VMaRS) Resource Monioring and Assessment Program (RMA)

PNW Research Station, USFS, Corvallis, OR2 Department of Forest Ecosystems and Society

Oregon State University, Corvallis, OR3 Region 6, USFS, Portland, OR; 4 BLM, Portland, OR

Funding contributed by: Region 6, USFS

PNW Research Station (WWETAC and ECOP)

Page 2: Spatial monitoring of  older  forest  for the Northwest Forest Plan

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

Page 3: Spatial monitoring of  older  forest  for the Northwest Forest Plan

Northwest Forest Plan of 1994• Conservation plan for older forests and

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?

Physiographic

provinces(57 mill. ac.,46 mill. ac

forest)

USA

Page 4: Spatial monitoring of  older  forest  for the 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.

Page 5: Spatial monitoring of  older  forest  for the Northwest Forest Plan

Overview of LSOG monitoring for 15-year report:

integration of map- and plot-based analysesMap-based analyses

Plot-based analyses• Successive

inventories (where available)

• FIA Annual inventory, all ownerships (no remeasurement)

Habitat and Watershed Condition

Page 6: Spatial monitoring of  older  forest  for the Northwest Forest Plan

Gradient Nearest Neighbor Imputation (GNN)

k=1

Page 7: Spatial monitoring of  older  forest  for the Northwest Forest Plan

Regional inventory plots for GNN modeling• Multiple data sources, unbalanced in time and space• One plot per location, matched to 94/96 or 06/07 imagery• Develop single gradient model with all plots• Apply model to each imagery year• Imagery is only source of change – assumes normalization

Imagery years

Page 8: Spatial monitoring of  older  forest  for the Northwest Forest Plan

Landsat Detection of Trends in Disturbance and Recovery (LandTrendr)*

• Temporal normalization and segmentation at pixel level

• Minimizes noise from sun angle, phenology

• Segments describe sequences of disturbance, regrowth

• Yearly time-step• Detects gradual

and subtle changes

• Normalized imagery for multiple years for GNN modeling*Kennedy et al. (2010), Rem. Sens. Env.

Page 9: Spatial monitoring of  older  forest  for the Northwest Forest Plan

Defining‘late-successional and old growth’ (LSOG) forest• Single, simple definition, applied to tree-level data

associated with GNN pixels

• LSOG ≠ habitat!• More ecologically-based definition => different answers

(but not necessarily more accurate)

Forest class

Conifer canopy cover

Avg. DBH of dom/codom

conifersOpen <10% --

Young >10% 0.0 to 19.9 in

LSOG >10% >20 in

Page 10: Spatial monitoring of  older  forest  for the Northwest Forest Plan

Mapping LSOG change

Not LSOGLSOG gainLSOG lossLSOG Nonforest

1996 B-G-W 2006 B-G-W Disturbance

1996 LSOG 2006 LSOG LSOG change

Land-Trendr

GNN

- - 10 miles - -

Page 11: Spatial monitoring of  older  forest  for the Northwest Forest Plan

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: – Area distributions in map vs. plot sample– Range of variation in map vs. plot

sample– Riemann et al. (2010) diagnostics– 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

1

2 3 4

5 6 7* 8 9

10 11 12

13

local(1-ha

plot) scale

regionalscale

landscape- or watershed-

scale

Oregon

Page 12: Spatial monitoring of  older  forest  for the Northwest Forest Plan

Results

Page 13: Spatial monitoring of  older  forest  for the Northwest Forest Plan

LSOG change from GNN ‘bookend’ maps, 1994/6 to 2006/7

• GNN models and change at 30-m pixel scale

• Recommend summarizing to coarser scales

• Example: 10-km hexagonsLSOGchange(% of forest)

Page 14: Spatial monitoring of  older  forest  for the Northwest Forest Plan

Spatial change in Klamath province,

1996-2006

Not LSOGLSOG gainLSOG lossLSOG Nonforest

• Change is dramatic in some landscapes (2002 Biscuit Fire)

• Spatial change is quite noisy

Page 15: Spatial monitoring of  older  forest  for the Northwest Forest Plan

Change in older forest on federal land• ~ 2/3 of total LSOG• Net loss of 1.9%, 7.3 to 7.1 mill. ac, from 33.2% to 32.5% of forest• >200,000 acres lost in large fires (LandTrendr disturbance), 90% in

reserves• Losses roughly offset by recruitment, but difficult to reliably map• Small amount of change relative to level of uncertainty

Page 16: Spatial monitoring of  older  forest  for the Northwest Forest Plan

Change in older forest on nonfederal lands• ~ 1/3 of total LSOG• Net loss of 9.9%, from 3.9 to 3.5 mill. ac.• >500,000 acres lost, mostly timber harvest (LandTrendr disturbance

maps)• Losses not offset by recruitment• Small amount of change relative to level of uncertainty

Page 17: Spatial monitoring of  older  forest  for the Northwest Forest Plan

Comparison of GNN and FIA Annual estimates• GNN shows less LSOG on federal, more LSOG on nonfederal,

very similar for all ownerships• Many reasons for differences: different plots, different dates,

sample- vs. model-based, unsampled area, nonforest area, etc. etc. etc.Federal Nonfederal All owners

Acknowledgment: Olaf Kuegler and Karen Waddell for FIA Annual estimates

Page 18: Spatial monitoring of  older  forest  for the Northwest Forest Plan

Change in older forest from successive inventories• National Forest and Oregon BLM lands only• Differences between estimates were not significant (all provinces,

states)• GNN estimates are within the sampling error (90% C.I.)• Except Calif. (Region 5 FIA vs. FIA Annual) – data problem?

??

Page 19: Spatial monitoring of  older  forest  for the Northwest Forest Plan

Change in habitat suitabilityNWFP Effectiveness

Monitoring

• Maxent (machine learning) models based on forest structure and composition attributes from GNN, trained with nest location data

• Subtract models to get change

Marbled murrelet

Northern spotted owl

Page 20: Spatial monitoring of  older  forest  for the Northwest Forest Plan

Error and Uncertainty in the Monitoring Data

“Those pixels are wrong!”

Page 21: Spatial monitoring of  older  forest  for the Northwest Forest Plan

How good are the GNN ‘bookend’ maps?• Local-scale accuracy (cross-validation)

– LSOG is 80% correct, kappa 0.49– Normalized RMSE for CANCOV = 0.33, QMDCDOM = 0.53– Best in closed-canopy, conifer-dominated, even-aged forest

(challenges in patchy stands of mixed ages and species)• Regional LSOG area estimates are comparable to FIA Annual • Need kNN bootstrapped variance estimators for kNN to statistically

compare two models (Magnussen et al. 2010)• How reliably can we map LSOG change?

– TimeSync validation tool (Cohen et al. 2010) to assess change spatially

Cohen, W.B.; Zhiqiang, Y.; Kennedy, R.E. 2010. Detecting trends in forest disturbance and recovery using yearly Landsat time series: 2. TimeSync—tools for calibration and validation. RSE 114: 2911-2924.

Magnussen, S.; McRoberts, R.E.; Tomppo, E.O. 2010. A resampling variance estimator for the k nearest neighbours technique. CJFR 40:648-658.

Page 22: Spatial monitoring of  older  forest  for the Northwest Forest Plan

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 LSOG definition: – Simple QMD threshold, can be affected by one or a few

trees– Disturbance can => LSOG gain, LSOG loss, or no change

Page 23: Spatial monitoring of  older  forest  for the Northwest Forest Plan

Monitoring: improving methods, rewriting history?• Capability to re-run models for previous years (can users stomach

it?)• 10-year* and 15-year monitoring data:

– Map analyses: similar estimates for WA/OR, very different for CA– Plot analyses: large amount of projected LSOG recruitment not

supported

Federal lands

Baseline LSOG map estimates10-year

report* (IVMP, CalVeg)

15-year report (GNN)

Difference

Thousand acres

Washington 2,131 2,131 +0.4Oregon 3,379 3,400 +20.5California 2,358 1,754 -603.4Range-wide 7,868 7,286 -582.4

* 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.

Page 24: Spatial monitoring of  older  forest  for the Northwest Forest Plan

Products from NWFP monitoring study • GNN models and diagnostics available for download

– 2006/7 now available, 1994/96 pending peer review and publication

• 15-year reports (PNW GTRs) in review:– LSOG, northern spotted owl, marbled murrelet, watershed

condition• Article (in prep.) for Forest Ecology and Management

http://www.fsl.orst.edu/lemma/nwfp

Page 25: Spatial monitoring of  older  forest  for the Northwest Forest Plan

Thanks for your attention!