full project text

19
Page i Integrated, observation-based carbon monitoring for wooded ecosystems in Washington, Oregon, and California A proposal for NASA Roses 2010 Carbon Announcement A.5 Carbon Cycle Science PI: Kennedy (Oregon State University) Co Is: Ohmann, Cohen (USFS) Franklin, Kane, Lutz (University of Washington) Powell (Montana State University) Table of contents Integrated, observation-based carbon monitoring for wooded ecosystems in Washington, Oregon, and California ................................................................................... i Table of contents........................................................................................................... i Introduction.................................................................................................................. 1 Objectives & Significance ........................................................................................... 3 Technical approach ...................................................................................................... 5 Task 1: Develop statewide disturbance, growth, and change agent maps for Washington, Oregon, and California from 1985 to 2010. ................................ 6 Task 2: Develop forest carbon maps using nearest-neighbor mapping approaches to link FIA and LandTrendr data for Washington, Oregon, & California ........ 9 Task 3: Develop lidar-based carbon surfaces for selected areas .......................... 10 Perceived impact in the field ..................................................................................... 13 Relevance to NASA................................................................................................... 13 Plan of work............................................................................................................... 14 References and Citations ........................................................................................... 16

Upload: duongthien

Post on 31-Dec-2016

236 views

Category:

Documents


3 download

TRANSCRIPT

Page 1: Full project text

Page i

Integrated, observation-based carbon monitoring for wooded ecosystems in Washington, Oregon, and California A proposal for NASA Roses 2010 Carbon Announcement A.5 Carbon Cycle Science

PI: Kennedy (Oregon State University) Co Is: Ohmann, Cohen (USFS) Franklin, Kane, Lutz (University of Washington) Powell (Montana State University)

Table of contents Integrated, observation-based carbon monitoring for wooded ecosystems in

Washington, Oregon, and California ................................................................................... i  Table of contents........................................................................................................... i  Introduction.................................................................................................................. 1  Objectives & Significance ........................................................................................... 3  Technical approach...................................................................................................... 5  

Task 1: Develop statewide disturbance, growth, and change agent maps for Washington, Oregon, and California from 1985 to 2010. ................................ 6  

Task 2: Develop forest carbon maps using nearest-neighbor mapping approaches to link FIA and LandTrendr data for Washington, Oregon, & California........ 9  

Task 3: Develop lidar-based carbon surfaces for selected areas .......................... 10  Perceived impact in the field ..................................................................................... 13  Relevance to NASA................................................................................................... 13  Plan of work............................................................................................................... 14  References and Citations ........................................................................................... 16  

Page 2: Full project text

Kennedy et al: Integrated, observation based carbon monitoring

Page 1

Introduction Predicting the fate of carbon in wooded ecosystems (both woodlands and forests;

from here forward, “forests”) under future climates is a fundamental scientific and management challenge, as these systems contain large reservoirs of carbon, provide many essential ecosystem services, and represent a potentially critical feedback in global climate change (Birdsey et al. 2007). Yet carbon storage in forested ecosystems is highly dynamic, affected by diverse anthropogenic and natural processes that can radically change the carbon trajectory of a landscape (Chambers et al. 2007; Environment-Canada 2006). Models and decision support tools that cannot realistically incorporate the full range of disturbance and growth processes will not stand the scrutiny of the public or policy makers (Joyce et al. 2009). Moreover, neither models nor management plans can exist in an historical vacuum: both must consider the how their potential future conditions relate to current trends and status (Birdsey et al. 2006).

Our understanding of possible forest carbon futures is hampered by our sparse observations of the recent carbon past. In the United States, the core tool used for national and international reporting of carbon flux has traditionally been the field measurements collected by the Forest Inventory and Analysis (FIA) program (Goodale et al. 2002; Woodbury et al. 2007). Though FIA measurements are now measured consistently, cover the nation, and are statistically defensible, their sparse spatial grain constrains estimation of carbon change to areas much larger than those needed by modelers and land managers. Additionally, their decadal repeat period is longer than that needed to capture anthropogenic and natural processes that significantly alter carbon pools every year (Houghton 2005).

Other tools have the potential to complement FIA plot information in capturing actual forest carbon conditions, but none can fully address carbon dynamics alone. Maps based on small-footprint lidar data have been shown to provide excellent estimates of forest biomass and hence carbon (Kim et al. 2009; Lefsky et al. 2002; Naesset 2002), but to date these data are available only for relatively small areas, and even in the future are unlikely to be repeated frequently enough over large areas to capture disturbance processes in a monitoring framework. Maps of forest disturbance from Landsat satellite imagery have fine enough resolution to capture individual patches of disturbance and growth, and also to meet the criterion of covering large areas (Huang et al. 2010). But Landsat-based maps of forest dynamics have focused almost exclusively on high-severity, abrupt disturbances such as clearcuts and fires (Cohen et al. 2002).

While abrupt disturbance processes are critical for understanding carbon dynamics, they omit key lower-severity anthropogenic disturbances such as mechanical thinning, subtle or long-lasting natural disturbances caused by insects and drought, and slow aggradation processes such as forest growth. Moreover, existing maps of forest disturbance provide little information on carbon status before disturbance nor on the agent causing the change, both of which can substantially alter the carbon consequences of a given change. Other efforts to map biomass or disturbance at broad spatial extents from satellite-based data (optical, radar, and lidar) have generally focused on single-date estimates of carbon stock (Goetz et al. 2009; Kellndorfer et al. 2004; Zheng et al. 2004), have spatial grains larger than that needed to adequately capture change (Blackard et al. 2008), and or do not distinguish the trajectory of biomass change nor the agents causing those changes.

Page 3: Full project text

Kennedy et al: Integrated, observation based carbon monitoring

Page 2

We propose to build an integrated system to monitor forest carbon dynamics for all forests of Washington, Oregon, and California for the period 1985 to present (Figure 2). This system will provide unprecedented understanding of the causes and carbon consequences of forest dynamics over areas large enough to 1) directly address fundamental science questions, and 2) evaluate the actual history of forest management across a range of ownerships, forest types, and climatic conditions. The design of the system will allow incorporation of new tools, algorithms, and data as each improves, and will explicitly provide estimates of uncertainty in carbon pools and fluxes, allowing independent evaluation of the significance of different trends across space and time.

Our team is uniquely positioned to build this system. Co-Investigator Ohmann is a recognized authority on statistical approaches to link FIA plot data with environmental and Landsat Thematic Mapper data to create rich forest maps over large areas (Ohmann and Gregory 2002b; Ohmann et al. 2007; Ohmann and Spies 1998; Pierce et al. in press). To date, her work has focused on single-date mapping, but recently has expanded into a multi-temporal realm in collaboration with Principal Investigator Kennedy. Co-Investigator Cohen has a long history of mapping forest changes, particularly related to carbon change (Cohen and Fiorella 1998; Cohen et al. 1998; Cohen et al. 1996; Cohen et al. 2002; Cohen et al. 1995). Recently, Kennedy developed new algorithms for change detection in dense Landsat image time-series (Kennedy et al. 2007; Kennedy et al. Accepted). These algorithms are unique in providing 1) new detail on forest dynamics needed to fully describe and label landscape change, and 2) the spectrally-stable annual data necessary to support Ohmann’s multitemporal mapping efforts. Co-Investigator Powell has experience mapping change in cover as well ass forest biomass from FIA plot data in the context of Landsat image time-series (Powell et al. 2010; Powell et al. 2008). Co-Investigators Franklin, Kane, and Lutz are emerging leaders in the science of forest mapping from small-footprint lidar, particularly in the forests of interest in this proposal (Kane, Bakker et al. 2010; Kane et al. 2008; Kane, McGaughey et al. 2010).

We emphasize that we have already developed the primary methodological pieces of the system, and have already implemented them separately at scales commensurate with the work proposed here. This proposal seeks to deepen methodological integration over larger areas, to better characterize uncertainties in carbon stocks and fluxes, and to use resulting maps to address several key science and management questions.

Page 4: Full project text

Kennedy et al: Integrated, observation based carbon monitoring

Page 3

Objectives & Significance Our objectives are to:

1. Quantify annual impacts of land management, land use change, and natural processes on carbon stocks and fluxes 2. Document and evaluate strategies used to reduce vulnerability of forests to unplanned

carbon loss 3. Establish a carbon monitoring structure and characterize uncertainties

Addressing project objectives relies on a set of deliverables derived from the proposed observational carbon monitoring system. Key deliverables include annual (1985 to present), 30m resolution maps for all wooded areas in Washington, Oregon, and California of:

• Forest disturbance year and magnitude • Agent of disturbance: clearcuts, partial cuts, insect mortality, fire, urbanization • Forest growth rate and period • Forest type, size, age • Live and dead aboveground forest carbon pools • Uncertainty estimation based on accumulated errors in monitoring as well as

comparison with small-footprint lidar maps Other important deliverables include robust lidar-based estimates of aboveground

carbon for several study areas, a unified forest/non-forest mask for the entire region, and evaluations of best practices for integrating lidar, Landsat, and FIA plot data.

Objective 1: Quantify annual impacts of land management, land use change, and natural processes on carbon stocks and fluxes.

Underlying science and management questions: 1.1.How much aboveground live forest carbon has been converted to other pools

by all disturbance at regional scales? How much is accumulating through post-disturbance growth of forest?

1.2.What are the observed effects of different forest management strategies on forest carbon? How are different agents of forest change distributed across ownership sectors and climatic gradients?

1.3.How much of the loss is caused by natural versus anthropogenic agents? How do different agents of carbon loss (clearcuts, thinnings, conversion, fire, insects, etc.) compare in cumulative losses? How much is cyclical (loss followed by growth) vs sustained (conversion of forest to other land uses)?

1.4.How variable are rates of disturbance-related carbon loss annually? To what extent are carbon loss rates at regional scales attributable to climatic, economic or policy factors?

Implementation: To address Questions 1.1-1.3, we will link our maps of change agent and annual live and dead forest carbon with ownership and other geospatial data to summarize carbon changes by time period, agent of change, and ownership. For areas with known policy shifts (such as the Northwest Forest Plan, implemented in the early 1990s), we will characterize changes over time relative to policy changes (Question 1.4).

Significance: This objective addresses section 3.1 in the Research Announcement, specifically research designed to address “effects of land management and land use on carbon and greenhouse gas fluxes” and the “tradeoffs between greenhouse gases and

Page 5: Full project text

Kennedy et al: Integrated, observation based carbon monitoring

Page 4

other products and services” in systems that are both experiencing high flux and that provide fiber production systems (wood products) and diverse other ecosystem services (water, wildlife, recreation).

Objective 2: Document and evaluate strategies used to reduce vulnerability of forests to unplanned carbon loss

Underlying science and management questions: 2.1.How much does forest thinning affect subsequent occurrence of fire or insect

attack? Are unmanaged forests more likely to burn or suffer insect outbreak than are managed forests?

2.2.Which ownerships or forest types have the greatest potential to accumulate carbon? Which have been most vulnerable to fire or insect-related loss?

Implementation: For Question 1, we will use our maps of change agent to identify patches in three populations distinguished by management strategies -- thinned, clearcut, and unmanaged – and then follow their subsequent fate in terms of fire and insect mortality. For Question 2, we will characterize carbon distributions by ownership and forest type strata, and then quantify fire and insect carbon loss within each stratum.

Significance: This objective addresses section 3.4 in the Research Announcement, specifically research designed to understand “the effects of human actions on …. carbon fluxes and ecosystem services” and “assessment of adaptation options in terms of their ability to reduce unwanted consequences while maintaining ecosystem services.”

Objective 3: Establish a carbon monitoring structure and characterize uncertainties

Underlying science and management questions: 3.1.Which components of the observational monitoring system contribute most to

uncertainty in estimates? 3.2.Can knowledge of recent forest history improve estimation of the current

condition of forest carbon uptake, storage, and loss? 3.3.Which characteristics of disturbance patches are most useful for ascribing causal

agents to change? 3.4.How can carbon estimation through small-footprint lidar best be used to evaluate

effectiveness of large-area carbon maps? Which approaches to mapping carbon with lidar data promote generic monitoring by transferability across different forest conditions?

Implementation: The system is set up in distinct analysis modules with internal uncertainty tracked by module. The impact of each module’s uncertainty on final map estimates can be evaluated (Question 1). Questions 2-4 will be evaluated within sub-tasks of the overall workflow by comparing products produced with different inputs or analytical approaches.

Significance: This objective addresses section 3.4 in the Research Announcement, specifically on

“development/evaluation of monitoring strategies for assessing the efficacy of or verifying reporting on mitigation actions.”

Addressing end-to-end climate change Forests represent a nexus between humans, climate, and ecosystem services. They

provide a range of services directly to humans and may blunt effects of climate change

Page 6: Full project text

Kennedy et al: Integrated, observation based carbon monitoring

Page 5

through carbon sequestration, but they are affected by changes in climate and by human responses to both climate and social pressures. This project addresses many key issues in the emerging interplay between humans, forests, and climate: • Causes: Because all carbon changes will be linked to change agent (attribution), our

data will provide information on the causes of forest carbon change -- critical for any U.S. climate observing system.

• Consequences: By documenting all observable disturbance and growth phenomena in the forested systems of our study area, we will provide fundamental information on the carbon consequences of management and natural processes.

• Interactions and feedbacks: Because we observe a range of processes in both space and time, we can document interactive effects that are likely to become increasingly important under climate change (for example, interactions between insect attack and subsequent fire).

Technical approach The proposed carbon monitoring program (Figure 2) will be built from three major

components: 1) annual Landsat-based maps and images that capture a wide range of change processes occurring on the landscape; 2) statistical modeling of key carbon-related variables through integration of imagery, spectral trajectories, environmental variables, FIA plot measurements, and empirical allometric equations; and 3) aboveground carbon mapping from small-footprint lidar and concurrent field measurements. We have already developed the core methods for all of these individual components and for much of the proposed integration of methods. Primary effort in this proposal focuses on improvement, integration, and expansion. Uncertainties in each component (module) will be characterized separately, and then combined to produce annual maps of both forest carbon estimates and uncertainties in those estimates. The entire system will be constructed in a modular fashion to allow future incorporation of improved change maps and imagery, new plot data, new allometric equations, and new lidar datasets. Constructing and validating maps under the system is divided into three Tasks.

Page 7: Full project text

Kennedy et al: Integrated, observation based carbon monitoring

Page 6

Task 1: Develop statewide disturbance, growth, and change agent maps for Washington, Oregon, and California from 1985 to 2010. All maps in the system will ultimately be based on outputs from LandTrendr (Landsat-based detection of trends in disturbance and recovery), a set of algorithms developed by the PI to characterize temporal patterns in dense time series Landsat Thematic Mapper (TM) data (Kennedy et al. Accepted). The core algorithms statistically evaluate each TM pixel’s spectral trajectory to identify time periods when consistent trends occur and when events occur that interrupt or change the direction of those trends (Figure 3). These periods of consistency and change are represented as simple straightline segments that remove unwanted year-to-year noise caused by sun-angle, phenological variation, and residual atmospheric effects, while maintaining the salient features of each pixel’s temporal signature. This process of “temporal segmentation” drives all sub-tasks. Sub-

Task 1. Create yearly maps of landscape change for Washington, Oregon, and California.

Landscape change maps derive from the temporal segmentation process. We identify segments of interest and then map their onset year, their direction of spectral change, their duration, or combinations or sequences of these factors. Under the auspices of other projects funded by NASA, the USDA Forest Service, the Department of Energy,

Page 8: Full project text

Kennedy et al: Integrated, observation based carbon monitoring

Page 7

and the National Park Service, we have developed all of the relevant pre-processing and change detection algorithms, and have or will have performed this work for most of the proposed study area by the time this project begins (Figure 1). Importantly, because the algorithm creates on-the-fly mosaics of multiple images per year, we can map with both partly cloudy and Landsat 7 SLC-off images. This will allow continued mapping even after Landsat 5 ceases to function.

We have already implemented large-area mapping for the entire area of the Northwest Forest Plan (NWFP) in western Washington, Oregon, and northwest California (Figure 4). Processes captured in these maps include: stand-replacing harvest, partial harvest, high and low severity fire, bark-beetle and defoliator-related forest mortality, and growth, as well as sequences of these processes (Figure 4b). For this project, we propose to bring all existing scenes up to the 2010 growing season, and to add the 10 remaining scenes needed to cover the study area. Thus, with relatively little new overhead, this project will provide landscape change maps for the entire contiguous western coastal states.

Note: A portion of the change detection work detailed in sub-tasks 1 and 2 is also proposed in two other responses to this same research announcement [Principal Investigator Law and Principal Investigator Gray]. Should more than one of these proposals be funded, overlap can be resolved.

Sub-Task 2 Label maps according to change agent (clearcut vs. thinning, insect vs. fire, growth vs. encroachment, etc.)

Labeling change agent is critical for modeling and management. We have developed and continue to improve approaches to distinguish among anthropogenic and natural processes (Figure 4c). Attribution of change is essentially a patch-wise supervised classification exercise that includes information on patch size and shape to infer agent. The time-series approach brings more information to bear on the process, including mean and variation in change magnitude, duration of change. Additionally, because events leading up to or following a particular landscape change can provide clues about the

Page 9: Full project text

Kennedy et al: Integrated, observation based carbon monitoring

Page 8

cause of the change, the segmentation approach allows us to include information about pre- and post-event trajectories in our predictor variable set.

Sub-Task 3: Create radiometrically stable, temporally-consistent, cloud-free, seamless Landsat image mosaics for every year from 1985 to 2010.

Because the temporal segmentation process largely removes unwanted noise, it can be used to produce Landsat-like “pseudo-images” that incorporate durable change in the landscape, but that have the ephemeral effects of cloudiness, sun-angle variability, phenology, atmospheric variability, etc. removed. These spectrally stable images allow any spectrally based classifier to be defensibly applied to all dates of imagery; in this project, we use them to drive nearest neighbor mapping in Task 2 below (Figure 5)

The segment-smoothed images are produced from the segmentation process conducted for Task 1, and simply require implementation of a single additional batch file. The resultant yearly image mosaics will be produced for the entire three state study area and provided to the community for independent use in any other projects.

Of critical importance for any mapping effort is delineation of the forest area (a forest/non-forest mask). Existing maps from NLCD for 2001 capture only the forest status at a single point in time; we have developed approaches for the NWFP to produce forest/non-forest masks based on the most forested condition in the entire 25-year period of segment-smoothed images. We will extend this approach geographically to the entire study area.

Sub-Task 4: Characterize errors in change maps and segment-smoothed images

For our existing work in the NWFP, we have developed an approach to validation based on direct interpretation of the Landsat imagery using a new tool called TimeSync, followed by opportunistic validation of TimeSync calls with existing databases (Cohen et

Page 10: Full project text

Kennedy et al: Integrated, observation based carbon monitoring

Page 9

al. Accepted). We have already characterized segmentation results in the NWFP and will extend our sample to the entire study area to characterize errors in the change maps.

For annual pseudo-images, will consider the difference between fitted and original images to be composed of undesirable error introduced by LandTrendr and desirable noise reduction of scene-wide phenological, atmospheric, and sun-angle artifacts. To capture the effects of uncertainty, we will produce multiple instances of each segment-smoothed image, each with a slightly different spectral value for each pixel drawn from the cumulative distributions of pixel-level uncertainty for that spectral value.

Task 2: Develop forest carbon maps using nearest-neighbor mapping approaches to link FIA and LandTrendr data for Washington, Oregon, & California

We will use nearest-neighbor imputation (Ohmann and Gregory 2002a) to link FIA plot data to spectral and environmental data. For our integrative work in support of monitoring for the NWFP, we have already developed approaches to feed LandTrendr pseudoimages into nearest-neighbor mapping efforts for two image years (1996 and 2006), and have already developed example maps of aboveground biomass and carbon for those two years (Figure 5). For this project, we will apply this same approach to every year of imagery and develop several planned improvements described below.

Sub-Task 1. Acquire and screen FIA plots for areas not currently being

mapped We have developed maps using the Gradient Nearest Neighbor approach (GNN) for

large areas in Oregon, Washington, and California, and thus have working knowledge of the FIA plot data for much of the area. For this project, we will expand to include remaining all remaining areas in the study area. Yearly imagery will speed plot database construction, as it reduces mismatch of image and plot condition caused by differences in year of measurement. Janet Ohmann (Co-I) oversees this work, currently has access to all FIA plot data in the study area, and works on a team in the USDA Forest Service’s

Page 11: Full project text

Kennedy et al: Integrated, observation based carbon monitoring

Page 10

Pacific Northwest Research Station with FIA analysts that will ensure her ongoing ability to work with these plot data. We will acquire all remaining relevant FIA and FIA-related historical plot data from 1980 to 2010 for all of Washington, Oregon, and California.

Sub-Task 2: Integrate LandTrendr-based dynamics into nearest neighbor predictor variable sets to improve mapping

We have already developed the methods to incorporate LandTrendr pseudo-images into a nearest-neighbor statistical modeling framework to produce aboveground biomass and carbon maps for two years (Figure 5). We will expand this same approach to every other year in the time series, and also include data from LandTrendr on recent spectral trends. This will allow distinction between stands that are spectrally similar at a given point in time, but that have arrived at that state from different starting points.

Sub-Task 3: Calculate aboveground live and dead carbon from FIA plot attributes for the entire mapped area

As shown in Figure 5, we have already developed the tools to create basic carbon maps from nearest-neighbor derived data. For this project, we will combine our existing approach with lessons learned specifically in the implementation of carbon mapping to Landsat time-series (Co-I Powell), exploring different allometric and multivariate mapping approaches in our carbon calculations (Powell et al. 2010). Because our maps will contain the full suite of FIA variables at each pixel, we can calculate for all pixels in our maps any variable that can traditionally be derived from plot data. These include aboveground tree biomass or carbon in live and dead pools. In the future, as plot data mature, our system will allow easy integration of any new plot-derived variables, including potentially belowground carbon. Also, because our carbon stock and change-in-stock values are tied to specific agents of change and to specific locations on the landscape, our system will also allow future incorporation of the fate of carbon removed from the site.

Sub-Task 4: Adding uncertainty Uncertainty in carbon maps comes from at least three sources: 1 ) Errors in spectral

data contained in pseudoimages from LandTrendr, 2) Choice of FIA plot for imputation in the nearest neighbor framework (including spectral ambiguity at high biomass values), and 3) Errors in allometric equations linking plot attributes to carbon. To capture these uncertainties, we will create multiple instances of aboveground carbon by varying all three of these components. Final maps will report median carbon for each period, as well as estimates of the 20th and 80th percentiles in the carbon distributions derived from these multiple instances of the maps.

Task 3: Develop lidar-based carbon surfaces for selected areas The Landsat and FIA-based maps of carbon developed under Task 2 must be applied to large areas (state wide), but to be relevant to the modeling and management communities, they must also be relevant at the local level of individual modeling cells and management actions. Thus, we must assess how well maps produced under Task 2 reflect actual carbon conditions on the ground. Extensive work over the past decade has shown that lidar data can be used to accurately estimate forest carbon stores across a wide range of forest types (Lefsky et al. 2002; Lefsky et al. 2005; Næsset and Gobakken 2008; van Aardt et al. 2008; Asner et al. 2009). We will calculate tree-biomass and carbon

Page 12: Full project text

Kennedy et al: Integrated, observation based carbon monitoring

Page 11

estimates at sites where existing lidar data and appropriate field data are available, and analyze errors in our larger-area maps with the more accurate lidar estimates. As new datasets emerge, both from small-footprint and from space-borne lidar systems, the structure developed in this project will allow easy integration of the new data to improve the overall monitoring system.

Sub-Task 1: Identify and acquire existing, appropriate small-footprint lidar datasets

For this study, we will use existing field and lidar datasets to estimate forest carbon for six study areas representing common forest types and dominant land management practices in the three states to be studied (Table 1). Study sites have been chosen where field and lidar data were collected within two years of each other to provide concurrency. All field sites will have been located using professional GPS units and locations will have been post processed to provide location accuracy with a few meters for most plots.

Table 1: Sites to be used for field- and lidar-based calculations of forest carbon. All data is either in the public domain or made available through study Co-PI Lutz or owner/agency collaborators.

Site / Owner / Collaborator

Forest types / disturbance regimes lidar coverage (date) / no. of field plots

Cedar River Watershed (WA) / City of Seattle /

Gersonde

Westside Cascade Mountains: Western hemlock-Douglas fir; Pacific silver fir; mountain hemlock.

Extensive clear cut harvesting from ~1900 to 1990s; selective thinning for habitat restoration since 2000; chronic small scale wind and pathogen disturbances; 16% of area remains in old-growth

36,700 ha1 (2001) / 115 plots2

Ellsworth Creek (WA) / Nature Conservancy /

Rolf

Pacific Northwest Coastal: Sitka spruce. 7% of area remains in old-growth. Chronic small scale

wind and pathogen disturbances. Clear cut harvesting 1940s-1988; selective thinning of 440 ha since 2007; road decommissioning and revegetation of 12 km of roads since 2006

2439 ha1 (2007) / ~200 plots1, 6

Yosemite (CA) / National Park Service /

Lutz

Westside Sierra: ponderosa pine mixed conifer, white fir mixed conifer, white fir/red fir, red fir.

Well documented history of natural and management fires; chronic small-scale insect outbreaks, documented timber harvest in some forests prior to 1940

9000 ha3 (2010) / ~100 1/10 ha plots2, 25 ha plot3

Deschutes (OR) / U.S. Forest Service /

Maffei

Eastside Cascade Mountains: ponderosa pine, white fir mixed conifer, mountain hemlock, lodgepole pine.

Well documented management treatments including thinning, regeneration harvest and salvage. Significant areas of stands with late old structure are present.

48,000 ha2 (2009-2010) / 240 FIA plots1

Colville (WA) / U.S. Forest Service /

McGaughey

Okanogan Highlands: Mixed conifer Ponderosa pine-Douglas-fir and western redcedar-hemlock.

Extensive management from 1910 to present. History of clear cut harvest from 1910 to 1995 with more selective harvest and patch treatments from 1995 to present. Extensive long-term disturbances from pathogens and fire.

43,000 ha2 (2008) / 72 plots3

BLM Oregon / SW Oregon costal mixed conifer moist site forest and 48,000

Page 13: Full project text

Kennedy et al: Integrated, observation based carbon monitoring

Page 12

McFadden

Klamath-Siskiyou mixed conifer dry site forest. Intensive management including clear cut harvests and

partial cut harvests from 1920 to present. Checkerboard ownership pattern has resulted in dramatic differences in land use and management. Chronic small scale disturbances from wind and pathogens with occasional small and large burns. BLM ownership has more mature stands with some old growth. Other ownerships within the acquisition area have extensive clear cut harvesting and more young forests

ha1 (2008) / 900-1000 plots1 to be sampled in 2010

Data ownership: 1Landowner/agency (available through collaborator); 2Public Domain; 3University of Washington, 6 This study funds survey-grade GPSing of plots at Ellsworth Creek.

Sub-Task 2. Develop carbon surfaces for selected acquisitions The creation of carbon maps will follow the general approach of (Powell et al.2010),

but applied to lidar rather than Landsat time-series data. We will use ancillary data such as forest type, elevation, climate, and geomorphology as predictive variables. For study areas with lidar coverage <100 square kilometers, the entire area of lidar coverage will be included in the study. For areas with greater coverage, a study site of approximately 100 square kilometers will be selected to include forest types common to the area.

Workflow will be identical for each site. After assurance of data quality and consistency, the following sequence will be followed: calculation of standing carbon from field measurements of individual tree using most appropriate allometric equations (Means et al. 1994, Kittredge 1944, Gholz et al. 1979, Westman 1987, Gower et al. 1992, Means et al. 1994, Miller and Urban 1999, Law et al. 2001, Acker et al. 2002, Jenkins et al. 2004); calculation of regressions linking height metrics to field-derived carbon under a variety of multivariate paradigms (Breiman 2001; Cohen et al. 2003; Ohmann and Gregory 2002a) ; and application of the regression to lidar height metrics across the entire dataset to produce biomass maps. Additional maps will be produced showing statistical distance of each pixel from the regression line or plot characteristics, and this information will be used in assessing the overall project error.

Uncertainties in lidar-derived carbon maps will be captured by varying: 1) the form and type of allometric equation linking field measurements to carbon, 2) the plot population used to develop regression equations (bootstrapping), and 3) the manner of developing site-specific regressions. We will produce carbon maps for each of these combinations.

Sub-task 3: Use lidar data to characterize error The goal in this task is to evaluate whether the estimates of uncertainty in

Landsat/FIA derived carbon distributions overlap with the much better estimates of carbon derived from the lidar data. Depending on the degree of overlap and mismatch, the analysis for this task could range from simple characterization of the carbon distribution overlap to more complex approaches to adjust estimates of uncertainty in the Landsat/FIA-based maps. Because we have chosen lidar study areas that encompass a wide range of forest areas and management histories, we will characterize agreement among maps according to forest type, estimated age, and recent management history.

Page 14: Full project text

Kennedy et al: Integrated, observation based carbon monitoring

Page 13

Perceived impact in the field We expect our work to have impacts in three main areas:

1) General science and management communities: Our approach will provide insight into both natural and anthropogenic controls on forest carbon change in an economically and ecologically dynamic region. 2) Carbon modeling community: Our observations will serve as constraints on predictions from models that will eventually be used to predict carbon futures. 3) State and federal carbon monitoring communities: As climate change and carbon become increasingly important components of policy and social frameworks, the agencies tasked with monitoring, adaptation, and mitigation must have robust data on which to both base decisions and track compliance. Our outputs will provide these data.

Particular stakeholders and interested groups Because of the relevance of our proposal to non-profit, state, and federal

management agencies, we have engaged collaborators expressing the need for consistent, cross-ownership tracking of forest carbon at a grain size small enough to capture individual harvest and management events. (Yost, Oregon Department of Forestry; Scott; California Air Resources Board; Rolph, The Nature Conservancy; Maffei, Deschutes National Forest; McFadden, Bureau of Land Management). Similar needs are expressed by carbon modelers (such as Collaborators Harmon and Turner). The PI will engage these collaborators over the course of the project to ensure that our products and methods of relaying our results are useful for management and modeling needs.

Relevance to NASA Our proposal directly addresses several goals of the NASA Earth Science Program:

• Objective 1 addresses carbon cycle science goals by quantifying changes in terrestrial carbon in response to land use, land cover change, and other human and natural events.

• Objective 2 addresses the goals of both the carbon cycle science program and the Applied Sciences program by quantifying impacts and consequences of human management on systems vulnerable to unplanned carbon loss, particularly those that serve multiple needs to society

• Objective 3 addresses the goals of the Applied Sciences program in setting up the understanding needed to implement a society- and management-relevant carbon monitoring system for forests

Our work is also relevant to broader NASA goals, including explicit characterization of uncertainty in products derived from space-based platforms and in providing a flexible framework that can incorporate new input or validation data (such as that from the upcoming LDCM and DESDyni missions).

Finally, the LandTrendr components of this project build in part on prior NASA support from the Earth Sciences program for Kennedy’s 2008 New Investigator Project (“Leveraging temporal variation in climate and management across national parks in the western U.S. to characterize three decades of landscape vegetation dynamics”) and on Dr. Sam Goward’s NACP-related project (“North American Forest Dynamics”).

Page 15: Full project text

Kennedy et al: Integrated, observation based carbon monitoring

Page 14

Plan of work The overall arc of the project progresses through the Tasks described in the

Technical Approach section and then uses the products and lessons learned through those tasks to address the science questions described in the Objectives/Significance section.

Key deliverables from the Tasks section include annual (1985 to present), 30m resolution maps for all forested areas in Washington, Oregon, and California of:

• Forest disturbance year and magnitude • Forest/Non-forest mask • Agent of disturbance: clearcuts, partial cuts, insect mortality, fire, urbanization • Forest growth rate and period • Forest type, size, age • Live and dead aboveground forest carbon pools • Uncertainty estimation based on accumulated errors in monitoring as well as

comparison with small-footprint lidar maps Other important deliverables include robust lidar-based estimates of aboveground

carbon for several study areas and evaluations of best practices for integrating lidar, Landsat, and FIA plot data.

In the Year 3 of the project, we will use these deliverables in a GIS analytical context to evaluate the science questions outlined Objectives 1 and 2 of the proposal above. Additionally, we will document lessons learned and alternative strategies throughout the project and summarize these to address science and management questions in Objective 3.

Data Sharing All map-based deliverables are noted on the project calendar below. These will be

provided on our websites (Kennedy: landtrendr.forestry.oregonstate.edu; Ohmann: www.fsl.orst.edu/~lemma) for public use once validation and documentation are complete. We have a proven record of providing geospatial data to stakeholders and the public under both the NWFP project and other prior projects. Additionally, we will explore new options to disseminate our map results in an interactive map format. An example of this type of online disturbance map for the NWFP can be viewed at http://meridian.forestry.oregonstate.edu/nwfp/ltviewer/ index.html.

Management and contributions of paid personnel: Kennedy (Oregon State University) is overall project lead, responsible for project

direction and implementation, and for timely completion of all tasks. Additionally, he will directly supervise the Post Doc and faculty research assistant (FRA) conducting image processing and analysis, and will supervise the field technicians collecting positional information on plots at the Ellsworth Creek site in July 2012. Finally, he will oversee analysis and integration of all components for publication in peer-reviewed journals in the final year of the project (Total time: 5.5 months). Co-Investigator Ohmann is responsible for completion of the nearest neighbor component of the project, working with both the Post Doc and FRAs. Co-Investigator Cohen is responsible for guiding the FRA on TimeSync interpretation. Both Ohmann and Cohen are Federal employees and will not receive salary under this project. Co-Investigator Franklin (University of Washington) is responsible for oversight of lidar processing and carbon surface mapping (total time: 1.3 months). Kane and Lutz will conduct lidar processing and carbon surface calculations, and will assist in integrating lidar carbon results into the

Page 16: Full project text

Kennedy et al: Integrated, observation based carbon monitoring

Page 15

larger carbon error analysis. (total time: ~21 months, plus 3 months of student hourly time for data management). Co-Investigator Powell (Montana State University) is responsible in Year 1 for assisting the OSU team in developing approaches to improve biomass estimation from FIA plot attributes, and in Year 3 for evaluating and analyzing overall biomass surfaces (total time: 4 months). Total time requested for a postdoc for all above processing is ~9 months, for FRAs ~32 months, and field technicians (6 person months).

Collaborators: We have assembled a group of collaborators to aid us in acquiring data needed for

lidar work, to guide our approaches to result in products useful to non-profit and state agencies, and to test the applicability of our maps in a training/validation context for carbon modeling. Collaborator Role Rolph, McFadden, Maffei, McGaughey

Provision of field measurements and/or lidar acquisitions in study areas (See Table 1 on lidar datasets)

Bigley, Pierce, Scott, Yost

Advise and assess relevance of disturbance, carbon, and forest type maps for state mandated greenhouse gas and wildlife goals

Harmon, Turner Advise and assess relevance of disturbance and carbon maps in carbon modeling framework

Project Calendar

Task* Detail Q1

2011

Q2

2011

Q3

2011

Q4

2011

Q1

2012

Q2

2012

Q3

2012

Q4

2012

Q1

2013

Q2

2013

Q3

2013

Q4

2013

Task 1 LandTrendr dynamics maps X X D**

Labeling and attribution X X X D**

Pseudo-images X X X

TimeSync error estimation X X

Task 2 Acquire/Screen plots X X

LandTrendr / NN integration X X X D

Carbon estimation X X X X X X DUncertainty modeling X X X

Task 3 Data acquisition/screening X X

GPS Field Plots X

Lidar-biomass X X X X DLink lidar / Landsat uncertainty analysi X X X X

Object-ives Geographic/temporal analysis in support of science questions X X D

**D Corresponds to expected date of deliverable

* Tasks detailed in techincal approach section, science questions described in Objectives/Significance Section

Page 17: Full project text

Kennedy et al: Integrated, observation based carbon monitoring

Page 16

References and Citations Birdsey, R., Pregitzer, K., & Lucier, A. (2006). Forest carbon management in the

United States: 1600-2100. Journal of Environmental Quality, 35, 1461-1469 Birdsey, R.A., Jenkins, J.C., Johnston, M., Huber-Sannwald, E., Amero, B., de John,

B., Barra, J.D.E., French, N., Garcia-Oliva, F., Harmon, M., Heath, L.S., Jaramillo, V.J., Johnsen, K., Law, B.E., Marin-Spiotta, E., Masera, O., Neilson, R., Pan, Y., & Pregitzer, K.S. (2007). North American Forests. In A.W. King, L. Dilling, G.P. Zimmerman, D.M. Fairman, R.A. Houghton, G. Marland, A.Z. Rose & T.J. Wilbanks (Eds.), The First State of the Carbon Cycle Report (SOCCR): The North American Carbon Budget and Implications for the Global Carbon Cycle (pp. 117-126). Asheville, N.C.: National Oceanic and Atmospheric Administration, National Climatic Data Center

Blackard, J.A., Finco, M.V., Helmer, E., Holden, G.R., Hoppus, M.L., Jacobs, D.M., Lister, A.J., Moisen, G.G., Nelson, M.D., Riemann, R., Ruefenacht, B., Salajanu, D., Weyermann, D., Winterberger, K., Brandeis, T.J., Czaplewski, R., McRoberts, R.E., Patterson, P.L., & Tymcio, R.P. (2008). Mapping U.S. forest biomass using nationwide forest inventory data and moderate resolution information. Remote Sensing of Environment, 112, 1658-1677

Breiman, L. (2001). Random forests. Machine Learning, 45, 5-32 Chambers, J.Q., Fisher, J.I., Zeng, H., Chapman, E.L., Baker, D.B., & Hurtt, G.C.

(2007). Hurrican Katrina's carbon footprint on U.S. Gulf Coast Forests. Science, 318, 1107

Cohen, W.B., & Fiorella, M. (1998). Comparison of methods for detecting conifer forest change with Thematic Mapper imagery. In R.S. Lunetta & C.D. Elvidge (Eds.), Remote sensing change detection: environmental monitoring methods and applications (pp. 89-102). Chelsea, MI: Ann Arbor Press

Cohen, W.B., Fiorella, M., Gray, J., Helmer, E.H., & Anderson, K. (1998). An efficient and accurate method for mapping forest clearcuts in the Pacific Northwest using Landsat imagery. Photogrammetric Engineering & Remote Sensing, 64, 293-300

Cohen, W.B., Harmon, M.E., Wallin, D.O., & Fiorella, M. (1996). Two decades of carbon flux from forests of the Pacific Northwest. BioScience, 46, 836-844

Cohen, W.B., Maiersperger, T.K., Gower, S.T., & Turner, D.P. (2003). An improved strategy for regression of biophysical variables and Landsat ETM+ data. Remote Sensing of Environment, 84, 561-571

Cohen, W.B., Spies, T., Alig, R.J., Oetter, D.R., Maiersperger, T.K., & Fiorella, M. (2002). Characterizing 23 years (1972-95) of stand replacement disturbance in western Oregon forests with Landsat imagery. Ecosystems, 5, 122-137

Cohen, W.B., Spies, T.A., Swanson, F.J., & Wallin, D.O. (1995). Land cover on the western slopes of the central Oregon Cascade Range. International Journal of Remote Sensing, 16, 595-596

Cohen, W.B., Zhiqiang, Y., & Kennedy, R.E. (Accepted). Detecting Trends in Forest Disturbance and Recovery using Yearly Landsat Time Series: 2. TimeSync - Tools for Calibration and Validation. Remote Sensing of Environment

Environment-Canada (2006). National Inventory Report 1990-2004: Greenhouse Gas Sources and Sinks in Canada, Greenhouse Gas Division, Environment Canada, Ottowa, Ontario Available at http://www.cc.gc.ca/pdb/ghg/inventory_

report/2004_report/toc_e.cfm

Page 18: Full project text

Kennedy et al: Integrated, observation based carbon monitoring

Page 17

Goetz, S.J., Baccini, A., Laporte, N.T., Johns, T., Walker, W., Kellndorfer, J., Houghton, R.A., & Sun, M. (2009). Mapping and monitoring carbon stocks with satellite observations: a comparison of methods. Carbon balance and management, 4, doi:10.1186/1750-0680-1184-1182

Goodale, C.L., Apps, M.J., Birdsey, R.A., Field, C.B., Heath, L.S., Houghton, R.A., Jenkins, J.C., Kohlmaier, G.H., Kurz, W., Liu, S.R., Nabuurs, G.J., Nilsson, S., & Shvidenko, A.Z. (2002). Forest carbon sinks in the Northern Hemisphere. Ecological Applications, 12, 891-899

Houghton, R.A. (2005). Aboveground forest biomass and the global carbon balance. Global Change Biology, 11, 945-958

Huang, C., Goward, S.N., Masek, J.G., Thomas, N., Zhu, Z., & Vogelmann, J.E. (2010). An automated approach for reconstructing recent forest disturbance history using dense Landsat time series stacks. Remote Sensing of Environment, 114, 183-198

Joyce, L.A., Blate, G.M., McNulty, S.G., Millar, C.I., Moser, S., Neilson, R.P., & Peterson, D.L. (2009). Managing for multiple resources under climate change: National forests. Environmental Management, 44, 1022-1032

Kane, V.R., Bakker, J.D., McGaughey, R.E., Lutz, J.A., Gersonde, R., & Franklin, J.F. (2010). Examining conifer canopy structural complexity across forest ages and elevations with LiDAR data. Canadian Journal of Forest Research, 40, 774-787

Kane, V.R., Gillespie, A.R., McGaughey, R.E., Lutz, J.A., Ceder, K., & Franklin, J.F. (2008). Interpretation and topographic compensation of conifer canopy self-shadowing. Remote Sensing of Environment, 112, 3820-3832

Kane, V.R., McGaughey, R.E., Bakker, J.D., Gersonde, R., Lutz, J.A., & Franklin, J.F. (2010). Comparisons between field- and LiDAR-based measures of stand structural complexity. Canadian Journal of Forest Research, 40, 761-773

Kellndorfer, J.M., Walker, W.S., Pierce, L.E., Dobson, M.C., Fites, J., & Hunsaker, C. (2004). Vegetation height derivation from Shuttle Radar Topography Mission and National Elevation data sets. Remote Sensing of Environment, 93, 339-358

Kennedy, R.E., Cohen, W.B., & Schroeder, T.A. (2007). Trajectory-based change detection for automated characterization of forest disturbance dynamics. Remote Sensing of Environment, 110, 370-386

Kennedy, R.E., Yang, Z., & Cohen, W.B. (Accepted). Detecting trends in forest disturbance and recovery using yearly Landsat time series: 1. LandTrendr - Temporal segmentation algorithms. Remote Sensing of Environment

Kim, Y., Yang, Z., Cohen, W.B., Pflugmacher, D., Lauver, C.L., & Vankat, J.L. (2009). Distinguishing between live and dead standing tree biomass on the North Rim of Grand Canyon National Park, USA using small-footprint lidar data. Remote Sensing of Environment, 113, 2499-2510

Lefsky, M.A., Cohen, W.B., Harding, D.J., Parker, G.G., Acker, S.A., & Gower, S.T. (2002). Lidar remote sensing of above-ground biomass in three biomes. Global Ecology & Biogeography, 11, 393-399

Naesset, E. (2002). Predicting forest stand characteristics with airborne scanning laser using a practical two-stage procedure and field data. Remote Sensing of Environment, 80, 88-99

Page 19: Full project text

Kennedy et al: Integrated, observation based carbon monitoring

Page 18

Ohmann, J.L., & Gregory, M.J. (2002a). 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, 724-741

Ohmann, J.L., & Gregory, M.J. (2002b). Predictive mapping of forest composition and structure with direct gradient analysis and nearest-neighbor imputation in coastal Oregon, USA. Canadian Journal of Forest Research, 32, 725-741

Ohmann, J.L., Gregory, M.J., & Spies, T.A. (2007). Influence of environment, disturbance, and ownership on forest vegetation of Coastal Oregon. Ecological Applications, 17, 18-33

Ohmann, J.L., & Spies, T. (1998). Regional gradient analysis and spatial pattern of woody plant communities of Oregon forests. Ecological Monographs, 68, 151-182

Pierce, K.B.J., Ohman, J.L., Wimberly, M.C., Gregory, M.J., & Fried, J.S. (in press). Mapping wildland fuels and forest structure for land management: a comparison of nearest-neighbor imputation and other methods. Canadian Journal of Forest Research

Powell, S.L., Cohen, W.B., Healey, S.P., Kennedy, R.E., Moisen, G.G., Pierce, K.B., & Ohman, J.L. (2010). Quantification of live aboveground forest biomass dynamics with Landsat time-series and field inventory data: A comparison of empirical modeling approaches. Remote Sensing of Environment, 114, 1053-1068

Powell, S.L., Cohen, W.B., Yang, Z., Pierce, J.D., & Alberti, M. (2008). Quantification of impervious surface in the Snohomish Water Resources Inventory Area of Western Washington from 1972-2006. Remote Sensing of Environment, 112, 1895-1908

Woodbury, P.B., Smith, J.E., & Heath, L.S. (2007). Carbon sequestration in the U.S. forest sector from 1990 to 2010. Forest Ecology and Management, 241, 14-27

Zheng, D.L., Rademacher, J., Chen, J.Q., Crow, T., Bresee, M., le Moine, J., & Ryu, S.R. (2004). Estimating aboveground biomass using Landsat 7 ETM+ data across a managed landscape in northern Wisconsin, USA. Remote Sensing of Environment, 93, 402-411