Integrating time series of Landsat-based information into FIA's estimation process RMRS: Gretchen Moisen, Todd Schroeder, Sean Healey, Ray Czaplewski PNW: Warren B. Cohen WO: Ken Brewer UMD: Sam Goward, Karen Schleeweis FIA Nat’l User Group Meeting— 7-8 March2012 1
38
Embed
Integrating time series of Landsat-based information into FIA's estimation process RMRS: Gretchen Moisen, Todd Schroeder, Sean Healey, Ray Czaplewski PNW:
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
1
Integrating time series of Landsat-based information into FIA's estimation process
RMRS: Gretchen Moisen, Todd Schroeder, Sean Healey, Ray CzaplewskiPNW: Warren B. Cohen
WO: Ken BrewerUMD: Sam Goward, Karen Schleeweis
FIA Nat’l User Group Meeting— 7-8 March2012
2
Status: How much is out there now, ….and where is it?
Change: What just happened?
Trend: What’s happening?
Some Simple Questions
Status:
Change:
Trend:
How’s FIA Doing?
A-
I
A …for effortI …for accomplishment
Outline
3. NAFD Phase 3
4. How can we integrate Landsat time series into FIA’s estimation processes?
1. Forest disturbance and monitoring
2. History of the North American Forest
Dynamics (NAFD) Project
• Impacts ~ 1-3% of a forest area per year• Occurs at different spatial scales, temporal scales, and intensities• Can impact canopy, understory and forest floor • Climate change and growing human population may alter the
frequency and severity of future disturbance regimes• Monitoring has taken on renewed importance
FireClearcut
1987
1989
1990
1991
1993
1994
1995
1997
1998
1999
2001
2002
2004
2006
2008
0
0.5
1
1.5
2
0
2
4
6
8
10
12ClearcutFire
Year
Ann
ual R
ate
of C
lear
cutti
ng
Annual Rate of Fire
Spatial Temporal
Forest Disturbance
6
Disturbance and Time
(Brewer, 2009)
7
Disturbance and Space
(Brewer, 2009)
Monitoring Through Plots
Unbiased estimates at broad scales
Sampling error is well understood
Measurement error can be assumed to be negligible for many variables
Results are not spatially explicit at the local level
Revisit frequency may miss disturbance events
Post-dating is problematic
Difficult to see upper canopy disturbances from the ground
Monitoring Through Landsat Time Series
16-day repeat cycle and 40-year historical archive allows development of dense image times series which can be used to detect changes in forest cover over large areas.
Spatial grain (30m) and variety of spectral bands allows detection and causal attribution of most natural and anthropogenic disturbances.
Can be used for mapping forest change and for collecting human interpreted reference data (e.g. Timesync).
There is no sampling error BUT measurement error is variable and often poorly understood.
Different monitoring methods are appropriate for different purposes
Joining traditional forest inventory data with temporally dense satellite data results in new information for monitoring change and trend
Outline
3. NAFD Phase 3
4. How can we integrate Landsat time series into FIA’s estimation processes?
1. Forest disturbance and monitoring
2. History of the North American Forest
Dynamics (NAFD) Project
North American Forest Dynamics (NAFD)(UMD, NASA-Goddard, FIA, PNW, NRS, CFS, CONAFOR)
• NASA-funded project designed to characterize disturbance patterns and recovery rates of forests across the continent
• Goal: Determine the role of forest dynamics in North American carbon balance
Phase I & II Sample Sites
Eastern Stratum
Western Stratum
Phase IPhase II
Phase IPhase II
Processed time series (1985-2008) of Landsat satellite imagery using FIA inventory data for validation and training
Vegetation Change Tracker
Year Disturbed
1985
1986
1987
1988
1989
1990
1991
1992
1993
1994
1995
1996
1997
1998
1999
2000
2001
Forest
Nonforest
Lake Anna, VA, 60 km NW of Richmond, VA
Major disturbance
0
5
10
15
20
25
1985 1990 1995 2000
Year
FI
Minor disturbance
0
5
10
15
20
25
1985 1990 1995 2000
Year
FI
(Huang et al. 2006, 2008)
NAFD “Science” (NASA, PNW, UMD, CONAFOR, CFS, and others)
Characterizing disturbance and regrowth patterns on US forests by analyzing a biennial time series of Landsat imagery over a sample of Landsat data cubes spread across US forests. Objectives include:
1. Produce nationwide estimates of forest dynamics for NACP2. Convert data cube reflectance to data cube biomass3. Develop nationwide maps predictions of forest dynamics4. Begin trials in Canada and Mexico5. Quantify forest component of woody encroachment nationally
NAFD “Applications” (NASA, PNW, UMD, all FIA units)
Illustrate how FIA data can be combined with temporal disturbance and biomass products to answer management questions relevant to FIA users. • Developed FIA monitoring
products that take advantage of satellite-derived disturbance and biomass data (storm-related loss, harvest rates across time and ownerships, fragmentation, carbon considerations)
• Note studies by Sean Healey, Mark Nelson, Randy Morin, Hobie Perry, Andy Lister, John Coulston, and others
Detour: A Model for Collaboration
• Pre-proposal communication with FIA• Engagement of FIA scientists and managers• Common problem identification• Memorandum of Understanding• Sensitivity to logistical and political constraints• Patience
Outline
3. NAFD Phase 3
4. How can we integrate Landsat time series into FIA’s estimation processes?
Ten sample scenes were identified as good candidates for testing, representing a range of causal agents and varying forest types and prevalence.
Outline
3. NAFD Phase 3
4. How can we integrate Landsat time series into FIA’s estimation processes?
1. Forest disturbance and monitoring
2. History of the North American Forest
Dynamics (NAFD) Project
Post-stratification
Disturbance Year Disturbance Type
28
Alternatives to the Moving Average0
1
0 5 10 15Year
true trend = 50% between years 6-7
(Czaplewski, 2008)
29
Endogenous Post-stratification(Breidt and Opsomer 2008; Dahlke et. al In Press, Tipton et. Al In Prep)
Using FIA as training data to make maps
Then using those maps to post-stratification that same FIA data
Mapping Plot Attributes Through Time
31
Strategic Timing of Ground Observation
Before After
AREBA: Accelerated Remeasurement and Evaluation of Burned Areas (RSAC 2009)
1994 2000 2002
Trend
Anomaly
LandTrendr (Kennedy et al.)TimeSync (Cohen et al.)
33
Plot History: Clearcut and recovery
Andy Gray
34
Plot History: Defoliation, delayed mortality, recovery, salvage, and recovery
Andy Gray
Very few disturbances are detected by both Timesync and FIA.
FIA records lots of disturbances which are undetectable by Landsat (e.g. animal damage).
82% of disturbances detected only by Timesync fall outside FIA’s observation window (i.e. disturbance date is > 5 yrs before or is after plot measurement date).
Disturbance is less common thus overall accuracy is inflated by high proportion of undisturbed plots.
A Utah Example: Comparing FIA and Timesync Observations of Disturbance
(Schroeder et. al, In Prep.)
36
Integrating Landsat time series into FIA’s estimation processes?
Integrating Landsat time series into FIA’s estimation processes?
1. Make best use of (endogenous) post-stratification2. Incorporate Landsat (photo)-based “observations” on
field plots3. Consider alternative sampling frequency for disturbed
strata 4. Develop alternatives to the MA and make best use of
RS data through model-assisted or model-based methods
5. Ensure compatibility between status maps and status estimates
6. Ensure compatibility in maps through time7. Explore ways to reduce costs through these processes
38FIA Nat’l User Group Meeting— 7-8 March2012
Third phase of NAFD is providing annual maps of forest disturbance along with attribution, validation, and re-growth analyses nationwide
We need to keep pushing our statistical tools beyond post-stratification and moving average into more integrated ground and RS approaches