Mapping Forest Disturbance.

advertisement
Action: Map forest disturbance (harvest, insect
damage, storm damage, fire, etc) across a
Landsat frame (timing, intensity, location) for 19842005
Biomass
Atmospheric
source
Regrowth Dominated
Disturbance rate ->
Forest age ->
Carbon Flux
Disturbance Dominated
Balanced
area histograms
Why? Interested in
knowing forest
carbon flux for North
America, and how it
is changing
Carbon Consequences
Biologic
C Flux
Spatial Sample of US Disturbance
~25 Sample Sites
Annual or Biennial
Image
Time Series
(1972-2004)
Disturbance
history / stand age
I.
Search for leaf-on, cloud-free annual
images for a given location; optimize
scene selection for anniversary dates.
Need to balance:
- cloud avoidance
- seasonality
- data quality (e.g. L7 vs L5)
- year selection
II. Download/buy data (Landsat-7 data now
free; Landsat-5 by end of year)
III. Orthorectify images to map base using GLS
data set and SRTM digital topography; clip
images to common X-Y spatial domain
- image-image registration essential;
orthorectification desirable (facilitates GIS
integration with other data)
IV. Calibrate each image to radiance, and apply
atmospheric correction to convert images to
surface reflectance
- not essential for change detection, but useful
for additional studies (e.g. integration with other
sensor data; canopy reflectance modeling)
Orthorectified (2)
Landsat 1Gs (2)
Atmospheric Correction
1990’s Landsat-5 mosaic
TOA reflectance
Surface reflectance
100 km
BOREAS Study Region
100
100km
km
V.
Find a sample of “mature” forest in each scene
either by visual inspection or automated
histogram thresholding; obtain mean reflectance
and standard deviation for this set of pixels, for
each scene
VI. Calculate per-pixel “Forestness Index” for each
reflectance image according to:
FI = 1/n * S [(ri – rmf_mean) / rmf_stdev]
Where ri is the reflectance for band i (from 1,n), and
rmf_mean, _stdev are the mean and standard
deviations of the mature forest band for that
image.
Time Trajectories of Forestness Index
Indicate Forest Dynamics
(b) Disturbance
Forestness index
(a) Permanent forest
Year index (19xx)
10
10
8
8
6
6
4
4
2
2
0
0
85 87 89 91 93 95 97 99
(c) Thinning
85 87 89 91 93 95 97 99
(d) Aforestation
(e) Permanent non-forest
10
10
10
8
8
8
6
6
6
4
4
4
2
2
2
0
0
0
85 87 89 91 93 95 97 99
85 87 89 91 93 95 97 99
85 87 89 91 93 95 97 99
VII Use rule-based system and thresholds to
identify disturbance events (e.g. “if FI
increases > 8 and stays >6 for at least three
consecutive years, then mark Year 1 as
disturbance”).
- agriculture identified by frequent large
changes in FI value
- clouds can be identified as large “single event”
changes in FI
VIII. Filter maps using sieve filter to remove
speckle (single pixel changes).
Fig. 1. (a) Location of
samples selected across the
conterminous United States,
where biennial time
series stacks of Landsat
images (LTSS) were acquired
and analyzed to map forest
disturbance
over the past three decades.
The background forest group
map shows that most of the
forest
groups in the United States
have been represented by the
samples. (b) An example
disturbance
map developed using the
LTSS in a 28.5-kilometersquare area south of Lake
Moultrie in South
Carolina. Persisting forest,
nonforest, and water are
shown in green, gray, and
blue, respectively. All
other colors represent
changes mapped in different
years. (c) Percent of forest
land disturbed annually,
calculated according to the
derived disturbance map for
the entire South Carolina
Landsat scene.
Download