View
5
Download
0
Category
Preview:
Citation preview
USING TIME-SERIES APPROACHES TO
IMPROVE LANDSAT’S CHARACTERIZATION OF
LANDSCAPE DYNAMICS
Project update, October 2013, by Kennedy.
Co-Author: John Rogan, Clark University
Team: Andréfouët, Fraser, Gómez, Hais, Helmer, Hostert,
Pflugmacher, Griffiths, Main-Knorn, Phinn, Scarth, Sonnenschein
LST Oct 2013 Sioux Falls 1
Landsat
Standard
Products
Motivation
Geometry
Clouds
Surface
Reflectance Status
Use
Dynamics
LST Oct 2013 Sioux Falls 2
Valid
Consistent
Generalizable
Valid
Consistent
Generalizable
LST Oct 2013 Sioux Falls 3
Overall workflow
LST Oct 2013 Sioux Falls 4
Landsat
images
Maps of trends
and events
Yearly maps
of state /
condition
Classification
Vertex
maps
Simplifying &
Filtering Temporal
smoothing
Temporally-
smoothed
imagery
Existing
landcover map
Pre-processing Data stack
LandTrendr
temporal
segmentation
Attribution
Human
interpretation
Maps of change
processes
Maps of change
processes
Yearly maps
of state /
condition
LST Oct 2013 Sioux Falls 5
Consistent
Generalizable
NBR: (NIR-
SWIR2)/(NIR+SWIR2)
Ledaps + Fmask: Good!
LST Oct 2013 Sioux Falls 6
Year of
Disturbance
LST Oct 2013 Sioux Falls 7
Generalizing
Use MapReduce framework and new
Python/GDAL coding
General framework will be useful for
any pixel-based time series approach
LST Oct 2013 Sioux Falls 8
LST Oct 2013 Sioux Falls 9
Year of disturbance: Abrupt
LST Oct 2013 Sioux Falls 10
LST Oct 2013 Sioux Falls 11
LST Oct 2013 Sioux Falls 12
Slic
e R
ate
of
Ch
an
ge
“History” image
LST Oct 2013 Sioux Falls 13
TC B
rig
htn
ess
TC
Gre
en
ne
ss
TC W
etn
ess
Define patch
LST Oct 2013 Sioux Falls 14
Interpret patch
Fluvial erosion
Balsam woolly
adelgid
Avalanche
Clearcut
Windthrow
Fire
Agriculture
Development
LST Oct 2013 Sioux Falls 15
Machine learning
LST Oct 2013 Sioux Falls 16
Overall workflow
LST Oct 2013 Sioux Falls 17
Landsat
images
Maps of trends
and events
Yearly maps
of state /
condition
Classification
Vertex
maps
Simplifying &
Filtering Temporal
smoothing
Temporally-
smoothed
imagery
Existing
landcover map
Pre-processing Data stack
LandTrendr
temporal
segmentation
Attribution
Human
interpretation
Maps of change
processes
Maps of change
processes
Yearly maps
of state /
condition
Yearly land cover mapping
LST 12/12/12 18
LandTrendr
Airphoto
C-CAP
White arrows
show
development
The only thing changing is the
spectral value over time
LST Oct 2013 Sioux Falls
0%
10%
20%
30%
40%
50%
60%
70%
80%
90%
100%
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
Pe
rce
nt b
y t
y
pe
Time S
i
nce D
i
sturbance
Forest M
a
nag ement
Herbaceous
Evergreeen Forest
Deciduous Forest
Barren L
a
nd
Developed
Perennial S
n
o w/Ice
Open Water
0%
10%
20%
30%
40%
50%
60%
70%
80%
90%
100%
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
Pe
rce
nta
ge b
y t
y
pe
Time S
i
nce D
i
sturbance
Increasing Urban
Herbaceous
Evergreeen Forest
Deciduous Forest
Barren L
a
nd
Developed
Perennial S
n
o w/Ice
Open Water
a)
b)
Figure 25. Land cover progression for patches labeled in the attribution phase as (a) forest management and (b) increasing urban. Forest management largely begins as coniferous forest and ends as either coniferous or deciduous forest, with barren and developed classes ref ecting brief periods of complete loss of vegetation cover. Patches modeled as increasing urban do sometimes get labeled as urban, but also include some propor-tion of vegetated classes. As noted in Figure 17, these are likely related to suburban, low-density development with associated lawns and trees.
Landsat
Standard
Products
Motivation
Geometry
Clouds
Surface
Reflectance Status
Use
Dynamics
LST Oct 2013 Sioux Falls 20
LST Oct 2013 Sioux Falls 21
LST Oct 2013 Sioux Falls 22
We will be forced to articulate
these for the temporal domain!
LST Oct 2013 Sioux Falls 23
* Landscape Change Monitoring System –
USGS and USFS
LST Oct 2013 Sioux Falls 24
Single image normalization: L5 vs. OLI
LST Oct 2013 Sioux Falls 25
Blue Green Red
NIR SWIR 1 SWIR 2
OLI
L
5
LST Oct 2013 Sioux Falls 26
NBR* From OLI NBR From L5
* NBR: (NIR-
SWIR2)/(NIR+SWIR2)
LST Oct 2013 Sioux Falls 27
NBR From OLI NBR From L5
LST Oct 2013 Sioux Falls 28
LST Oct 2013 Sioux Falls 29
LST Oct 2013 Sioux Falls 30
Consistent
Recommended