USING TIME-SERIES APPROACHES TO - Landsat 7 · 2015-06-02 · USING TIME-SERIES APPROACHES TO...

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