Interannual multitemporalapplications of Landsat to forest

ecosystem monitoring and management

Randolph H. Wynne

Department of ForestryVirginia Polytechnic Institute and State University

wynne@vt.edu

Key Co-Is & Collaborators

• Chris Potter & Rama Nemani, NASA Ames• Christine Blinn, Valquiria Quirino, Susmita Sen,

Jess Walker, Tom Fox, Carl Zipper, & John Seiler, Virginia Tech

• Feng Gao & Jeff Masek, NASA Goddard• Sam Goward, UMD• Jason Moan, NC Division of Forestry• Buck Kline, VA Department of Forestry• Jim Ellenwood, USDA Forest Service

Overall Objective

• R&D in support of increased utilization of Landsat data for forest science and management, with particular focus on US Southeast, including forest industry

Key Themes

• Explicitly multitemporal

• Ties with rest of LDCM team

• Collaborators across agencies & organizations (coops, NGOs)

• Applied orientation

Update Summary

• IGSCR in support of USDA Forest Service FIA Phase I in areas of rapid change (mid-NLCD cycle, indiv. scenes or mid-decadal)– Imagine automation complete (Musy et al. 2006;

ASPRS Leica paper award)– Shared memory parallelization complete and

publicly available (Phillips et al. 2007, 2008)– Parlaying this success into NASA AISR proposal

• REDD pilot on 20,000 ha in SA• DSM pilots in Mojave and WV underway

DSM Involvement(Blinn)

• Landsat & ASTER mosaics

• Band ratio development (TM, ASTER, other)

• Regression tree modeling

• Model design and development

• Spatial statistics

• Temperature modeling

Leaf-off Mosaic

Leaf-on Mosaic

Spring Mosaic

• Remote Sensing for Forest Carbon Management– Landsat downscaling prototyped using STARFM for

three scenes; pine age class evaluation of algorithmic performance

– Downscaled product being used for reducing variance in FS models

– Regionwide CO2 efflux and LAI acquired at network of monospecific sites with alternative forest management strategies

– Assessment of VI’s empirical relationship with LAI and stand growth within and across years

– Interface with Ames LAI/fPAR team– CASA modeling with Landsat

Decision Support for Forest Carbon Management: From Research to OperationsMODELS

ESE NASA-CASAGYC PTAEDA 3.1 FASTLOBUSDA Forest Service FORCARB

ESE MISSIONS Aqua Terra Landsat 7 ASTER

Analysis Projects IGBP-GCTE IGBP-LUCC USDA-FS FIA USDA-FS FHM

Ancillary Data SPOT AVHRR NDVI Forest inventory data VEMAP climate data SRTM topographic data

DECISION SUPPORT:

Current DSTs•COLE (county-scale)•LobDST (stand-scale)

• Growth and yield• Product output• Financial evaluations

•CQUEST (1 km pixels)• Ecosystem carbon

pools (g C/m2)• Partitioned NPP (g C/m2/yr)• NEP (g C/m2/yr)

Linked DSTs and Common Prediction

Framework (multiscale)

• Growth• Yield• Product output• Ecosystem carbon pools• Partitioned NPP • NEP• Total C sequestration• Forecasts and scenarios

Information Products, Predictions, and Data

from NASA ESEMissions:

- MODAGAGG- MOD 12Q1- MOD 13- MOD 15A2- ETM+ Level 1 WRS- AST L1B and 07

VALUE & BENEFITS

Improve the rate of C sequestration in managed forests

Decrease the cost of forest carbon monitoring and management

Potentially slow the rate of atmospheric CO2 increase

Enhance forest soil quality

Inputs Outputs Outcomes Impacts

É

É

É

É

É

Ý

b

b

b

b

19BBE1Kp

58DBP6L

34SCW3Kp35RBW3Kp

25RBE2Kp

26RBE3aKp

34SCW3Kp

34SCW3Kp

26BBE2Kp

34SCW4Kp+

19LBW3Kp

35SEW3Kp25SCW3Kp+

35SEW3Kp

58DBP6L

26SCE2Kp

35SEW3Kp

23SBW3Kp

25SEW3Kp35SEW3Kp

23SCW3Kp

19SBW2Kp23LBW3Kp

23SCW3Kp58DBP6L 35LBW3Kp+

58DBP6L

23SBW3Kp

23SCW3Kp

23SCW3Kp

1:12118

Map: 24004ALat: 32° 10.5Long: 84° 37.9

Sand Pine Guidelines

Sand Pine - Well Suited

Sand Pine - Mod. Well Suited

Not Suited for Sand Pine

É

É

É

É

É

Ý

b

b

b

b

Sand PineLAI 2.4

No Fert LobLAI 1.3

RW18 Study AreaLAI 1.7

96PL 1

34BH16

96PL 182BH16

86PL 1

86PL 1

79NP 4

82BH16

82BH16

96PL14

82BH16 34BH1686PL 182BH16

34BH16

96PL 1

34BH1634BH16 34BH16

37BH16

99PL 137BH16

99PL 137BH16

1:12118

Map: 24004ALat: 32° 10.5Long: 84° 37.9

Dec_2004

0.333 - 0.667

0.668 - 1

1.001 - 1.333

1.334 - 1.667

1.668 - 2

2.001 - 2.333

2.336 - 2.667

2.668 - 3

3.001 - 4

No Data

LandSat Image

LAI (Leaf Area Index):•stand stratification for inventory•identification of poor-performing stands for early harvest•identification of stands with high levels of competition

LAI plus GE (Growth Efficiency)Provides ability to estimate stand-level response to silviculture: (fertilization, release, tillage)

Growth = 7.2 tonsGrowth = 3.9 tons

Growth = 5.1 tons

A Regionwide Evaluation of A Regionwide Evaluation of Vegetation Indices for the Vegetation Indices for the

Prediction of Leaf Area Index in Prediction of Leaf Area Index in the Southeastthe Southeast

Study AreaStudy Area

17 path/row combinations

LAI and VI Relationships

Fertilization Effect on LAI and VI Relationship

15 plots from 8 stands in three L5 images acquired either April 11 or May 2, 2006

SR

LAI

6.56.05.55.04.54.03.53.02.5

4

3

2

1

0

-9901

FERT

LAI vs SR by Fertilization

STARFM NDVI Downscaling(Quirino, Gao)

Predicted Landsat Composites (left images) and Landat Images Used for Their Validation (right images): (a) Sample 1, (b) Sample 2, (c) Sample 3.

(a) (b) (c)

STARFM for Pines

Part of the Image Included Sample 1

Sample 2

Sample 3

Mean

Entire Image 65.8 52.5 79.2 65.8 Pine Area 46.4 49.3 56.1 50.6 Pines between 0 and 5 years 60.5 48.8 85.4 64.9 Pines between 5 and 10 years 67.4 72.8 77.1 72.4 Pines between 10 and 15 years 113.0 132.4 119.0 121.5 Pines older than 15 years 150.4 184.2 172.0 168.9

Part of the Image Included Sample 1

Sample 2

Sample 3

Mean

Entire Image 10.8 2.7 4.4 6.0 Pine Area 3.4 2.1 0.1 1.9 Pines between 0 and 5 years 4.5 -0.4 1.1 1.7 Pines between 5 and 10 years 3.2 2.2 -0.2 1.7 Pines between 10 and 15 years 3.4 2.7 -0.1 2.0 Pines older than 15 years 2.4 2.6 0.1 1.7

Results Good Enough to Move Toward FS Lag Analysis

Average Correlation Between Predicted and Daily Landsat NDVI

0.82

0.86

0.80

0.76

0.58

0.41

Entire Study Area

Pine Area

Pine Age Class 1- 0 to 4.9 years

Pine Age Class 2- 5 to 9.9 years

Pine Age Class 3- 10 to 14.9 years

Pine Age Class 4- older than 15years

Av

era

ge

R-s

qu

are

Retrospective Analysis of Growth and Yield (Blinn,

Goward, Masek)

52 sites fall inside more than 1 Landsat scene

Retrospective Analysis cont...

Available Imagery: 5 Landsat scene stacks were provided by the University of Maryland (UMD) courtesy of Sam Goward & Chengquan Huang

Each scene stack contained a time series of at least 12 images from 1984 through 2005 (at least 1 image every other year)

Scene stacks were already: coregistered (aligned with each other) converted to surface reflectance

Image Date

Sim

ple

Rat

io (

SR)

200019981996199419921990198819861984

20

15

10

5

TOA RefSurface Ref

Variable

SeptMay

JulOctSept

Aug

OctOctSept

OctSept

SeptOct

OctSept

Aug

Sept

May

Jul

Oct

Sept

Aug

OctOct

Sept

Oct

Sept

Sept

Oct

Oct

Sept

Aug

Plot 2109 SR Comparison

13 years old in 1984

Summary Conversion of Landsat data to surface reflectance resulted in more scene to scene variation then TOA reflectance.

Thus far, it appears that the stands used to develop the LPG&Y Coop’s growth and yield models had relatively low leaf areas.

Caveat: The Flores LAI equation was not validated for use with earlier Landsat TM scenes.

Updated growth and yield models may be needed to accurately predict the outcomes of current silvicultural regimes.

Update Summary III(Moan, Ellenwood)

• Forest Health– Southern Pine Beetle hazard rating pilot (Landsat +

NC Statewide Lidar)– Slight improvement in Landsat-based host BA models

(Cubist) when statewide lidar-derived DEM used – Also integrating forest age into hazard rating using DI

for age classes deemed susceptible to SPB from CART/Cubist models

Update Summary IV• OSM Abandoned Minelands Reforestation

Pilot (Sen, Zipper, Masek)– Focus on WV and western VA– Detection of previously mined areas (first

phase)– Assessment of reclamation quality– Using explicit trajectory-based approach

Inde

x V

alue

Time

Clear-cutting

Vegetated

Rev

eget

atio

n

Re-vegetated

0

Y

X

Mined

Fig.1 Ideal Spectral Trajectory of a reclaimed mined vegetation

•Disturbance like mining has distinctive temporal progressions of spectral values

• Temporal trend of vegetation recovery termed as “spectral trajectory”

• Spectral trajectories function as a diagnostic of the change

•Multitemporal trends are more stable than individual image pixel change reading (Kennedy et al, 2007).

Multitemporal approach-Rationale

WV site 3

0.0000

0.1000

0.2000

0.3000

0.4000

0.5000

0.6000

0.7000

0.8000

0.9000

1.0000

0 2 4 6 8 10 12

Landsat scene

ND

VI

Sample 1

Sample 2

Sample 7

Sample 8

Sample 9

Sample 10

Sample 3

June 1987

Aug. 1988

Sept. 1994

Sept. 1998

Mar. 2000

Apr. 2000

June 2000

Oct. 2000

May 2002

June 2003

June 1988

Landsat scene dates

1988, points not mined 1991, points already mined

0.0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1.0

NDV

I

Years

NDVI

64

67

-20

-15

-10

-5

0

5

Dis

turb

ance

Inde

x va

lues

Years

Disturbance Index

64

67

Search for the ideal index value:

0.00

0.10

0.20

0.30

0.40

0.50

0.60

0.70

0.80

0.90

1.00

ND

VI v

alue

s

Date

ndvi_mined_64

ndvi_mined_67

ndvi_notmined_25

ndvi_notmined_26

ndvi_notmined_27

ndvi_notmined_28

Comparison of NDVI of mined and surrounding un-mined points

Location of the mined and un-mined points on the False color image of 1998

•Points 64 and 67 are mined points

•Points 25, 26, 27 and 28 are un-mined points.

Comparison of NDVI values of mined and un-mined points

-0.8

-0.6

-0.4

-0.2

0.0

0.2

0.4

0.6

0.8

1.0

ND

VI a

nd N

orm

aliz

ed N

DV

I val

ues

Years

64_Norm_NDVI

67_Norm_NDVI

64_NDVI

67_NDVI

NDVI and normalized NDVI values for pts.64 and 67

0.00

0.10

0.20

0.30

0.40

0.50

0.60

0.70

0.80

0.90

1.00

0.00

2.00

4.00

6.00

8.00

10.00

12.00

14.00

16.00

ND

VI v

alu

es

RSR

val

ue

Date

rsr ndviPt. 64

0.00

0.10

0.20

0.30

0.40

0.50

0.60

0.70

0.80

0.90

1.00

0

50

100

150

200

250

9/17/84 6/6/87 6/8/88 9/21/91 9/29/94 9/5/97 9/24/98 6/9/00 5/22/02 6/2/03 9/11/05

ND

VI

ND

MI

Years

ndmi ndviPt.64

Reduced Simple Ratio (RSR) and NDVI values for pt.64

NDMI and NDVI values for pt.64

Some other promising indices:

-20

0

20

40

60

80

100

120TC

2

Years

Tasseled Cap 2

6467

TC2 (Greenness) values for pts. 64 and 67

Update Summary V• MEASURES (Management-scale Ecological

Assessment Using Remote Sensing; Potter, Kline, Blinn, Saleh, Walker)

• Key features: web-delivery, RS (lidar, digital orthoimagery, Landsat), scale, validated numbers, real-time modeling

• Foundation, state, and Forest Service support• Initially just C and water; biodiversity in development• Linkage with BSI and NGO Ecosystem Services

model clearinghouse (TNC)

Conclusions

Landsat-scale = management scale (almost perfect spatial resolution)

Seasonal and interannual time series now pro forma (function, process)

Importance of data set continuity(Why not 8-day revisit akin to Resource21)

Synergism from working more as team

Multitemporal in situ data sets greatly increase value of Landsat data

Thanks!

Randolph H. Wynne