Remote Sensing Crop Residue Cover - USDA · Remote Sensing Crop Residue Cover Guy Serbin, E....

Remote Sensing CropResidue Cover

Guy Serbin, E. Raymond Hunt Jr., andCraig S. T. Daughtry

USDA/ARS Hydrology and RemoteSensing Lab, Beltsville, MD

What are crop residues?

• Crop residues arestalks, cobs, andother plant parts leftbehind after aharvest.

• They are alsoreferred to as non-photosyntheticvegetation.

Why are crop residues important?

• When left on the soil surface, they:

– Protect the soil from wind and water erosion.

– Reduce evaporation by acting as a mulch.

– Their breakdown helps sequester carbon tothe soil.

• This also recycles nutrients.

– Improve soil structure and water retention.

• When removed from the soil:

– They do not benefit the soil.

– But, they can be used for cellulosic ethanolbiofuels.

Tillage systems and residues

A. Intensively tilled field B. No-tilled field

• Intensive tillage removes residue, exposes soil to erosion.

• Conservation tillage (e.g., no-till) leaves residue on fields.

• With conservation tillage, farmers save money on fuel, cansell carbon credits, and receive monetary benefits.

CTIC and USDA-NRCS tillagedefinitions

• Intensive tillage (< 15% residue cover)

• Reduced tillage (15 – 30% residue cover)

• Conservation tillage (> 30% residue cover)

Where else is non-photosyntheticvegetation important?

• Dry vegetation is animportant indicator ofrangeland quality and soilhealth.

• Dry plant material easilycatches fire:– Prescribed burning is an

important management practicein Western US.

– In Oct. 2007, California wildfirescaused over $1 billion indamage.

– Wildfires also occurring thisyear.

Prescribed rangeland burn, imagecourtesy Wyoming Wildlife andNatural Resource Trust

Simi Valley, CA, Oct. 14, 2008.(Associated Press)

Verification of residue cover

A. Line-point transect.

B. Photographic. C. Photo comparison.

Landsat TM-based indices:• NDI5 (McNairn and Protz, 1993)*

• NDI7 (McNairn and Protz, 1993)

• NDSVI (Qi et al., 2003)*

• NDTI (van Deventer et al., 1997)

*Only NDI5, NDSVIappropriate for AWiFS/LISS III

Remote sensing of crop residue coverR

efle

cta

nce

Facto

r

0.0

0.1

0.2

0.3

0.4

0.5

FF

Refle

cta

nce

Facto

r0.0

0.1

0.2

0.3

0.4

F

AS

TE

R 1 2 3 4 56789 5 6 7 8 9

Wavelength, nm

400 1000 1600 2200

Lan

dsa

tT

M

123 4 5 7

Wavelength, nm

2000 2200 2400

7

2.0 2.1 2.2

75

75

TMTM

TMTMNDTI

54

545

TMTM

TMTMNDI

74

747

TMTM

TMTMNDI

35

35

TMTM

TMTMNDSVI

Remote sensing of crop residue cover• ASTER: Lignin-Cellulose

Absorption (LCA) Index

• Hyperspectral SWIR:Cellulose AbsorptionIndex (CAI)

– CAI most effective inmeasuring residuecover:

• Shortwave infrared

• Narrowband

8562100 ASTERASTERASTERLCA

Refle

cta

nce

Facto

r

0.0

0.1

0.2

0.3

0.4

0.5

FF

Refle

cta

nce

Facto

r0.0

0.1

0.2

0.3

0.4

F

AS

TE

R 1 2 3 4 56789 5 6 7 8 9

Wavelength, nm

400 1000 1600 2200

Lan

dsa

tT

M

123 4 5 7

Wavelength, nm

2000 2200 2400

7

2.0 2.1 2.2

210122112031 2/100 RRRCAI

CAI, surface soils

• Crop residues contrast well with all soils, greenvegetation.

• N = 893 surface soils from Brown et al. 2006.

Surface soil samples

CA

I

-12

-10

-8

-6

-4

-2

0

2

4

6

8

Alfi

sol

And

isol

Arid

isol

Ent

isol

His

toso

lIn

cept

isol

Mol

lisol

Oxi

sol

Spo

doso

lU

ltiso

lV

ertis

olU

ndet

.R

esid

ues

Live

corn

LCA, surface soilsSurface soil samples

LC

A

-8

-6

-4

-2

0

2

4

6

8

10

12

14

Alfi

sol

And

isol

Arid

isol

Ent

isol

His

toso

lIn

cept

isol

Mol

lisol

Oxi

sol

Spo

doso

lU

ltiso

lV

ertis

olU

ndet

.R

esid

ues

Live

corn

• Some overlap seen between crop residues,soils, and green vegetation.

NDTI, surface soilsSurface soil samples

ND

TI

-0.2

-0.1

0.0

0.1

0.2

0.3

0.4

0.5

0.6

Alfi

sol

And

isol

Arid

isol

Ent

isol

His

toso

lIn

cept

isol

Mol

lisol

Oxi

sol

Spo

doso

lU

ltiso

lV

ertis

olU

ndet

.R

esid

ues

Live

corn

• Green vegetation has strongest response.

• Crop residues overlap some soils andgreen vegetation.

NDSVI, NDI5, surface soilsSurface soil samples

ND

SV

I

-0.2

0.0

0.2

0.4

0.6

0.8

1.0

Alfi

sol

And

isol

Arid

isol

Ent

isol

His

toso

lIn

cept

isol

Mol

lisol

Oxi

sol

Spo

doso

lU

ltiso

lV

ertis

olU

ndet

.R

esid

ues

Live

corn

Surface soil samples

ND

I5

-0.8

-0.6

-0.4

-0.2

0.0

0.2

0.4

0.6

Alfi

sol

And

isol

Arid

isol

Ent

isol

His

toso

lIn

cept

isol

Mol

lisol

Oxi

sol

Spo

doso

lU

ltiso

lV

ertis

olU

ndet

.R

esid

ues

Live

corn

• Green vegetation has strongest response.

• Crop residues overlap most soils.

• These indices only usable in limited areas.

2006-2007 study areas• Airborne hyperspectral SpecTIR

imagery were acquired in north-central Indiana.

• Imagery acquired shortly afterplanting (May/June).

• Most fields were soybean or corn.

• Ground truth of residue coveracquired at > 50 fields using line-point transects, 2 locationsmeasured per field.

• Soil and residue samples alsoacquired at select locations.

• Hyperspectral bands convolved toequivalent ASTER VNIR andSWIR, and Landsat TM bands.

2006-2007 field analysis methods• Pixels within 30 m of sampling locations analyzed

for:

– NDVI for live green vegetation cover.

– Indices residue cover.

1. Compared with line-point transect fR using linearregression.

2. Inversion to determine fR for CAI:

fR = (CAIpixel - CAIsoil)/(CAIresidue - CAIsoil)

– Two CAIsoil endmembers: low- and high-soil organic carbon(SOC).

– Two CAIresidue endmembers: Corn and soybean.

2006-2007 field analysis methods• Linear regression and inversion fR

compared against line-point transect fRestimates using:

– Correlation coefficients (r2).

– Root-mean-square errors (RMSE).

• Data points aggregated into two residuecover classes:

fR < 0.30 (Intensive and reduced till)

0.30 ≤ fR (Conservation till)

• Classifications were assessed foraccuracy.

Indiana 2006 results

• NDVI showed that live green vegetation wasminimal, ranged from 0.10 to 0.29, mean of 0.16.

NDSVI

0.25 0.30 0.35 0.40 0.45 0.50 0.55

0

20

40

60

80

100

CAI

-1 0 1 2 3 4

Perc

entre

sid

ue

cover

0

20

40

60

80

100r2 = 0.80RMSE = 9.9%Accuracy = 93.7%

NDTI

-0.05 0.00 0.05 0.10 0.15 0.20

0

20

40

60

80

100r2 = 0.39RMSE = 17.1%Accuracy = 77.9%

LCA

1 2 3 4 5 6 7 8 9

Perc

en

tre

sid

ue

cove

r

0

20

40

60

80

100r2 = 0.47RMSE = 15.9%Accuracy = 89.5%

NDI5

-0.35 -0.30 -0.25 -0.20 -0.15 -0.10 -0.05

0

20

40

60

80

100

NDI7

-0.4 -0.3 -0.2 -0.1 0.0 0.1

0

20

40

60

80

100r2 = 0.03RMSE = 21.6%Accuracy = 68.4%

r2 = 0.05RMSE = 21.4%Accuracy = 70.5%

r2 = 0.04RMSE = 21.5%Accuracy = 69.5%

Indiana 2006 statistical data

-0.71

0.684

0.215

0.05

NDI7

0.38

0.695

0.219

0.02

NDI5

4.66

0.779

0.171

0.39

NDTI

0.51

0.705

0.216

0.04

NDSVI

9.4016.3112.24z-stat

0.8950.9580.9372-class

accuracy

0.1590.0990.099RMSE

0.470.820.80r2

LCACAIcal.

CAIreg.

Index

Indiana 2007 results• In 2007 aircraft data

acquired later than in2006.

• Scene was significantlygreener: NDVI range:0.17 to 0.84, mean of0.36.

• CAI, LCA gaveacceptable results.

• NDTI showedimprovement afterremoval of NDVI > 0.5pixels, and subpixelgreen cover correction(NDTIgc).

NDTI

-0.3 -0.2 -0.1 0.0 0.1 0.2 0.3

f R

0.0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1.0

NDTINDTI

gc

NDTI

0.0 0.1 0.2 0.3

ND

VI

0.0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

Indiana 2007 fR, spectral index

CAI

-1 0 1 2 3 4

f R

0.0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1.0

LCA

0 1 2 3 4 5 6 7 8 9 10

f R

0.0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1.0

Indiana 2007 statistics

5.973.044.4711.8711.40z-statistic

0.7280.6250.6760.8600.8532-class

accuracy

0.2190.2350.1560.1070.107RMSE

0.090.180.640.830.83r2

NDTIgcNDTILCACAI 4-way

CAIreg.

Index

Spectral index results• Spectral index

performance (frombest to worst):

1. CAI

2. LCA

3. NDTI

4. NDI5*, NDI7, NDSVI*

*Only NDI5, NDSVIappropriate forAWiFS/ LISS III

Conclusions• CAI works best for crop residue cover estimation.

• CAI might also work well for fire risk assessmentand rangeland quality research.

• Future Resourcesat sensors could include thethree CAI bands.

• NDTI (TM bands 5 and 7) works well; may becorrected for vegetation.

• Current AWiFS/LISS bands may estimate cropresidue cover only at a few locations.