Estimating Cotton Defoliation with Remote Sensing

1
Estimating Cotton Defoliation with Remote Sensing

• Glen Ritchie1 and Craig Bednarz2

1UGA Coastal Plain Experiment Station, Tifton,
GA 2Texas Tech, Lubbock, TX
Funding provided by
2
Cotton Defoliants

• Facilitate machine harvest
• Reduce weather-induced losses
• Cost
• May be unnecessary in places

3
Estimating Defoliation

• Visual Estimates
• Inexpensive
• Quick
• Accurate???
• Remote Sensing
• Can cover broad areas
• Estimates vegetation well
• Relatively untested on defoliation

4
Vegetation Indices

• Plant vs. soil reflectance
• Ratios
• Differences
• Derivatives
• Separate plant and soil

Red Edge
NIR
Visible
Blue
Green
Red
Plant
Soil
5
Red Edge
Red
Blue
Green
NIR
Soil
Plant
6
Normalized DifferenceVegetation Index (NDVI)

• (NIR-R)/(NIRR)

or (?2 - ?1)/(?2 ?1)
0.5
0.5
0.5
?2
(0.5 - ) (0.5 )
(0.5 -0.04) (0.50.04) 0.85
( - ) ( )
?1
0.04
0.04
0.04
Rouse et al., 1973
7
NDVI

• Which wavelengths?
• Red and NIR
• Green and red edge variants
• Higher order models

8
Potential Confounding Factors

• Atmospheric effects
• Green leaves on ground
• Desiccated leaves on plant
• Plant height, orientation
• Leaf structure

9
Materials and Methods

• Four locations in Tifton 2003-2004
• DP 555 and Stoneville 4892
• Reflectance on 0.91 m of row
• Visual estimates
• Leaves removed by hand
• LAI from leaf area meter
• NDVI regressed against LAI

10
Materials
Apogee PAR/NIR Spectrometer
350-900 nm range 1.5 nm resolution
LI-COR LI-3100 leaf area meter
2 m fiber optic cable
Photos courtesy Apogee Instruments, Inc.
11
Regression Analysis
(?2 - ?1)/(?2 ?1)

• If ?1 and ?2 are arbitrary, 250,000 NDVI
  wavelength combinations possible
• If ?2 is fixed, 500 NDVI possibilities

?2
820 nm
12
Regression Analysis

• LAI estimates were compared using the coefficient
  of determination (r2).
• r2 1 perfect relationship between x and y
• r2 0 no relationship between x and y

r2 0.0
r2 1.0
13
Results
Quadratic (y ?0 ?1x ?2x2)
Linear (y ?0 ?1x)
All dates, all locations
14
Results

• High quadratic correlation flattening of NDVI at
  high LAI levels.
• Red NDVI did not increase above LAI of 1.2.
• Red edge NDVI continued to trend upward with LAI.

15
Results Quadratic Model
Maximum Estimated LAI
Minimum Estimated LAI
Crop Reflectance
16
Results All Combinations

• Comparison of all ?1 and ?2
• Highest correlations Combinations of red edge
  and near-infrared reflectance bands

17
Results Visual Estimates
Reviewer 1 Reviewer 2 Reviewer 3
18
Results Visual Estimates
Reviewer r2 (all LAI) r2 (LAIlt0.5) Slope (LAIlt0.5) r2 (LAIgt0.5) Slope (LAIgt0.5)
NDVI710 nm 0.90 0.87 0.50 0.50 0.17
1 0.73 0.64 -78.9 0.03O NS
2 0.94 0.76 -94.1 0.81 -75.7
3 0.90 0.55 -40.4 0.48 -26.0
O Not significant at 0.05 level
19
Conclusions

• Red edge NIR wavelength combinations most
  consistently estimate LAI
• Individual reviewers are generally very good at
  estimating changes in LAI
• Estimates vary between reviewers
• Precision defoliant application

20
Acknowledgments

• Georgia Cotton Commission
• Cotton physiology technical staff
• Steve Brown and Stanley Culpepper