Comparing Different Remote Sensing Approaches for Early Season Nitrogen Deficiency Detection in Corn

1
Comparing Different Remote Sensing Approaches for
Early Season Nitrogen Deficiency Detection in
Corn
NUE Workshop Improving NUE using Crop Sensing,
Waseca, MN
Yuxin Miao1, David J. Mulla1, Gyles W. Randall2,
Jeff A. Vetsch2, and Roxana Vintila3
1. Precision Agriculture Center, University of
Minnesota. 2. Southern Research and Outreach
Center, University of Minnesota. 3. Research
Institute of Soil Science and Agrochemistry
(ICPA), Romania.
2
Different Sensing Approaches for Precision N
Management

• Chlorophyll Meter

SPAD 502

• GreenSeeker

• CropScan Multispectral Radiometer

3
Different Sensing Approaches for Precision N
Management

• Aerial or Satellite -based Remote Sensing

High spatial resolution remote sensing images are
potentially cheaper, more efficient and more
spatially detailed than chlorophyll meter or
other ground-based hand-held sensors.
Hyperspectral Remote Sensing
(Image from http//www.eoc.csiro.au/hswww/Overview
.htm)
4
Objectives

• Identify hyperspectral bands (wavelengths), band
  ratios and vegetation indices that are sensitive
  to early season corn plant N status

• To compare the effectiveness of different sensing
  approaches to monitor early season corn plant N
  status and detect N deficiency

• SPAD Meter
• GreenSeeker
• CropScan multispectral radiometer
• Aerial hyperspectral remote sensing and,
• Aerial multispectral remote sensing (simulated).

5
Materials and Methods Study Sites
Field 1 Corn-Soybean Rotation
Field 2 Corn-Corn Rotation
6
Materials and Methods N Treatments
Field 1, Corn-Soybean Rotation
3 x 15.2 m
7
Materials and Methods N Treatments
Field 2, Corn-Corn Rotation
8
Materials and Methods Data Collection

• N Concentration

• V9 Whole plant sampling, 10 plants N
  concentration, biomass
• R1 Ear leaf, 10 leaves
• Harvest grain and stover.

• SPAD Meter

• F1 V9, V11, R1, and R3
• F2 V7, V9-10, V12, R1, R3
• Collected 30 readings from each plot.

• GreenSeeker

• F1 V9 and V11
• F2 V8, V9-10, V11-V12 and V12.

• CropScan Multispectral Radiometer

• V6 and V9
• About 50 cm above the canopy, three samples each
  plot.

9
Materials and Methods Data Collection

• Aerial Hyperspectral Remote Sensing

AISA-Eagle (AE) Hyperspectral Imager
61 bands from 392 982 nm, at 8.76 9.63nm
At 0.75 m spatial resolution
V9, R1, R2 and R4
Pixels of the central two rows in each plot were
averaged

• Simulated Multispectral Remote Sensing

Landsat ETM sensors four broad bands

• Blue 450-515nm
• Green 525-605nm
• Red 630-690 nm
• NIR 775-900nm.

10
Materials and Methods Band Combinations

• Simple Ratio (SR)

Ratio Definition Reference
Green Index Zarco-Tejada Miller (ZTM) PSSRa PSSRb PSSRc SPRI SR1 SR2 SR3 SR4 SR5 SR6 SR7 R554/R677 R750/R710 R800/R680 R800/R635 R800/R470 R430/R680 NIR/Red R801/R670 NIR/GreenR800/R550 R700/R670 R740/R720 R675/(R700 x R650) R672/(R550 x R708) R860/(R550 x R708) Smith et al., 1995 Zarco-Tejada et al., 2001 Blackburn, 1998 Blackburn, 1998 Blackburn, 1998 Penuelas et al., 1994 Daughtry et al., 2000 Buschman and Nagel, 1993 McMurtrey et al., 1994 Vogelman et al., 1993 Chappelle et al., 1992 Datt, 1998 Datt, 1998
11
Materials and Methods Band Combinations

• Difference Index (DI) and Normalized
  Difference Index (NDI)

Index Definition Reference
DI1 DVI NDVI Green NDVI PSNDb PSNDc NPCI NPQI SIPI mND705 mSR705 NDI1 NDI2 NDI3 R800-R550 R800-R680 (R800-R680)/(R800R680) (R801-R550)/(R800R550) (R800-R635)/(R800R635) (R800-R470)/(R800R470) (R680-R430)/(R680R430) (R415-R435)/(R415R435) (R800-R445)/(R800-R680) (R750-R705)/(R750R705-2 x R445) (R750-R445)/(R705-R445) (R780-R710)/(R780-R680) (R850-R710)/(R850-R680) (R734-R747)/(R715R726) Buschman and Nagel, 1993 Jordan, 1969 Lichtenthaler et al., 1996 Daughtry et al., 2000 Blackburn, 1998 Blackburn, 1998 Penuelas et al., 1994 Barnes et al., 1992 Penuelas et al., 1995 Sims and Gamon, 2002 Sims and Gamon, 2002 Datt, 1999 Datt, 1999 Vogelman et al., 1993
12
Materials and Methods Band Combinations

• Integrated Index (II)

Index Definition Reference
MCARI TCARI OSAVI TCAVI/OSAVI TVI MCARI/OSAVI RDVI MSR MSAVI MTVI (R700-R670)-0.2x(R700-R550)(R700/R670) 3x(R700-R670)-0.2x(R700-R550)(R700/R670) (10.16)(R800-R670)/(R800R6700.16) 0.5x120x(R750-R550)-200x(R670-R550) (R800-R670)/SQRT(R800R670) (R800/R670-1)/SQRT(R800/R6701) 0.5x2xR8001-SQRT((2xR8001)2-8x(R800-R670)) 1.2x1.2x(R800-R550)-2.5x(R670-R550) Daughtry et al., 2000 Haboudane et al., 2002 Rondeaux et al., 1996 Haboudane et al., 2002 Broge and Leblanc, 2000 Zarco-Tejada et al., 2004 Rougean and Breon, 1995 Chen, 1996 Qi et al., 1994 Haboudane et al., 2004

• Broad Band Combinations

• NIR/Green
• NIR/Red
• Blue NDVI
• Green NDVI
• Red NDVI.

13
Materials and Methods Analysis

• Correlation analysis

• Multiple linear regression

• Nitrogen Sufficiency Index (NSI)

Yield, Plant N, SPAD, or index
NSI
x 100
Reference value
F1 the average of the highest two preplant N
rates 168 and 202 kg ha-1.
F2 224 kg ha-1.
NSI of Plant N Concentration as standard
14
Results and Discussion Plant N Variability
36.1
34
31.9
28.2
24.9
CV 9.33
CV 14.06
18.7
202 kg ha-1
224 kg ha-1
15
Results and Discussion Impact of N Rate on
Reflectance
CropScan MSR
Hyperspectral RS
16
Results and Discussion Sensitive Wavelengths
Correlation between plant N concentration and
CropScan reflectance at V9
760-1000nm
560-710nm
17
Results and Discussion Sensitive Wavelengths
Correlation between plant N concentration and
hyperspectral reflectance at V9
742-982nm
554 and 563nm
695nm
18
Results and Discussion Sensitive Indices
Correlation with Plant N Concentrations at V9
Field 1
Field 2
0.51
CropScan
GNDVI
0.72
NIR/Green R800/R550
0.50
0.71
NDI2
0.71
0.39
Hyperspectral
NDI1
0.78
0.47
NDI2
0.79
0.41
SPAD Meter
0.58
0.85
GreenSeeker NDVI
0.26
0.49
0.31
Simulated Landsat ETM
GNDVI
0.68
19
Results and Discussion Multiple Regression
Correlation Coefficient
Field 1
Field 2
SPAD Meter
0.58
0.85
CropScan
0.77
6 (F1)/5(F2) bands
0.79
0.67
5 (F1)/4(F2) Indices
0.79
Hyperspectral
4 bands
0.73
0.89
0.66
4 (F1)/3(F2)indices
0.88
Simulated Landsat ETM
0.56
3 bands
0.89
0.57
2 indices
0.88
20
Results and Discussion N Sufficiency Index
21
Results and Discussion N Deficiency Detection
Treatment Level, Field 1, Corn-Soybean Rotation
Hyper
Landsat
CropScan
ID
Plant N
SPAD
GS
TVI
NIR/G
Yield
G
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
22
Results and Discussion N Deficiency Detection
Treatment Level, Field 2
Hyper
Landsat
CropScan
ID
Plant N
SPAD
GS
DI1
NIR
Yield
SR7
1
2
3
4
5
6
7
8
9
10
11
12
13
14
23
Results and Discussion N Deficiency Detection
Plot Level, Field 1, Corn-Soybean Rotation
Deficient Plots
Sufficient Plots
D?S
S?D
Overall Accuracy ()
34
26
100
(44)
(16)
Plant N Content
0
0
SPAD Meter
3
26
48
(35)
0
0
GreenSeeker
15
17
53
(47)
19
9
CropScan MCARI
22
14
60
(60)
12
12
Hyperspectral MCARI
25
11
60
(55)
9
15
Landsat ETM Green
17
15
53
(62)
17
11
24
Results and Discussion N Deficiency Detection
Plot Level, Field 2, Corn-Corn Rotation
Deficient Plots
Sufficient Plots
D?S
S?D
Total Accuracy ()
29
27
100
(46)
(10)
Plant N Content
0
0
SPAD Meter
21
25
82
(71)
8
2
GreenSeeker
10
26
64
(43)
19
1
CropScan TCARI/OSAVI
18
22
71
(59)
11
5
Hyperspectral SR7
21
22
77
(61)
8
5
Landsat ETM NIR/Green
17
20
66
(57)
12
7
25
Results and Discussion Promising Indices

• Normalized Difference Index 2 (NDI2)

R850-R710
NDI2
R850-R680
26
Results and Discussion Promising Indices

• Simple Ratio 7

R860
SR7
R550 x R708
27
Conclusions

• The sensors performed better in corn-corn
  rotation field
• than in corn-soybean rotation field at V9

• The NIR region was most sensitive to N
  deficiency at V9

• Reflectance at around 550-560 nm, 696 nm, and
  NIR region
• was highly correlated with corn plant N
  concentration at V9

• SPAD meter readings and GreenSeeker NDVI data
  had the
• highest and lowest correlation coefficients
  with corn plant N
• concentration, respectively

• Hyperspectral aerial remote sensing has a good
  potential to
• monitor spatial corn N variation, and
  identify N deficiency
• at V9, especially in corn-corn rotation
  fields

• SR7 and NDI2 were promising indices for N
  deficiency
• identification and deserve further testing.
  

28
Thank you for your attention!
Any Questions?