Integrated Assessment of the Utility of Soil, Crop and Remote Sensing Data for Precision Management

1
Integrated Assessment of the Utility of Soil,
Crop and Remote Sensing Data for Precision
Management of Corn

• Researchers
• Antoni Magri, Harold van Es, Bill Cox
• and Michael Glos

Collaborators Doug Freier, Kevin Swartley, and
Craig Richards
Supported by USDA-SARE, USDA-AES, and
Mpower3-Emerge
2
Research Objectives

• Determine information sources that are useful in
  identifying fertility patterns for precision
  nutrient management on NY fields
• Digital multi-spectral images
• Soil survey information
• Yield maps
• Grid soil sampling
• Investigate relationships among these sources of
  information, and its management implications

3
Materials and Methods Sites

• Studies were conducted for five fields in Central
  New York, from1999 until 2001
• Two fields on a dairy farm (manure). Three fields
  on cash grain farms (no manure)
• All fields planted to maize
• during all three years

Onondaga County fields
4
Materials and Methods Bare-soil images

• Three-band (G, R, NIR) digital aerial images of
  bare soil (pre-emergence) were taken in June 1999
  and 2000 (courtesy of MPower3-Emerge).

5
Materials and Methods Soil Sampling

• Soil samples were taken on a regular grid (45 by
  45 m) in mid June every year and analyzed for pH,
  OM and available P, K and NO3.
• Soil types were digitized from County Soil Survey
  maps.

Soil sample sites (45 by 45 m grid)
6
Materials and Methods Yield

• Yield monitors (combined with DGPS units) were
  used to produce yield maps of each field in each
  of 3 years.

7
Results Field-average fertility trends
8
Variogram modeling
9
Soil fertility variogram summary
Dairy fields Onondaga 1 and 2 Cash crop (CC)
fields Seneca 1, 2 and 3.
Spatial dependence total sill / nugget. strong, 0.25 to 0.75 moderate, 0.75 to 0.99
weak, 1 random
10
Krigedmaps
11
Soil properties temporal consistencyat sampling
sites (r-values)
ONONDAGA 1
ONONDAGA 2
X 00
X 01
X 00
X 01
PH 1999
0.86

0.95

PH 1999
0.91

0.22
PH 2000
0.91

PH 2000
0.33

P 1999
0.65

0.57

P 1999
0.83

0.77

P 2000
0.42

P 2000
0.75

K 1999
0.81

0.81

K 1999
0.53

0.65

K 2000
0.73

K 2000
0.56

OM 1999
0.90

0.95

OM 1999
0.88

0.86

OM 2000
0.94

OM 2000
0.82

NO3 1999
0.20
0.58

NO3 1999
0.29

0.44

NO3 2000
0.32
NO3 2000
0.31

Manured sites
12
Soil properties temporal consistencyat sampling
sites (r-values)
SENECA 1
SENECA 2
X 00
X 01
X 00
X 01
PH 1999
0.80

0.81

PH 1999
0.85

0.80

PH 2000
0.83

PH 2000
0.83

P 1999
0.68

0.58

P 1999
0.41

0.51

P 2000
0.76

P 2000
0.76

K 1999
0.63

0.63

K 1999
0.67

0.87

K 2000
0.72

K 2000
0.71

OM 1999
0.97

0.91

OM 1999
0.95

0.96

OM 2000
0.91

OM 2000
0.95

NO3 1999
0.44

-0.01
NO3 1999
0.13
0.09
NO3 2000
0.17
NO3 2000
0.62

Non-manured sites
13
Soil properties cross-correlations
Manured sites
14
Soil properties cross-correlations
SENECA 1
SENECA 2
PH 1999
P 1999
K 1999
OM 1999
PH 2000
P 2000
K 2000
OM 2000
SENECA 3
Non-manured sites
15
Aerial image analysis
Individual bare soil bands
Principal Component image
Principal components was used to combine all 3
bands into one image, which was then used to
estimate correlations of soil reflectance to soil
fertility indicators.
16
Principal component analysis
ONONDAGA 1
1999
2000
PCA1
PCA2
PCA1
PCA2
NIR
0.51
-0.78
0.44
-0.90
RED
0.65
0.07
0.75
0.32
GREEN
0.56
0.63
0.50
0.30
VARIANCE EXP
1061.52
40.95
1108.93
81.10
VARIANCE EXP
96.29
3.71
93.19
6.81
ONONDAGA 2
1999
2000
PCA1
PCA2
PCA1
PCA2
NIR
0.60
-0.78
0.43
-0.88
RED
0.59
0.27
0.67
0.15
GREEN
0.54
0.57
0.60
0.45
VARIANCE EXP
1074.67
44.44
3537.60
79.05
VARIANCE EXP
96.03
3.97
97.81
2.19
All PCA 1s are equally weighted combinations
of the 3 bands
17
Principal component analysis
SENECA 1
1999
2000
PCA1
PCA2
PCA1
PCA2
NIR
0.57
-0.82
0.54
-0.84
RED
0.59
0.40
0.62
0.42
GREEN
0.57
0.41
0.56
0.34
VARIANCE EXP
1210.86
278.25
4142.67
78.80
VARIANCE EXP
81.31
18.69
98.13
1.87
SENECA 2
2000
1999
PCA1
PCA2
PCA1
PCA2
NIR
0.76
-0.66
0.63
-0.77
PCA1s generally explain over 95 of the
variance, except for Seneca 1 in 1999, which had
a pronounced weed infestation at the time when
the image was taken.
RED
0.49
0.54
0.58
0.39
GREEN
0.43
0.52
0.52
0.51
VARIANCE EXP
4124.09
108.33
5979.93
69.38
VARIANCE EXP
97.44
2.56
98.85
1.15
SENECA 3
1999
2000
PCA1
PCA2
PCA1
PCA2
NIR
0.70
-0.68
0.65
-0.76
RED
0.53
0.27
0.57
0.44
GREEN
0.49
0.68
0.50
0.49
VARIANCE EXP
2593.26
65.31
3756.89
41.10
VARIANCE EXP
97.54
2.46
98.92
1.08
18
Integrated analysis

• Analysis techniques
• Correlations
• PCA and multiple regression
• Multivariate Adaptive Regression Splines (MARS)

19
Results aerial image and soil property
correlations
SARE01
SARE03
SARE06
SARE07
SARE08
PCA1 1999
PCA1 1999
PCA1 1999
PCA1 1999
PCA1 2000
PH1999
-0.36
-0.03
-0.40

-0.07
-0.53

P1999
-0.38
0.06
-0.37

-0.27

-0.18
K1999
0.12
-0.03
-0.06
-0.35

0.17
OM1999
-0.64

-0.52

-0.73

-0.62

-0.56

NO31999
-0.31
-0.45

-0.03
-0.55

-0.10
PH2000
-0.36
0.05
-0.35

-0.13
-0.57

P2000
-0.33
0.12
-0.17
-0.28

-0.26
K2000
0.16
-0.13
0.00
-0.27

0.12
OM2000
-0.49

-0.52

-0.74

-0.62

-0.68

NO32000
0.29
-0.20
-0.08
-0.40

0.05
PH2001
-0.37
-0.04
-0.41

-0.12
-0.47

P2001
-0.27
0.12
-0.19
-0.10
-0.30
K2001
0.27
-0.05
0.04
-0.27

0.19
OM2001
-0.62

-0.61

-0.71

-0.64

-0.74

NO32001
-0.21
-0.29

0.06
0.04
-0.14
Reflectance mostly correlates with OM content
20
Results Aerial image and yield correlations

• Correlations between PCA1 and yield are good for
  each season, but inconsistent from year to year
  Correlations are negative for yields for 1999 and
  2001, and positive for yields in 2000. This
  effect is stronger for the cash crop fields
  (Seneca 1, 2 and 3) than for the dairy fields
  (Onondaga 1 and 2)

• Explanation
• Year 2000 was wet and cold
• wet areas yielded low.
• Years 1999 and 2001 were dry ? wet areas yielded
  high

21
Results Relationship of soil type to yield

• Yields averaged over each soil type show few
  significant differences between soils.
• Trends over the 3 year period are strongly
  related to weather patterns (1999 was hot and
  dry, 2000 was humid and cold, 2001 was warm and
  dry).

22
Conclusions

• Spatial distributions (variograms) of soil
  properties show potential for site-specific
  fertilizer ands lime application. Dairy fields
  show higher spatial variability.
• pH, P and OM (in all fields) and K (in the cash
  crop fields only) show the most consistent
  relationships. NO3 is the least consistent
  property.
• High temporal consistency exists for soil
  properties over time, except for NO3 content on
  the dairy fields. This suggests that field
  characterization via intensive soil sampling does
  not need to be repeated annually in order to
  support management decisions.
• Correlations between P and K are consistently
  significant with years, suggesting that they may
  be managed jointly through fertility management
  zones. No correlations existed with NO3.

23
Conclusions

• Since the PCA1 images are equally weighted
  combinations of all 3 bands, any individual band
  or a panchromatic image (black and white) would
  have provided the same information. However,
  hyper-spectral images may provide additional
  information, which has not been explored in this
  project.
• Aerial images of bare soil provide valuable
  information regarding drainage patterns and OM
  content, but not of fertility patterns.
• PCA1 reflectance correlates well with yield in
  any given year, but is not consistent between dry
  and wet years.
• Soil survey data show little relationship to
  yield or fertility patterns.

24
Data Mining

• Precision agriculture generates large data sets
• Can we use this data more effectively?
• Data mining methods are designed to find patterns
  in large data sets
• MARS (Multivariate Adaptive Regression Splines)
  is a regression technique that allows for rapid
  analysis of large data sets using flexible
  functions

25
MARS Analysis
26
MARS Analysis
27
MARS Analysis Results (Preliminary)
28
Thank You!