Sensor Use for Determining Crop Variability

Download Report

Transcript Sensor Use for Determining Crop Variability

DETECTING VARIABILITY
WITH CROP SENSORS
(TOOLS TO IMPROVE SOIL AND FORAGE
SAMPLING TECHNIQUES)
Stewart Reed
Oklahoma State University
Biosystems & Agricultural Engineering Dept.
7/16/2015
1
Outline
•
•
•
Sensor Technology Basics
The Toolbox
Applied Research and Studies
– Wheat pasture forage biomass sampling
– Site specific soil sampling for variable rate
lime application
7/16/2015
2
Sensor Basics
Three key sensor properties for applied use:
1. Provides scaled data by electronic and/or
mechanical means
2. Sensors’ data are always numerical, but are
often shown visually (i.e. maps, light bars,
electronic display)
3. Data is typically in electronic form, indicating
potential value when used by electronic
control systems or management software
applications (i.e. GIS)
7/16/2015
3
Five Requirements of Sensor
Application
1. Know the sensing conditions
– position, lighting, power, external components, etc.
2. Know the data collection procedure
– sensing time, data storage, warm-up, automation/manual
settings, etc.
3. Know the data handling techniques
– storage media, transfer methods to other devices, read-outs,
displays
4. Know the reliability and quality
– calibration, periodic maintenance checks, signs of poor data
5. Know the support
– accessibility, availability, adequate, multiple levels (dealers,
manufacturers, educational)
7/16/2015
4
What's in the Toolbox
• Field Tools
–
–
–
–
–
Yield Monitors
Crop Sensors
Soil Sensors
Weather Sensors
GPS (positioning sensor)
• Support Tools
– GIS (data management software)
– Handheld computer devices (PDA)
– Personal computer
• Online Tools
– Nutrient calculators
– Weather data
– Geographical data
7/16/2015
5
I have these “tools” but don’t
quite understand how they are
intended to be used.
• Tools are to be used in combination to address a
problem or complete a specific task.
• Case in point: determine yield variability within a
single field
– Sensor tools needed: yield monitor and GPS
– Support tools needed: personal computer and GIS
software
7/16/2015
6
Determining Yield Variability
Collect data via
combine equipped
with Yield Monitor
and GPS
Transfer data using
storage media or cable
to PC-GIS software
Use GIS software
statistics to
calculate variability
7/16/2015
Use GIS software to
visually observe
variability
7
Applied Research
Study #1: Improving conventional “clipping”
method for determining standing
forage biomass in a wheat
pasture.
Study #2: Develop a site specific soil
sampling strategy for variable rate
lime application on an average
wheat field.
7/16/2015
8
Study #1
Sampling for Wheat
Forage Biomass
7/16/2015
9
Wheat Forage Biomass
Defining the Problem:
– Standing forage biomass is currently measured by a
randomized “clipping” method
– Sample size is 4ft2, and few samples are collected to
represent an entire field
• Example: if 10 samples were collected from an 80 acre field,
then 40ft2 is being used to represent 3.48 million square feet.
(1acre = 43,560ft2)
– This method does not accurately represent true
forage biomass due to plant growth variability and
variability from plant-animal interaction (grazing).
7/16/2015
10
Wheat Forage Biomass
Research Objectives
1. Illustrate plant growth variability using a crop
sensor (GreenSeeker)
2. Compare results of conventional clipping
method via analysis conducted with crop
sensor data
3. Show that clipping method coupled with crop
sensor data can be used to enhance the
accuracy of measuring forage biomass
7/16/2015
11
Wheat Forage Biomass
• Tools Needed
– Crop Sensor (GreenSeeker)
– GPS
– Personal Computer
– GIS Software
• Clipping data provided by cooperative
research
7/16/2015
12
OSU Wheat Pasture Research Station
Marshall Oklahoma (March 6, 2006)
Custom ATV GreenSeeker Mapper
•NDVI (Normalized Difference Vegetative Index, Red and NIR wavelengths)
•0.418 m2 (4.5 ft2) sensor resolution
•3.66 m (12 ft) boom width
•14.63 m (48 ft) paralleled swaths
•total sample area = 3.66/14.63 = ¼ of field
GPS
Crop Sensors
7/16/2015
13
Wheat Forage Biomass
Using GIS Software
With Paddock Layout
7/16/2015
Raw NDVI Sensor Values
14
Wheat Forage Biomass
Converting Raw Data to a Surface Map
7/16/2015
15
Wheat Forage Biomass
• Analysis
– Use average NDVI for each paddock to determine
relative forage biomass
– Calculate CV (from NDVI data) for each paddock to
illustrate forage biomass variability
– Compare average NDVI to clipping value
• High avg NDVI should correlate with high clipping value
– Compare using CV
• Avg NDVI and clipping value comparison should correlate
well with low CV values and poorly with high CV values
7/16/2015
16
Paddock 3 shows to
have more forage
biomass than
paddocks 2 and 4.
More plant biomass
in turn rows because
of higher seeding
rates.
Increased plant
biomass in areas
where soil moisture
is more available
due to water
holding/runoff
characteristics
relative to terrain and
terrace location.
7/16/2015
NDVI → Wheat Pasture
Forage Variability
•NDVI represents plant vigor.
Photosynthetic activity, total live
plant biomass, plant water
stress, etc. can be represented
by NDVI with proper calibration
techniques.
•Small and large scale variability
can be attributed to many
factors such as stocking density,
soil moisture variability, fertilizer
inputs, seeding rate, preferential
grazing, and water tank/mineral
feeder location.
17
Mid CV
Low CV
High CV
7/16/2015
18
Wheat Forage Biomass
Research Results
Conclusions:
Evaluating Standing Crop Measurements using NDVI and Measured Variability Obtained from
NDVI Sensor Readings
•Crop sensor NDVI data is capable
of showing forage variability both
small and large scale.
750
•Paddocks with less variability had
a significantly stronger relationship
between mean NDVI and clipping
measurements.
700
650
R2 = 0.7988
Standing Crop (lbs/acre)
per Paddock
600
Low CV
550
Mid CV
High CV
R2 = 0.2485
Linear (Low CV)
500
Linear (Mid CV)
R2 = 0.0196
Linear (High CV)
450
400
•Paddocks with a medium or large
variability had a poor relationship
between mean NDVI and clipping
measurements.
•Clipping procedure can be
improved by assessing variability
first and then determining proper
sample population and location.
350
300
0.29
0.3
0.31
0.32
0.33
0.34
0.35
0.36
0.37
0.38
0.39
Mean NDVI per Paddock
7/16/2015
19
Study #2
Soil Sampling for VR Lime
7/16/2015
20
Soil Sampling for VR Lime
Defining the Problem:
– 4 to 16 spot samples are collected per field
(~80 to 160 acres) and commonly mixed
resulting in one single sample sent to the lab
– Sample collection sites are sometimes
arbitrarily chosen in the most convenient
locations
– pH levels in fields often vary on a large area
basis
7/16/2015
21
Soil Sampling for VR Lime
Defining the Problem (continued):
– Conventional soil sampling techniques are not
capable of detecting soil pH variability (or any
soil nutrient variability)
– Flat rate lime applications do not address soil
pH variability and can result in economic loss
and/or a reduction in crop performance
7/16/2015
22
Soil Sampling for VR Lime
Research Objectives
1. Show that yield and crop sensor data can be
used to identify potential soil pH variability
2. Compare intense soil sampling results with
improved site sampling method to verify
strategy is sufficient for VR lime
7/16/2015
23
Soil Sampling for VR Lime
• Tools Needed
–
–
–
–
–
Crop Sensor (GreenSeeker)
Yield Monitor
GPS (one or two units)
Personal Computer
GIS Software
• Other Tools
– Soil sample kit (probe and bags)
– Hand held device with GPS
– Navigation software (installed on hand held device)
7/16/2015
24
160 Acre Oklahoma Wheat Field
7/16/2015
2003 NAIP Aerial Photo
25
Soil Sampling for VR Lime
Collect yield
and NDVI
data
Load data into
GIS software
Visually and statistically
interpret variability
Develop soil
sampling regions
from analysis
7/16/2015
Load sample site
locations into
hand held device
Collect soil samples from
field and send to lab
Load lab results
into GIS software
Analyze lab results and
generate VR lime map
26
Yield and NDVI Data Collected
with Sensor Tools
Shown in GIS Software
Grain Yield, 2005
Average: 49 bu/ac
CV: 0.21
7/16/2015
GreenSeeker NDVI, Dec 2004
Average: 0.65
CV: 0.12
27
Analytical Check List
• Look for large scale variability
• Identify common trends shown in both yield
and NDVI maps
• Identify small scale areas with low variability
• Identify small scale areas with high variability
– CAUTION: it is the most difficult to collect reliable
soil samples from small areas with high variability
7/16/2015
28
Analyzing Variability and
Determining Sampling Sites
Mean=54.65 bu/acre Mean=44.12 bu/acre Mean=0.69 NDVI
Mean=0.31 NDVI
CV=0.17
CV=0.13
7/16/2015
CV=0.20
CV=0.10
29
1.Define regions based
off visual
interpretation.
CV = 0.083
0.119
2.Analyze regions’
variability using CV.
0.084
0.105
0.108
• Total Regions: 22
• Mean CV: 0.110
• Max CV: 0.167
• Min CV: 0.078
0.120
• CV>0.11: 7
0.120
• Use yield data to
further characterize
potential variability.
0.094
0.093
0.102
0.110
0.073
0.092
7/16/2015
0.137
30
1.Calculate CVs from
yield data for each
region.
CV = 0.16
0.15
0.12
2.Use CVs to reaffirm
high or low
variability.
0.20
0.18
3.Divide regions where
necessary based on
yield data.
• Mean CV: 0.15
0.16
• Max CV: 0.20
0.19
• Min CV: 0.10
• CV>0.17: 6
0.17
0.16
0.18
0.15
0.14
0.14
0.19
7/16/2015
31
Two options for regions with potentially high variability:
1. Increase number of samples (i.e. instead of 16 samples per region, use 32)
2. Divide regions into smaller areas with more homogeneity
7/16/2015
32
New Region Layout
7/16/2015
33
Final Layout
Total Regions: 30
Small Regions: 5
•30 Samples @
$10 each=$300
7/16/2015
34
pH Results
Lime Application Map
5.7
6.4
0
5.3
6.7
5.6
5.5
0
5.6
5.9
5.8
1 ton/acre (50% ECCE)
5.7
6.2 7.8
6.8
6.5
7.7
5.8
7.0
6.7
6.9
5.3
5.8
5.8
6.6
6.0
5.3
7.1
6.6
7.1
0
0
0
~115 acres needs 1 ton/acre of lime
7/16/2015
35
Verify Using Intensive Soil
Sampling
• Divide into small sized regions using
multiple years of yield, NDVI, and Veris
(EC) data.
• Recalculate VR lime application map using
intensive soil sample results.
• Compare to previous method that used
only 30 regions (30 soil samples).
7/16/2015
36
7/16/2015
2005
2006
49 bu/ac
16 bu/ac
37
Dec 2004
7/16/2015
April 2006
38
NDVI 2006
7/16/2015
Yield 2006
Veris EC 2006
39
Soil Samples: 217
Approximately ¾
acre per sample
site (similar to grid
sampling).
~$2170
7/16/2015
40
Soil Sample Results (pH Surface Map)
<6.5
>7.0
>6.5 and
<7.0
7/16/2015
41
VR Lime Application Map
Approximately 114 acres need
1 ton of line (50% ECCE).
(41.5 acres needs 0)
@$15/ton, ~$620 savings
Few small areas need 1.5-2
tons of lime.
7/16/2015
42
Large Region Sampling vs. Intensive Sampling
1 ton
1 ton
0 ton
7/16/2015
0 ton
43
Over applied: 12.7 acres
(@$15/ton, ~$190 loss)
Under applied: 9.7 acres
(@2 bu/ac loss, @$9/bu,
~$175 loss
Total loss: ~$365
7/16/2015
44
Conventional Method
• Total Samples: 12
(1 lab sample)
• Avg pH: 5.9
• Max pH: 7.5
• Min pH: 5.0
• BI: ~6.7
• Recommendation
Rate: 1ton/acre
(50% ECCE)
• Lime Cost: $2332
7/16/2015
45
Conclusion
• Sensor Data (i.e. yield and NDVI) can be
used to assess variability in a field
• Site sampling strategies can be
significantly improved using sensor data,
especially for pH management and forage
analysis.
• Improved site sampling has similar results
to intensive sampling, with significantly
reduced costs.
7/16/2015
46