FieldScale Sensor Evaluation

1
Field-Scale Sensor Evaluation

• Ken Sudduth, Newell Kitchen, Scott Drummond
• USDA-ARS
• Columbia MO

2
Objectives

• Investigate row-to-row variability in field-scale
  reflectance sensor data
• Document differences between data collected with
  Holland Scientific Crop Circle (amber) and NTech
  GreenSeeker (green)

3
N Application System

• 6-row system with sensors mounted over rows 2 and
  5
• System tested on 7 producer sites in 2004

4
Control Hardware
Green GreenSeeker 1
Green GreenSeeker 2
Crop Circle 3
Crop Circle 4
GPS
Laptop Computer
Application Control System
Stored Data All sensor data GPS data Processed
data Valve commands
1x, 2x, and 4x Solenoid Valves
5
Analysis of Response Plot Sensor Data

• Each field site included two strips of N-rate
  response plots
• Reflectance data were collected at the time of
  sidedress N application
• Mean reflectance ratio and NDVI were calculated
  for each of the four sensors for each 50-foot plot

6
Response Plot Reflectance Ratio Data

• N application at the Diederich (D) field was done
  near dusk, with only diffuse lighting. Work at
  all other field sites was completed before 6 pm.

7
Response Plot Reflectance Ratio Data
8
Response Plot Reflectance Ratio Data
9
Response Plot Reflectance Ratio Data

• Row-to-row differences are apparent
• Is there an ambient light effect?
• Within a row, relative differences in sensor
  output are generally consistent between sensor
  types
• Scaling differences are apparent between sensor
  types
• Amber reflectance vs. green reflectance?
• Normalize data - divide by mean of each sensor
  reading within each field

10
Normalized Reflectance Ratio Data
Row 5
Row 2
11
Normalized Reflectance Ratio Data
12
Normalized Reflectance Ratio Data
13
Normalized Reflectance Ratio Data

• Within-site, by-sensor normalization removed much
  of the sensor-type variability in many (but not
  all) cases
• In practice, a similar normalization is
  accomplished using reference strip data
• Well-fertilized as opposed to unfertilized
• How well does it work?

14
Comparing Sources of Variation
Row 2
Row 5
Sensor Variation
SE 0.13
SE 0.11
GreenSeeker
Crop Circle
Row-to-row Variation
SE 0.13
SE 0.10
15
Comparing Sources of Variation

• Considerable variability in ratio (or NDVI)
  readings between sensor types
• Mean normalization removed much of the variation
• The remaining variation was of similar magnitude
  as the variation between corn rows 90 inches
  apart
• How many sensors are needed to adequately
  describe variability?
• More in MO where we cant seem to get uniform
  corn stands?

16
Does Between-Sensor Variability Affect N Rate?
17
Does Between-Sensor Variability Affect N Rate?

• In this case, there was not much effect when
  looking at large-scale patterns of N rate changes

18
Does Between-Sensor Variability Affect N Rate?

• Strong relationship between rates from the two
  sensors, but somewhat offset from 11 line

19
Does Between-Sensor Variability Affect N Rate?

• In some fields, GreenSeeker N rate range was
  considerably reduced compared to Crop Circle
• Diederich field was a worst-case example, perhaps
  because of a different relationship between the
  two sensor outputs in low light

20
Summary

• Sensor types are different
• So are individual crop rows, at a similar
  magnitude
• Application rates with the different sensors are
  similar in some field conditions, but not in
  others
• Are sensors interchangeable within algorithms, or
  do we need to consider them as a package?