Case Study Evidence of New
Opportunities for Farm Management
Specialists in Spatial Analysis of Onfarm Trial Data
Terry Griffin 1 , Craig Dobbins 2 , Jess Lowenberg-DeBoer 2
1 University of Arkansas – Cooperative Extension Service
2 Department of Agricultural Economics, Purdue University
National Farm Management Conference
• On-farm trials often violate statistical assumptions
• Farmers continue to conduct on-farm trials
– PA and GPS lead to resurgence of on-farm trials
– Harvested with yield monitors w/out interference
• Reduced public research funding
– More weight on local on-farm experiments
• Farmers’ objective: make best decision
• Better understand the motivations of farmers for conducting field-scale experiments and document their perceptions of spatial analysis
• Propose plan to Extension and industry for spatial analysis service
• Case studies evaluated perceptions of conducting on-farm research and spatial analysis
• Qualitative research methods
• Lack of general information to conduct quantitative survey
• Reference and comparison group farmers
– IL, IN, KY, Ontario
• Direct 3 year observation of reference group
– During farm-visits, frequent communication
– Farmers conducting field-scale on-farm trials
– Spatial analysis reports provided to farmers
• One-on-one interviews of both groups
• Two yield monitor data analysis workshops
• Spatial statistics assume that data is spatially correlated and explicitly included in analysis; in contrast to independent observations assumption.
• Yield monitor and site-specific data is spatially correlated. If that correlation is not accounted for in the analysis, results will be biased and misleading.
• Yield monitor data with appropriate spatial analysis can lead to more reliable decision making with limited replications.
• Description of on-farm trial
– Design, treatment, data available
• Data handling procedures
– Yield data filtering, data assimilation
• Spatial statistical analysis and diagnostics
• Economic analysis
• Production recommendation
• Central Indiana soybean seeding rate trial
– 80, 100, 120, 140, and 160K seeds per acre
– 4 replications in 1700 foot strips
– 30 inch rows
• End result is more reliable information
– A production recommendation
– Not a map
Photo: Griffin – Twilight Farms
Raw yield monitor data
• As-is from the combine
• No cleaning or filtering
Yield data in GIS after removing erroneous observations
Yield data in GIS after removing erroneous observations
Study area
Yield monitor data used in analysis
Rate trial: 80K to 160K seeds per acre
Four replications of five population rates
1
2
3
4
{
{
{
{
Major soil
Secondary soil
All five rates on each soil “zone”
{ Minor eroded soil
Spatial Error Estimation
Variable
CONSTANT
POP
POP_SQ
Soil
3
Soil
2
POP*Soil
3
POP_Soil
2
ELV
*Soil
3
ELV
*Soil
2
POP*ELV
ELV
LAMBDA
Spatial error
Coefficient t-Statistic
64.141***
0.176***
17.38
3.50
-0.001***
-246.321
-356.830***
-2.81
-1.01
-3.92
0.117***
0.023***
0.271
0.417***
-0.002
-0.500***
0.781
5.44
2.59
0.94
3.86
-1.49
-3.21
63.11
70
Major soil :
100K profit max
Major soil:
130K yield max
65
60
Secondary soil:
150K yield max
Secondary soil:
120K profit max lowering seeding population from 130K to about 100K on most of the field, increasing planting timeliness
120 140
Seeding rate (000 seed ac
-1
)
160
Major soil Secondary soil Minor soil
180
• November 2005 and March 2007
• Farmers, consultants, university personnel
• Farmers
– some farmers perform own spatial analysis
• University Extension
– Technical skill, but can only work directly with a few farmers
• Private industry
– farmers, co-op, dedicated analysts, consultants
• Assist with designing experiments
• Network of research collaborators
– Regional research projects
• Continued education/training for analysts
• Troubleshooting and problem solving
• Teaching interpretation of analysis
• Assist with decision making
• Already dealing with vast amounts of data
• Familiar with assisting interpretation of results
• Assist with whole-farm decision making
• Will farmers be willing to pay a fee that entices qualified analysts to offer service?
Farmer D F W P T
What would you expect to pay for full-service spatial analysis?
$3 per acre
$5 per acre or $500 per trial.
Percent of predicted value. Up to several hundred dollars.
$2 per acre.
Up to $500 per trial.
$5 to $10 per acre or 40 to
50% of payback.
Third-party Spatial Analysts Questions
• What software to assemble the data?
• What software to statistically analyze the data?
– R, GeoDa, Stata, SpaceStat, MATLAB, SAS
• What type of spatial analysis conducted?
– How characteristics of neighboring data used?
• Where received spatial analysis training?
• How confident are you in the results?
• Does client receive copy of original or raw data?
• Is focus on providing maps or recommendations?
• More confidence in data and decisions when using spatial analysis
– Farmers made more decisions more quickly
• Unclear who will be primary analysis provider
• Role for Extension and private sector industry
– Depends upon farmers’ willingness to pay
– Will any tax money be put into this?
– Are any public goods generated?
• Farm management specialists may be among the first to demonstrate benefits of the yield monitor
Wish to thank USDA-SARE for providing funding to evaluate alternative on-farm designs and develop spatial analysis methods
Terry Griffin tgriffin@uaex.edu
501.249.6360(O)
Craig Dobbins cdobbins@purdue.edu
Jess Lowenberg-DeBoer lowenbej@purdue.edu