fieldworkAGRICULTURAL INTELLIGENCEStart a data assessment

WORK / PUBLIC-DATA DEMONSTRATION

Follow the evidence.
Inspect the decisions.

Worked examples from one public field study, followed by prospective trial planning. Real measurements, explicit screening choices and reproducible methods. Inspect both the data decisions and the design assumptions.

Open the interactive study ↗Download analysis sources ↓

Analytical examples

EXAMPLE 01HENRYS / 2018
50.8%

of retained yield records matched

PUBLIC DATA / EXPLORATORY ANALYSIS
AGRICULTURAL DATA / SPATIAL EVIDENCE

Point matching exposes a coverage gap.

Challenge
Application locations and harvest observations were recorded separately. Proximity alone does not guarantee useful coverage.
Approach
Screen yield records and compare spatial matching radii. Use the balanced profile and a 12 m matching radius for this example.
Observed result
30,133 yield records retained; 15,302 matched to application points. The remaining 14,831 stay visible as unmatched.

METHODS / Python · spatial matching · browser maps

Inspect the 2018 report ↗
EXAMPLE 02HENRYS / 2016
88.7%

of retained yield records matched

PUBLIC DATA / EXPLORATORY ANALYSIS
AGRICULTURAL DATA / SPATIAL EVIDENCE

Polygon matching changes the analytical coverage.

Challenge
A second harvest uses application polygons rather than points. It needs a different spatial matching method.
Approach
Apply documented screening and polygon-based matching. Preserve the source geometry and the limitations of the available machine attributes.
Observed result
19,970 yield records retained; 17,709 matched. Speed and recording time were not supplied in this harvest batch.

METHODS / Python · polygon matching · reproducible reporting

Inspect the 2016 report ↗
EXAMPLE 03HENRYS / 2018 / PROTEIN
563

original CropScan protein records

PUBLIC DATA / EXPLORATORY ANALYSIS
AGRICULTURAL DATA / SPATIAL EVIDENCE

Keep measured crop quality distinct from modelled response.

Challenge
Grain protein is a useful crop-quality measurement. It should not be silently treated as a yield measurement or a synchronized treatment response.
Approach
Map the original protein values as a separate layer. Expose recorded date and time on inspection and keep them outside the yield-response model.
Observed result
563 measured protein records available for inspection. No time alignment with yield or causal treatment inference is claimed.

METHODS / CropScan CSV · Python · interactive inspection

Explore the protein layer ↗
METHODS PROJECTQA / TRIAL PLANNING
52,827

original machinery yield records audited

MEASURED QA / CONDITIONAL POWER
MACHINERY DATA / EXPERIMENTAL DESIGN

From coordinate alignment to defensible replication.

Challenge
Inspect the errors and ambiguities in real machine records, then decide what a small yield difference would require from a future trial.
Approach
Preserve a full row audit, align the 2018 coordinate frames, expose quality rules and compare exact paired-block power scenarios.
Observed result
2,724 unique QA-flagged records. An illustrative 2 bu/ac difference requires 52 paired blocks at 80% power if block-difference SD is 5 bu/ac. Final counts need plot-scale variance.

METHODS / Python · GIS QA · exact power · simulation checks

Inspect the project and full report ↗
WHAT THIS DEMONSTRATES

Measurement discipline.
Inspectable spatial analysis.

Original measurements are retained. Screening and matching decisions are exposed, results carry uncertainty, and reports link back to their source.

BOUNDARIES OF THE EVIDENCE

This is an independent demonstration, not a client portfolio or a live sensor feed. Trial assignment and independent experimental units are not established. The 2018 application-rate mass basis is unresolved in the source documentation. No dose recommendation, causal treatment effect or economic return is claimed.

OFPEDATA © 2020 Montana State University, MIT license. Inspect the pinned source manifest ↗

START WITH YOUR QUESTION

Begin with a data assessment.

Define the question, review the available evidence, and identify a practical route to the result you need.

Start a data assessment