When can similar fields improve yield predictions?

A traceable correction for frozen yield models, scored forward in time.

How it works ↓Background, data, evaluation, method. Correct one field →Choose a correction, then see the harvest.

2,062 field-seasons3 crops4 countries2016–2024

Scroll Imagery © Esri, Maxar · illustrative

Compare corrections

The best correction differs by crop.

–individual methods in the chart clearly help all three cropsIn-season · τ1–τ4

Choose from past seasons

Choosing from past seasons reduced error in all three crops.

–crops with lower MSE
  • Corn
  • Soybean
  • Wheat
95% gain intervals above zero In-season · τ1–τ4 Exploratory replay

Understand the error

Year and field shifts account for most Base error.

–of Base MSE is a year or field offset, by cropIn-season · τ1–τ4

Predict earlier

Corn and soybean improve early, wheat later.

Check the risks

Some years and fields get worse.

–of fields with higher MSERange across crops and methodsIn-season · τ1–τ4

Add a few labels

A few labels help wheat the most.

–less wheat error when 1 in 10 fields is labelledIn-season · τ1–τ4

Limitations

Traceable corrections can still fail.

Wheat 2020 · field 37 · full season · chosen after the fact to show a failure

1Similarity can mislead.

Its closest past fields were over-predicted; this one was under-predicted.

2Abstaining is still a decision.

Both Routers corrected a Base that was nearly unbiased on average. Reliably deciding when to keep the Base remains unresolved.

3Traceable is not fully explained.

We can replay the yearly choice and the correction. Explanations of the Router remain unvalidated.

Evidence scope11 test crop-yearsOne benchmarkExamples chosen after the factOne Base per yearSelection rule chosen post hocNo user study

Explorer

Now correct one field yourself.

Pick a correction, set λ, lock. Then see the harvest.

Open the Explorer →