q_agro results · deployment adaptation

Background

YieldSAT dataset coverage map and table — **Dataset coverage.** Miranda et al. (2026). YieldSAT field coverage across countries, crops, years, and Sentinel-2 images.

YieldSAT data modalities for crop yield modeling — **Data modalities.** Miranda et al. (2026). Satellite, weather, soil, and terrain inputs used for field-level yield modeling.

Najjar-style multimodal yield prediction pipeline — **Base multimodal model.** Najjar et al. (2025). Transformer-based multimodal crop-yield model used as the base encoder.

Calibration vs analogue correction

Calibration

global shift

Same correction for all fields.

+β_tField A

+β_tField B

+β_tField C

+β_tQuery

Uses alllabeled target calibration fields estimate one shared bias shift.

Analogue correction

local analogue vote

Query-specific analogue correction.

w .52Top-1

w .31Top-2

w .17Top-3

local ΔQuery

Uses Top-3analogues are ranked by agro-feature distance; closer fields get larger weights.

Why scenes differ

Fixed q_agro retrieval space for LOYO and LORO domain structure — **LOYO/LORO structure.** Retrieval space separates temporal and regional shifts.

Main result

label efficiency across crops and scenes

Label-efficiency of residual domain adaptation policies — **Label efficiency.** Analogue correction improves RMSE with few target labels; low-SR LORO remains hardest.

Modality evidence

Signed q_agro Shapley utility by modality — **Signed Shapley utility by modality.**← worse0better →

All-subset modality audit for q_agro retrieval — **All-subset modality audit.**← subset better0full better →

Decision workflow

leakage-safe deploy-time evidence

Base model

Start from the Najjar-style multimodal Transformer prediction.

Target calibration

Estimate target-domain bias from a small labeled calibration bank.

Analogue retrieval

Find similar fields in target-domain agro-feature space.

LLM audit

Review LOO self-test, geometry, neighbour consistency, modality agreement, and historical prior.

Prefix-k rule

Trim distant neighbours or fall back when evidence is not reliable.

Recommendation

Apply analogue correction or keep calibration-only at deploy time.

Compact result table

mean ± std field-level RMSE

crop	protocol	SR	base MML	target-cal	DA q_agro	DA binary	best	TC - q_agro	TC - binary

Analogue correction as decision support