Crop Yield Estimation

Early forecasts (V6–V8 in maize, flag leaf emergence in wheat) are most influential for management changes but have high uncertainty (RMSE ~30–40% of mean yield). Mid-season forecasts (grain fill stage) have moderate precision (RMSE ~15–20%). Late-season forecasts (R5–R6 in maize, soft dough in wheat) have good accuracy (RMSE ~10%) but limited time for adjustment. Use early forecasts for strategic decisions (irrigate?), late forecasts for tactical decisions (harvest timing, marketing).

Use grid or stratified random sampling: divide field into zones (e.g., 4–10 zones based on topography, soil, or management), then sample randomly within each zone. Sample 10–15 locations per hectare minimum. At each location, count plants in a 1 m² quadrat (or smaller if plants are dense), count grain-bearing organs, and estimate kernel number. Average across replicates for field estimate.

Leaf area at flowering (LAI, leaf area index) correlates moderately with yield (R² = 0.4–0.6) because photosynthetic area drives grain fill. However, late-season LAI is less informative than mid-season LAI; most grain is filled 2–4 weeks after flowering, so stress after this window has minimal yield impact. NDVI or leaf color can estimate LAI non-destructively.

Regression models using weather can explain 40–60% of yield variation, useful for large-scale forecasts (regional, national) but less precise for individual fields. Local soil properties, cultivar, and management explain the remaining variance. Combine weather models with field sampling for best accuracy: weather models set the baseline expectation, field sampling fine-tunes the prediction.