Brazil Corn Safrinha Yield Outlook

1. Select Harvest Year - Choose the corn safrinha harvest year (e.g., 2025 = planted Feb 2025, harvested Jul-Aug 2025)
2. Choose Analysis Mode - Quick (5 municipalities), Standard (25), or Full (50)
3. Set Brazil Area - Input total Brazil corn safrinha area (kha)
4. Run National Analysis - Click button to analyze all 5 states automatically

Core-Safrinha Proxy - Estimates Brazilian 2nd crop corn by analyzing the structural center of Safrinha production (PR, SP, MG, GO, MS, MT) with defensible methodology.

Why This Approach:

• Geographic Focus: These 6 states represent the core Safrinha production region
• Dynamic Selection: Top municipalities chosen by IBGE corn production (not hardcoded)
• Pooled Calibration: Single β sensitivity learned across all core states
• Transparent Process: Full methodology and assumptions clearly displayed

Output:

• Core-region yield & production estimate with confidence intervals
• Brazil extrapolation with explicit assumptions stated
• State-by-state breakdown with quality diagnostics
• Complete defensibility documentation (coverage %, calibration quality, etc.)

• Custom Harvest Year: Select "Custom year..." from harvest year dropdown to type any year (2026, 2027, 2028+). Useful for planning future seasons or analyzing historical years not in the dropdown list.
• Custom Trend Period: Select "Custom" from trend calculation period dropdown to specify exact start and end years. Allows precise control over historical data range used for trend calculation.

Tips: Custom inputs update labels in real-time and validate input ranges. Invalid entries are clearly marked. Both features work seamlessly with all analysis modes and trend models.

• Quick (~30s): 2 municipalities per state (12 total)
• Standard (~90s): 5 municipalities per state (30 total)
• Full (~3min): 10 municipalities per state (60 total)

Important: Modes control analysis granularity only. Calibration (β learning) always uses 10 municipalities per state regardless of mode selected, ensuring consistent calibration quality across all modes.

Note: Rate limiting delays prevent API errors. More municipalities = better statistical reliability for predictions.

• Precipitation Anomaly (default): Measures cumulative deviation from normal rainfall. Formula: Σ(P_observed - P_climatology). Simple and intuitive - positive means wetter than normal.
• Water Balance: Accounts for both precipitation AND evaporative demand. Formula: Σ(P - ET₀×Kc) where Kc is crop coefficient varying by growth stage. More physiologically meaningful as it captures actual water available to the crop.

When to use each:

• Precipitation Anomaly: Good baseline, works well when ET₀ is relatively constant or when you want simpler interpretation.
• Water Balance: Better during hot/dry seasons when evaporative demand varies significantly. Captures drought stress more accurately.

Important: Each signal type has its own calibration (β, μ, σ). Switching signals triggers re-calibration. Clear cache if switching doesn't show different results.

• IBGE: Historical yield, production, area per municipality
• Open-Meteo: Current + forecast precipitation, ET₀, climatology baseline
• Calendar: Embedded planting dates (DAP0) per state/year

• Trend baseline (2000-2024): IBGE yield data fitted with selected trend model to estimate y_trend per municipality
• β calibration (2005-2024): Climate sensitivity learned by regressing yield residuals on Z-scores, pooled per state
• Climatology reference (1981-2010): WMO standard 30-year period for "normal" precipitation/ET₀ (P_clim, ET₀_clim)

Why different ranges? Trend uses more years for stability. β calibration starts 2005 due to Open-Meteo data availability. Climatology uses standard WMO reference period.

Backtesting: When analyzing past harvest years, the calibration period automatically excludes the target year to prevent data leakage. For example, predicting 2025 will use calibration from 2005-2024 maximum.

• Calibration Guardrails: Minimum n≥10 observations per state, MAD≤1.5 threshold
• Area-weighted Z: Uses area for aggregation (not production) to avoid circular feedback
• MAD Diagnostics: Median Absolute Deviation measures fit quality; lower = better calibration
• Trend-only Flag: Municipalities lacking climate data use trend-only estimates

Status indicators: OK (Z≥-1.0), Watch (-1.5≤Z<-1.0), Alert (Z<-1.5)

• Configuration Display: Shows baseline window, custom harvest year, custom trend period, DAP end, lag settings used
• Sample Quality: Municipality count, coverage %, trend-only %
• Calibration Metrics: Per-state MAD, n observations, fit quality
• Guardrail Status: Validation results and recommendations

Export: State details expandable for detailed per-state calibration quality

Running the analysis weekly during the season produces different results because:

• Observed data accumulates: Each week, more actual rainfall replaces forecast data
• Forecast window shifts: The 16-day forecast always looks ahead from today
• CumDev updates: Cumulative precipitation anomaly changes as real weather unfolds

What stays constant: Trend baseline (y_trend), climate sensitivity (β), municipality selection, and climatology reference.

Accuracy improves over time: Early-season runs have more forecast uncertainty; late-season runs reflect actual weather.

• Upward bias: Top municipalities have higher yield than state average
• Point-based climate: Open-Meteo uses municipal centroids
• Coverage: 87% of Brazil production (5 states)
• Rate limits: Re-run if 429 errors occur

IBGE Baseline Data: Historical area and yield data from IBGE reflects total corn production (all seasons combined), not safrinha-specific data.

• MT/GO/MS: Safrinha = ~85% of total corn (good proxy)
• PR/MG: Safrinha = ~60% of total corn (less precise)

Weather Analysis: Uses safrinha-specific parameters:

• Planting calendar: Feb-Mar (after soybean harvest)
• Critical window: VT-R3 in Jun-Jul (dry season drought risk)
• Crop coefficients: Corn-specific Kc values

Control how climate signals are applied to trend predictions:

• Pre-Season Mode: When enabled, uses trend-only prediction before Feb-Mar planting (recommended)
• Analysis Mode:
  • Climate-Adjusted: Full climate signal applied (current behavior)
  • Trend-Only: Always use historical trend, never apply climate
  • Hybrid: Apply climate when confident, trend when data is sparse
• Climate Weighting: Adjust signal strength from 0% (trend-only) to 100% (full signal)

Trend Model Types:

• Theil–Sen: Robust linear regression (default)
• OLS: Ordinary Least Squares linear regression
• EWMA: Exponentially Weighted Moving Average
• Piecewise: Linear with breakpoint detection
• Yield-Dependent: Non-linear trend that adapts based on yield level (NEW)

Yield-Dependent Trend: Implements saturating technology model where the rate of yield improvement depends on the current yield level. Uses sigmoid function to model diminishing returns as yields approach technological limits.

Use Cases:

• Pre-Season (Jan): Use "Trend Only" mode for planning
• Planting (Feb-Mar): Use "Climate-Adjusted" with 50% weighting
• Growing Season: Use "Climate-Adjusted" with 100% weighting