Weather-Disease Modeling

Instructions

Disease Risk Threshold Reference

Implement disease models using these established phytopathology thresholds:

Late Blight (Phytophthora infestans)

1. Blitecast model (modified):

• Track consecutive hours where: temperature 45–80°F AND relative humidity ≥ 90%
• Low risk: < 6 consecutive hours meeting conditions
• Moderate risk: 6–10 consecutive hours
• High risk: > 10 consecutive hours
• Severity values: accumulate daily; trigger spray advisory at severity value ≥ 18

1. Rain interaction: any rainfall event > 0.1 inches during a high-humidity period escalates risk by one level

Powdery Mildew (Erysiphales)

1. Favorable conditions:

• Temperature: 60–80°F (optimum 70°F)
• Relative humidity: 40–100% (spore germination does NOT require free water)
• Inhibited by: temperatures > 95°F, direct rainfall washing spores off leaves

1. Risk calculation:

• Count hours per day in the 60–80°F range with RH > 50%
• Low risk: < 6 hours/day
• Moderate risk: 6–12 hours/day
• High risk: > 12 hours/day for 3+ consecutive days

Downy Mildew (Peronosporaceae)

1. Favorable conditions:

• Temperature: 50–75°F (optimum 65°F)
• Leaf wetness duration: ≥ 6 hours
• High humidity (> 85%) combined with cool nights

1. Risk triggers:

• Night temperature drops below 65°F AND morning leaf wetness persists > 4 hours
• 3+ consecutive days meeting conditions = high risk

Bacterial Spot (Xanthomonas)

1. Favorable conditions:

• Temperature: 75–86°F
• Leaf wetness: ≥ 12 hours (rain-splashed transmission)
• Wind-driven rain dramatically increases spread

1. Risk calculation:

• Track days with: max temp > 75°F AND any rain event AND leaf wetness > 8 hours
• High risk: 2+ qualifying days in a 5-day window

Growing Degree Day (GDD) Calculations

1. Standard formula:

   GDD = max(0, ((T_max + T_min) / 2) - T_base)

1. Base temperatures by crop:

• Tomato: 50°F
• Pepper: 55°F
• Corn: 50°F
• Squash/Cucumber: 50°F
• Lettuce/Greens: 40°F

1. Accumulation tracking:

• Sum GDD daily from a defined biofix date (transplant date or emergence date)
• Use accumulated GDD to predict: days to flowering, days to fruit set, days to maturity

1. Upper threshold cutoff: cap T_max at 86°F for most vegetables (growth slows above this)

Disease Severity Index (DSI)

1. Composite scoring (0–100 scale):

   DSI = (w1 × late_blight_score) + (w2 × powdery_mildew_score) +
         (w3 × downy_mildew_score) + (w4 × bacterial_spot_score)

Default weights: w1=0.35, w2=0.20, w3=0.25, w4=0.20

1. Individual disease scores (0–100):

• Map each disease’s multi-day risk accumulation to a 0–100 scale
• Low risk days contribute 0–2 points
• Moderate risk days contribute 3–5 points
• High risk days contribute 6–10 points

1. Decay function: reduce accumulated score by 10% per day when conditions are unfavorable
2. Alert thresholds:

• DSI 0–25: Green (normal monitoring)
• DSI 26–50: Yellow (increase scouting frequency)
• DSI 51–75: Orange (preventive treatment recommended)
• DSI 76–100: Red (active treatment required)

Weather Data Integration

1. Required data points (minimum hourly):

• Temperature (°F or °C)
• Relative humidity (%)
• Precipitation (inches or mm)
• Wind speed (mph or km/h)
• Leaf wetness duration (hours) — estimate from dew point if sensor unavailable

1. Leaf wetness estimation when no sensor:

• If RH ≥ 90% AND temperature is within 3°F of dew point → assume leaf wetness
• Duration = consecutive hours meeting the above condition

1. Data sources:

• Weather API (OpenWeatherMap, Visual Crossing, Tomorrow.io)
• Local weather station (Davis Instruments, Ambient Weather)
• National Weather Service API (forecast.weather.gov) for free forecasts

1. Forecast integration:

• Pull 7-day hourly forecast data
• Run disease models on forecast data to produce predictive risk scores
• Label forecast-based scores distinctly from observed-data scores

Implementation Steps

1. Define crop profiles: map each crop to its susceptible diseases and GDD base temperature
2. Build weather data pipeline: fetch hourly observations and forecasts, normalize units
3. Implement individual disease models: one function per disease returning a 0–100 score
4. Calculate composite DSI: weighted sum with configurable weights per crop
5. Generate alerts: compare DSI to thresholds, produce user-facing advisory messages
6. Store historical data: retain daily DSI values for season-over-season comparison
7. Visualize: time-series chart of DSI with color-coded zones and forecast projection

Inputs Required

• Geographic location (latitude/longitude or zip code for weather data)
• Crops being grown (for disease susceptibility mapping and GDD base temps)
• Weather data source API credentials
• Planting/transplant dates (for GDD accumulation start)
• Optional: local weather station data for higher accuracy

Output Format

Disease Risk Advisory

{
  "date": "2026-04-14",
  "location": { "lat": 33.749, "lon": -84.388 },
  "data_source": "observed" | "forecast",
  "gdd": {
    "daily": 12.5,
    "accumulated": 342.0,
    "base_temp": 50,
    "crop": "tomato"
  },
  "disease_scores": {
    "late_blight": { "score": 35, "level": "yellow", "trend": "rising" },
    "powdery_mildew": { "score": 12, "level": "green", "trend": "stable" },
    "downy_mildew": { "score": 48, "level": "yellow", "trend": "rising" },
    "bacterial_spot": { "score": 8, "level": "green", "trend": "falling" }
  },
  "composite_dsi": 28,
  "dsi_level": "yellow",
  "advisory": "Increase scouting frequency. Downy mildew pressure building — monitor morning leaf wetness closely.",
  "recommended_actions": [
    "Scout for downy mildew symptoms on lower leaves",
    "Consider preventive copper spray if forecast shows continued cool wet nights"
  ]
}

Anti-Patterns

• Using daily averages instead of hourly data — disease models depend on consecutive-hour thresholds; daily averages mask critical periods
• Ignoring leaf wetness duration — humidity alone is insufficient; leaf wetness duration is the key driver for most fungal diseases
• Hardcoding disease weights — different crops have different susceptibility profiles; weights must be configurable
• Treating forecast scores the same as observed scores — always label the data source; forecast-based scores carry more uncertainty
• Skipping the decay function — without decay, scores only go up, creating permanent false alarms
• Using a single weather data point per day — hourly granularity is the minimum for accurate disease modeling
• Alerting without actionable recommendations — a risk score without treatment guidance creates anxiety, not action