Risk Factors

Risk factors

Every Canopi prediction includes risk factors — the top environmental conditions most strongly influencing tree survival at that specific site. These aren’t generic warnings. They’re derived from the model’s actual decision process for that exact combination of location, species, method, and time horizon.

How risk factors work

Canopi uses SHAP (SHapley Additive exPlanations) to decompose each prediction into the contribution of every input feature. SHAP assigns each feature a value indicating how much it pushed the survival probability up or down relative to the baseline average.

From the full set of 17 features, the API returns the top 3 by absolute magnitude, filtered to ecologically actionable factors. Features like state code and species code are excluded from the risk factor output because they aren’t factors you can influence or plan around — they’re structural inputs, not site risks.

Reading risk factors

Each risk factor has three components:

{
  "factor": "Vapor pressure deficit",
  "impact": "negative",
  "magnitude": 0.089
}

• factor — The environmental condition, in plain language.
• impact — Whether this factor is pushing survival probability down ("negative") or up ("positive").
• magnitude — How strong the effect is. Higher magnitude means more influence on the prediction.

A negative-impact factor is a risk to survival at this site. A positive-impact factor is a condition supporting survival.

Common risk factors

Vapor pressure deficit — A measure of atmospheric dryness. High VPD means the air is pulling moisture from leaves faster than roots can replace it. This is the primary driver of drought stress in seedlings and shows up frequently as a risk factor on dry, exposed sites.

Precipitation — Annual precipitation at the site. Can be a positive factor (sufficient moisture) or a negative one (insufficient rainfall). Very high precipitation can also be negative in some contexts (waterlogging, fungal disease).

Soil water capacity — The soil’s ability to retain plant-available moisture. Sandy or rocky soils with low water capacity increase drought risk, especially during the critical first growing season.

Dew point temperature — Related to atmospheric moisture. Low dew points indicate dry conditions. This factor often appears alongside vapor pressure deficit on moisture-limited sites.

Elevation — Affects temperature regimes, frost exposure, and growing season length. High-elevation sites face shorter growing seasons and greater cold stress.

Soil organic matter — Higher organic matter generally improves water retention and nutrient availability. Low organic matter can be a risk factor on disturbed or degraded sites.

Crown ratio — Reflects the vigor of existing trees at similar sites in the training data. Lower crown ratios suggest stressed conditions. This factor captures site productivity signals that other features may miss.

Soil drainage — How quickly water moves through the soil profile. Both extremes (excessively drained or poorly drained) can be negative depending on species.

Risk factors change with time

The same site may show different dominant risk factors across horizons. First-year risk is often acute stress (drought, frost, competition). Five-year risk may shift toward chronic conditions (sustained climate trends, soil limitation, disease pressure). Compare risk factors across horizons to understand how the threat landscape evolves.