Chapter 15: Modelling Natural Disturbances

Learning Objectives

After reading this chapter, you should be able to:

  • Explain the role of natural disturbances in forest estate modelling

  • Distinguish deterministic, stochastic, and scenario-based disturbance approaches

  • Implement disturbance transitions and salvage logic in a ws3-style model

  • Design resilience-oriented indicators for plans under disturbance uncertainty

  • Connect disturbance outputs to downstream spatial and supply-chain analysis

Why Model Disturbances?

Natural disturbances (wildfire, insect outbreaks, windthrow, drought-related mortality) are first-order drivers of long-term forest dynamics. Strategic plans that ignore disturbance dynamics often overstate sustainable harvest and understate system risk.

Ignoring disturbance can lead to:

  • Over-optimistic harvest schedules: Models assume all inventory remains intact, but disturbances regularly remove it

  • Inadequate reserve planning: No buffer for disturbance-related inventory losses

  • Poor risk assessment: Cannot quantify the probability of meeting harvest targets

  • Weak adaptation planning: Silviculture and salvage decisions are not stress-tested against plausible disturbance regimes

        graph TD
 FOREST["Forest landscape state"] --> FIRE["Wildfire"]
 FOREST --> INSECT["Insects/pathogens"]
 FOREST --> WIND["Wind/ice events"]
 FOREST --> DROUGHT["Drought stress"]
 FIRE --> IMPACT["Mortality, quality loss, access constraints"]
 INSECT --> IMPACT
 WIND --> IMPACT
 DROUGHT --> IMPACT
 IMPACT --> TRANSITIONS["New transitions + salvage options"]
 TRANSITIONS --> PLAN["Updated harvest, risk, and regeneration plan"]

Disturbance Modelling Paradigms

Three common paradigms are used in strategic forest modelling.

  1. Deterministic deductions A fixed annual loss percentage is applied to affected strata. This is simple and fast, but masks volatility and spatial clustering.

  2. Stochastic event simulation Random events are sampled from probability distributions (occurrence, size, severity). This captures variability and tail risk but is computationally heavier.

  3. Scenario envelopes Several plausible disturbance futures are imposed (e.g., low/medium/high fire pressure). This is practical for policy communication and sensitivity analysis.

In practice, organizations often combine 2 and 3: stochastic simulation inside named scenario envelopes.

Disturbance Types

Wildfire

Wildfire is often modelled as a hazard process with event size and severity sub-models.

Key dimensions:

  • Probability: Annual probability of ignition per hectare

  • Spread/size: Distribution of fire sizes and weather-conditioned spread

  • Severity: Surface vs crown effects with class-specific mortality

  • Seasonality: Higher risk in dry, hot periods

# Example hazard proxy for scenario screening.
# Use calibrated regional models in production.

def wildfire_probability(age, moisture_index, slope_aspect):
    """Estimate annual wildfire probability."""
    # Fuel load proxy rises with stand age.
    base_prob = 0.001 * (1 + age / 100)

    # Moisture reduces risk.
    moisture_factor = max(0.1, 1 - moisture_index)

    # South/west aspects are often drier.
    aspect_factor = {"S": 1.2, "SW": 1.1, "W": 1.0,
                     "NW": 0.9, "N": 0.8, "NE": 0.85,
                     "E": 0.9, "SE": 0.95}.get(slope_aspect, 1.0)

    return base_prob * moisture_factor * aspect_factor

Insect Outbreaks

Insects and pathogens can generate multi-year mortality pulses and quality degradation.

Typical features:

  • Host specificity: Some insects attack specific species

  • Cyclicity: Outbreak cycles can repeat over decades

  • Threshold effects: Damage can accelerate beyond density/age thresholds

  • Climate sensitivity: Warmer winters reduce insect mortality

Windthrow

Wind and ice damage produce abrupt structural loss, especially in exposed stands.

Important predictors:

  • Exposure: Coastal and ridge-top stands are more vulnerable

  • Soil depth: Shallow soils increase uprooting risk

  • Tree height: Taller trees are more susceptible

  • Frequency: Return intervals of 50-200 years for major events

From Disturbance to Transition Logic

In an aspatial strategic model, disturbance effects are typically represented by area transfers between development types, with optional salvage pathways.

Conceptually, for development type \(d\) in period \(t\):

\[A_{d,t+1} = A_{d,t} - H_{d,t} - D_{d,t} + R_{d,t} + S_{d,t}\]

where:

  • \(A_{d,t}\) is standing area,

  • \(H_{d,t}\) is planned harvest,

  • \(D_{d,t}\) is disturbed area,

  • \(R_{d,t}\) is regeneration inflow,

  • \(S_{d,t}\) is salvage-related transfer.

This bookkeeping is the core of disturbance-aware planning.

Integrating Disturbances with ws3

ws3 can represent disturbance through actions, transitions, and yield impacts. The exact API varies by project conventions, but the pattern is stable.

Common implementation pattern:

  1. State classes: Add post-disturbance development types

  2. Disturbance actions: Define FIRE/INSECT/WIND actions that transfer area

  3. Salvage actions: Define optional salvage transitions with reduced yields

  4. Regeneration pathways: Route disturbed classes into managed recovery

from ws3.forest import ForestModel

# Disturbance actions and transitions are defined in Woodstock-format
# section files and imported into the ForestModel.
#
#   model = ForestModel("disturbance_model", "/path/to/data", 2024,
#                       horizon=20, period_length=10)
#   model.import_areas_section()       # includes post-disturbance DTs
#   model.import_yields_section()
#   model.import_actions_section()     # includes FIRE, SALVAGE actions
#   model.import_transitions_section() # maps DF-SI50 -> DF-SI50-disturbed
#
# The ACTIONS section file defines:
#   *action FIRE Wildfire disturbance
#   *operable df si50 volume
#
#   *action SALVAGE Salvage harvest after disturbance
#   *operable df si50-disturbed volume
#
# The TRANSITIONS section file defines:
#   *case FIRE
#   *source df si50
#   *target df si50-disturbed
#
#   *case SALVAGE
#   *source df si50-disturbed
#   *target df si50-regen

# The simulation controller applies FIRE stochastically and SALVAGE
# according to policy and operational constraints.

Scenario Design for Disturbance Planning

A useful strategic experiment set usually varies four levers:

  • Disturbance intensity: expected annual disturbed area

  • Disturbance severity: merchantability loss and regeneration delay

  • Operational response: salvage capacity, access constraints, replanting lag

  • Climate trend: non-stationary shift in disturbance frequency/severity

Example scenario matrix:

Disturbance Scenario Envelope

Scenario

Disturbance rate

Salvage capacity

Interpretation

S1 Baseline

Historical mean

Current

Continuation of recent regime

S2 Elevated fire

+30%

Current

Stress-test under warmer/drier conditions

S3 Elevated + response

+30%

Expanded

Tests operational adaptation investment

S4 Compound risk

+30% fire + outbreaks

Expanded

Multi-disturbance pressure case

Resilience Metrics

Beyond total harvest, disturbance-aware plans should track resilience metrics:

  • Reliability: probability of meeting minimum harvest commitments

  • Recovery time: years to return to baseline harvest capacity after shock

  • Structural diversity: area balance across age/species cohorts

  • Salvage dependence: share of realized harvest from salvage rather than planned entries

  • Regeneration debt: deferred re-establishment area

These metrics often reveal plan fragility that average-volume indicators hide.

Challenges

Disturbance modelling remains difficult because:

  1. Stochasticity: Disturbances are inherently random

  2. Scale mismatch: Strategic models are often aspatial while disturbance is spatially clustered

  3. Data sparsity: Long clean time series are rare for many regions and agents

  4. Non-stationarity: Climate trend breaks historical frequency assumptions

  5. Compounding effects: Fire-after-beetle, drought-after-thinning, and access failures interact

Worked Example: Simple Disturbance Stress Test

Assume a planning unit with 100,000 ha operable area and baseline planned harvest capacity of 2,000 ha/year.

  • Baseline disturbance: 1.0%/year (1,000 ha/year)

  • Elevated disturbance: 1.5%/year (1,500 ha/year)

  • Salvageable fraction: 40% of disturbed area

  • Salvage utilization: 70% of salvageable area

Under elevated disturbance:

  1. Disturbed area increases by 500 ha/year.

  2. Salvage recovered area is \(500 * 0.4 * 0.7 = 140\) ha/year.

  3. Net additional unavailable area is approximately 360 ha/year.

If this persists, the model should test whether regeneration and treatment programs can offset this gap fast enough to maintain medium-term flow targets.

Future Directions

Potential extensions to ws3 for disturbance modelling:

  • Risk-constrained optimization: add reliability constraints on key outputs

  • Adaptive re-planning: periodic parameter updates from monitoring streams

  • Coupled models: integrate with SpaDES for event realism and feedback

  • Policy levers: explicit salvage, reserve, and regeneration investment controls

Exercises

  1. Easy: Add a disturbed state class and define FIRE and SALVAGE transitions for one species/site class in a toy model.

  2. Medium: Build three disturbance scenarios (baseline, elevated, elevated with increased salvage) and compare reliability and recovery time.

  3. Hard: Formulate a planning objective that maximizes discounted harvest value subject to a minimum probability of meeting flow commitments under stochastic disturbance.

Further Reading