Each line shows the running average of damage-weighted burn area as Monte Carlo runs complete. Lines converge as more runs accumulate — stable convergence by ~60-80 runs indicates sufficient sample size. The dashed baseline (no treatment) provides the control. Weather is sampled independently per run from a fire-season distribution (70% normal summer, 30% Diablo wind events — Moritz et al. 2010).

White lines on the map show firebreak locations — where Burnbot units performed prescribed burns. Each strategy places treatments differently. Click a strategy below to see its spatial pattern. Treatment budget = fleet size × 140 hrs at 1 acre/hr. Scattered strategies lose hours to travel (20 km/h between parcels).

Damage = Σ(burned cells × value-at-risk), where WUI cells carry weight up to 10 and wildland = 1 (Thompson et al. 2022). Reduction = % decrease vs. no-treatment baseline. All strategies share the same 100 ignition points (paired comparison) and weather draws, isolating treatment placement as the sole variable. Green row = lowest mean damage.

Bar height = mean damage across 100 runs. Error bars = ±1 standard deviation. Fire spread uses a simplified Rothermel model (Rothermel 1972; Andrews 2018): ROS = baseROS × moisture damping × slope factor × wind alignment × fuel load. Treatment reduces spread probability by 35–90% depending on wind speed and fuel moisture (meta-analysis benchmark: 62–72%, Fernandes & Botelho 2003).

The fundamental tradeoff: protecting communities (low WUI damage, Y-axis) vs. preserving wildland (low wildland burn, X-axis). Points closer to the bottom-left are better on both axes. The Pareto front reveals which strategies are dominated. This tradeoff framework follows Pais et al. (2021) "Downstream Protection Value" and Thompson et al. (2022) risk-based prioritization.

WUI Saved/Acre = (baseline WUI damage − strategy WUI damage) / acres treated. Higher = more efficient use of Burnbot time. Wildland Cost/WUI = increase in wildland burn per unit of WUI damage prevented. Strategies that cluster near WUI may sacrifice wildland protection. Novel metric: no published work evaluates treatment efficiency under robotic operational constraints (travel time, slope feasibility).

Ignition points are stratified by NASA FIRMS hotspot density into quartiles (Q1=low-risk, Q4=high-risk). Q4 receives 40% of samples; Q1 receives 10% — testing both historical fire zones and unexpected ignition locations. A robust strategy performs well across all strata, not just where fires historically start. Stratified sampling follows standard Monte Carlo variance reduction techniques (Fishman 1996).

CSV includes strategy comparison, efficiency metrics, and per-run raw data. PDF includes summary tables and experiment parameters.