# Deep Dive #5 — Weather-Driven Evacuation & Impact Emulation: a decision theater for disasters *NSF NCAR / RAL · strategy working document · draft July 2026* **One line:** Couple RAL's physics-based hazard models (wildfire spread, hurricane wind/surge) to human-systems models (evacuation traffic, agent behavior) in an interactive, browser-native "decision theater" — so an emergency manager, a planner, or the public can *watch* what happens and *tune the what-ifs*: what if the winds pick up, what if one neighborhood leaves 30 minutes late, what if it rains. This document is built directly on your framing: an evacuation you can visualize and emulate, road congestion when one area is slow to get out, wildfire spread driven by *empirical* weather, and tunable scenarios (wind up/down, rain) — with the hurricane analog, and an honest note on tornadoes. --- ## Why this could be your project — and why it's exciting This is the most visual and most integrative idea in the portfolio, which makes it a strong fit for someone who does viz + software integration: - **Visualization is the product**, not a garnish. The whole value is making an unfolding disaster and a human response *legible in real time*. Your `wrf-viewer` / `webgl_viewer` work is the seed. - **It's systems integration** — stitching a fire model, a traffic model, weather forcing, and a UI into one coherent experience. That's engineering, not a single science domain. - **RAL already owns the hazard half.** You wouldn't be building fire physics from scratch — you'd be putting an interactive, human-coupled front end on models the lab is already a world leader in. It's also the highest-ambition idea (feasibility 3/5, ease 3/5). The way to make it real is to **start with visualization + scenario replay** (achievable now) and add live coupling in phases. --- ## The problem, precisely When a wildfire or hurricane threatens, the two questions that decide whether people live are *how will the hazard evolve* and *can everyone get out in time* — and today those are answered by **separate communities with separate tools that don't talk to each other**: - **The hazard side** has excellent physics. RAL's **WRF-Fire / CAWFE** couples the atmosphere to wildfire spread (firebrand/spotting added in v4.4, and it's the basis for NOAA's new **Community Fire Behavior Model**); **SWUIFT** extends it to the wildland-urban interface; hurricane wind and storm-surge models are mature. - **The human side** has its own tools — agent-based evacuation models and traffic simulators (MATSim, SUMO, and the transportation-planning literature on evacuation clearance times). - **They are almost never coupled**, and almost never *interactive*. Emergency managers get static maps and a clearance-time estimate, not a living model they can interrogate: "if this ridge community waits, does the one road out gridlock while the fire reaches it?" That coupling gap — physically-consistent hazard **×** human response **×** interactivity — is the opportunity, and it's squarely an "impact coupling" role where a trusted public institution beats both Big Tech (no domain physics) and startups (no trust for public-safety use). --- ## What exactly to build A **decision theater**: a browser-native environment that runs (or replays) a coupled hazard-and-response scenario and lets the user tune it live. **A. Weather-driven hazard core.** Wildfire spread forced by *empirical* weather — real observed/forecast winds, temperature, humidity, fuel moisture — via WRF-Fire/CAWFE (or a fast surrogate for interactivity). The key user-facing capability you described: **tunable meteorology** — nudge the wind up or down, add rain, shift direction, and watch the fire's behavior respond. For hurricanes, the analog is track/intensity and the resulting wind/surge footprint. **B. Human-response / evacuation model.** An agent-based / traffic layer over the real road network: population leaves on a departure-time distribution, routes to safety, and **roads congest**. Your specific scenario — *one area takes too long to get out and its single egress road jams while the hazard closes in* — is exactly the emergent behavior this surfaces. Outputs: clearance time, who is still exposed when the hazard arrives, bottleneck roads. **C. The what-if engine (the heart).** Interactive sliders and scenario presets: wind speed/direction, rain onset, ignition location, evacuation start time, staged vs. simultaneous evacuation, a blocked road, a contraflow lane. Each change re-runs (or re-replays) and updates the visualization. This is what turns a model into a **rehearsal tool** and a **public-persuasion tool** — "here's what happens if we wait." **D. The visualization.** A time-animated map: hazard footprint advancing, traffic flowing and jamming, exposure highlighted, danger colored intuitively. 2D first; a 3D/terrain view for fire-in-canyon or surge-over-topography as a stretch. This is the deliverable that makes people *take evacuation seriously* — your stated goal. ### Phased build (how to make an ambitious idea real) - **Phase 0 — replay + viz (achievable now):** take a *pre-computed* fire or surge scenario + a pre-computed evacuation run and build the interactive time-animated visualization and the what-if UI over cached scenario variants. This alone is a compelling demo and a fundable prototype — no live coupling required. - **Phase 1 — live tunable hazard:** wire a fast fire-spread surrogate so meteorology sliders re-run the hazard in near-real-time. - **Phase 2 — coupled human response:** integrate the traffic/ABM layer over a real road network for a pilot community; surface congestion and exposure. - **Phase 3 — operational / hurricane analog:** harden for a real emergency-management partner; add the hurricane track/surge/evacuation case. --- ## Hurricanes, and the honest word on tornadoes - **Hurricanes fit well.** Longer lead times (days), large spatial scale, and evacuation *is* the response — a strong second application once the fire case is proven. Track/intensity uncertainty maps naturally onto the what-if engine (evacuate for which scenarios?). - **Tornadoes are genuinely different, and you're right that they're harder.** Lead times are minutes, scale is sub-kilometer, and predictability is low — so evacuation-by-road isn't the response; *sheltering in place* is. A road-evacuation decision theater doesn't fit tornadoes. If tornadoes are ever in scope, the useful version is different: shelter-access and warning-response modeling, not clearance-time/traffic. Better to scope tornadoes out of v1 and say so plainly than to over-promise. --- ## Why a public institution wins here Public-safety decision tools must be **trusted, neutral, transparent, and physically defensible** — you cannot run a real evacuation off a startup's black box. NCAR/RAL brings the hazard physics (WRF-Fire/CFBM), the neutrality, and the verification culture; it's a natural, fundable public good with clear life-safety value. ## Technical risks & unknowns - **Interactivity vs. fidelity** is the core tension — full WRF-Fire is not real-time. Surrogates/emulators (ties to #4/#8) are the bridge; be explicit about where the tool is "indicative" vs. "validated." - **Human-behavior modeling is uncertain** — departure timing and compliance are hard; present ranges, not false precision. - **Data assembly** — road networks, population, fuels, real-time weather — is real integration work (your lane) and a partnership question. - **Misuse/over-trust risk** — a convincing visualization can be believed too much; the UI must communicate uncertainty and provenance. This is an ethical design requirement, not optional. ## What success looks like An emergency-management partner uses it to plan or train; a planning board watches the "if we wait" scenario and changes a decision; and — the goal you named — people take an evacuation order seriously because they've *seen* what delay does. --- # Positioning & funding — #5 ## The narrative "The forecast is not the point — the *decision* is. When a fire or hurricane threatens, communities need to see how the hazard and the human response unfold together, and to rehearse the what-ifs before the day. RAL has the world-class fire and storm physics; we can put an interactive, trustworthy decision theater on top of it that saves lives by making the consequences of delay impossible to ignore." ## Target sponsors & why each buys | Sponsor | The pitch to them | Vehicle | |---|---|---| | **DHS / FEMA** | Evacuation planning, training, and real-time decision support | DHS S&T, FEMA resilience programs | | **DOT / state DOTs** | Evacuation traffic, contraflow, clearance-time planning | DOT / FHWA; RAL's road-weather (MDSS) relationships | | **State emergency mgmt / Cal Fire** | Fire evacuation rehearsal and public communication | State contracts; wildfire-state partnerships | | **NOAA** | Impact-based decision support services (IDSS) for hurricanes/fire weather | NOAA weather program / EPIC | | **NSF** | Smart & connected communities; disaster-resilience research | NSF (S&CC, CIVIC, hazards) | | **Insurance / reinsurance** | Evacuation & exposure modeling for catastrophe risk | Commercial | | **Philanthropy** | Climate-adaptation & disaster-resilience public tools | Climate funders | This idea has the **broadest, most mission-driven funder set** of any in the portfolio — which is exactly why it scored high on fundability (5/5) despite the difficulty. ## The differentiated value prop - **To emergency managers:** a rehearsal and real-time tool that couples the hazard and the roads — which they currently do in their heads. - **To the public / officials:** the persuasion artifact — *seeing* gridlock meet fire changes behavior in a way a map never will. - **To funders:** life-safety impact from software, built on physics only NCAR has. ## The ask (illustrative) Start **internal + one agency pilot**: ~2 FTE for a Phase-0 replay-and-viz prototype on one real historical fire (e.g., a well-documented WUI evacuation), then pursue DHS/DOT/state funding for the coupled, live version. The Phase-0 prototype is the fundraising instrument. ## Risks to the funding case - "Too ambitious" → the phased plan (replay first) is the answer; show a working Phase-0 before promising live coupling. - Liability/trust concerns for real evacuations → position v1 as *planning & training & communication*, not an operational dispatch system, and build uncertainty communication in from day one. - Overlap with existing evacuation-modeling groups → partner with them; RAL's unique contribution is the *weather-driven hazard coupling* and the verification/trust layer, not reinventing traffic simulation.