He Named My Wildfire Prediction Model After Himself — Then the Fatality Inquiry Required the 6-Year Burn Probability Dataset

The fire shelter was clipped to Dr. Elena Reyes’s belt — a compact, reflective emergency shelter pack, 25cm × 12cm, the silver surface worn to a matte finish from years of contact with her field gear. The pull tab was slightly worn from being tested monthly, as required by NWCG protocol. She had carried it through 23 fire assignments across four states and had never deployed it.

The FlamMap display at the Arroyo Seco Fire briefing tent showed the fire spread model she had run the previous afternoon — the fuel moisture inputs, the slope data, the wind pattern from the previous 48-hour observations, and the rate-of-spread projections for the current burn conditions. The spread model had produced a timeline: a 4-hour window from the current fire position to the drainage channel, with the drainage flagged as a high-probability rate-of-spread corridor. She had colored it red in the visualization.

Anthony Redhawk, the crew boss for Sector 7, was studying the display. He was responsible for 22 firefighters working the east flank.

She pointed at the red corridor. “The drainage channel here. The slope-to-wind alignment creates a natural draft that will accelerate the fire’s rate of spread when it enters the channel. The fuel loading in the drainage is heavy — shrub and chapparal at 8 to 12 feet, cured, with the fine fuel moisture below 6%. Once the fire enters the drainage, the rate of spread will increase. The model shows the fire exiting the drainage northeast in approximately 35 minutes at current conditions.”

He said: “Four hours from now.”

“Four hours from the current fire position to the drainage head,” she said. “Less if the wind increases, which the 24-hour forecast shows as a moderate possibility between 1600 and 1800 hours.”

He looked at the corridor. He said: “We’re pulling back from the drainage at 1400.”

“That’s what the model recommends,” she said. She pointed at the briefing printout — the same map, printed, with the timeline written in the margin: 1400 — corridor clearance, 1600 — fire entry drainage head. She had written it in grease pencil on the laminate cover sheet. He took the printout.

She touched the pull tab on her fire shelter without thinking. She had done it thousands of times. The worn tab was always there.

—

The burnover had happened at 1730 hours.

Three weeks earlier, when the cluster of 14 firefighters had been reported as trapped in the drainage, Donovan had called her on the radio from the incident command center. He had said: “Your model was right. The drainage is the problem.” She had said: “I flagged the corridor in the spread model at 1400 hours. The spotfire at the drainage head at 1600 hours was within the predicted envelope.” He had said: “The crew is trapped in the drainage. We’re pulling everyone back.” She had said: “The model shows the fire exits the drainage northeast in approximately 35 minutes at current rate of spread.” He had said: “Okay. Okay. Good work, Elena.” She had gone back to the FlamMap output and updated the real-time spread path with the 1730-hour position data. She had noted the time. The shelter was on her belt.

That call had been at 1803 hours. He had gone to direct the helicopter rescue. She had continued updating the model.

The after-action report had been published four weeks after the fire was contained.

She read it on her phone during the drive to a new assignment — a grass fire in a different county, an initial attack where she would be the fire behavior analyst on the incident management team. She had read: “Donovan Fire Behavior Assessment — incident fire behavior analysis conducted under Operations Chief M. Donovan’s direction.” She read “fire behavior support: Dr. Elena Reyes, NWCG-FBAN-ER-7741.”

She was reading on her phone at a rest stop. She was not driving. She put the phone in her jacket pocket. She touched the pull tab on the fire shelter. She had not thought about whether she had touched it because she had read the report, or because she was at a rest stop between assignments, or because she touched it without reason at the start of every field deployment. She got back in her vehicle. She drove to the new assignment briefing.

—

She was in her vehicle with her field laptop when she checked the FlamMap project file for the Arroyo Seco analysis. NWCG-FBAN-ER-7741 in the model documentation header — her NWCG fire behavior analyst certification number, embedded in the project documentation at the time the model was saved. The model had been saved 47 times over the course of the fire. Every saved version had the certification number in the header.

She closed the laptop. She touched the fire shelter pull tab. She drove to the new assignment briefing tent.

The NWCG fire behavior analyst certification — NWCG-FBAN — was one of the more demanding technical certifications in the wildfire management system. It required successful completion of several prerequisite NWCG courses, documented experience on fire assignments at progressively complex incident management levels, and a certification exam administered through the NWCG qualification system. The certification was renewed every five years through continuing education and documented fire assignment hours.

She had been NWCG-FBAN certified for 11 years. She had completed the certification the second year of her research career, when she had been running fire behavior modeling studies for a federal research lab, and had maintained it through every subsequent assignment. NWCG-FBAN-ER-7741 had been her certification number from the first renewal cycle.

The FlamMap software was a federal interagency tool — developed by the Rocky Mountain Research Station and maintained by the USDA Forest Service — that ran fire spread simulations based on surface fuel data, terrain data, canopy cover data, and atmospheric inputs. The spread model used the Rothermel fire spread equations, the primary physical models for fire behavior prediction in the wildfire management community. She had run FlamMap analyses on 23 fire assignments. Each model run produced a rate-of-spread map, a flame length map, and a spread timeline for the current fuel and weather conditions.

The Arroyo Seco model had been one of the more complex she had produced — the terrain had two distinct drainage systems, the fuel loading was highly variable between the canyon bottom and the ridge crests, and the wind data had shown significant topographic channeling effects that required manual adjustment of the wind inputs at each terrain segment. She had spent eight hours building the model before the briefing.

The wildfire management conference was in Sacramento — two days, 300 attendees, a plenary session on lessons from the previous year’s fire season. Donovan presented in the second plenary: “Drainage Corridor Risk: A Case Study in Pre-Fire Behavior Prediction.”

His presentation had 11 slides. Slide 4 was Elena’s FlamMap spread map — the Arroyo Seco model output with the drainage corridor in red, the timeline annotations she had added in the briefing tent the morning before the burnover. He had the original file. He had displayed it.

He said: “Our fire behavior predictions identified the drainage corridor as a high-probability rate-of-spread zone four hours before the burnover event. The model output guided the decision to begin crew withdrawal from the corridor.”

He said “our fire behavior predictions.”

He said “the model output” without naming the model, the analyst, or the NWCG certification.

He did not show the 4-hour timeline warning she had written in grease pencil. He did not show NWCG-FBAN-ER-7741. He did not say: the spread model was produced by Dr. Elena Reyes, NWCG fire behavior analyst, whose certification number is in the model header.

Elena was not at the conference. She was at her desk doing a fuel moisture analysis for a different district.

—

The subpoena arrived by email from Jocelyne Vásquez, plaintiffs’ attorney.

“Dr. Reyes — I represent three families in a federal wrongful death action arising from the Arroyo Seco Fire burnover on [date]. In connection with our discovery, we are seeking the following: (1) the original FlamMap fire spread model files for the Arroyo Seco Fire, (2) documentation of your NWCG fire behavior analyst certification, and (3) your deposition on the fire spread prediction and the timeline warning issued prior to the burnover. The FlamMap project documentation identifies NWCG-FBAN-ER-7741 as the certifying analyst. Please confirm your availability.”

She read “NWCG fire behavior analyst certification.”

She read “the certifying analyst.”

She opened the FlamMap project file. NWCG-FBAN-ER-7741. 47 saved versions. The 1400-hour model run — the version that showed the drainage corridor in red — was version 31. She had saved it at 1342 hours on the day of the burnover.

She touched the pull tab on the fire shelter.

She did not call Donovan.

She opened a reply to Jocelyne Vásquez. She confirmed her availability and attached: the FlamMap project file (all 47 versions), the model documentation with the certification header, the NWCG-FBAN-ER-7741 certification certificate and scope, and a two-page timeline summary documenting the 1400-hour corridor warning, the 1600-hour spotfire confirmation, and the 1730-hour burnover event.

She sent it.

She went to the current district’s fuel moisture analysis. It was incomplete. She finished entering the data.

—

The state agency’s legal team had told Donovan the same week: the plaintiff’s attorney had subpoenaed the FlamMap files. The files identified the NWCG-certified fire behavior analyst. NWCG-FBAN-ER-7741 was Elena’s certification. They would need her to be deposed.

He had said: “What about the after-action report? It says ‘Donovan Fire Behavior Assessment.'”

The team lead had said: “The after-action report is how the incident commander’s office described the analysis. The FlamMap files have the certifying analyst’s NWCG number in every header. NWCG-FBAN-ER-7741 is Elena Reyes. She is the fire behavior analyst of record. You cannot be deposed on a FlamMap spread model methodology.”

He had looked at the slide he had presented in Sacramento: “our fire behavior predictions.”

He had said nothing for a long time.

—

She had not called Donovan before preparing the discovery documentation. The subpoena had been addressed to the NWCG-certified fire behavior analyst identified in the FlamMap documentation. That was her. She had responded.

The timeline documentation had taken her three hours to prepare — she had gone through the 47 model versions, identified the key decision points (version 31 at 1342 hours, the drainage corridor designation; version 38 at 1618 hours, the first real-time spread update after the spotfire at the drainage head; version 43 at 1803 hours, the update she had run after Donovan’s call), and had written a two-page timeline narrative that tracked the model’s predictions against the actual fire behavior at each recorded observation point.

The model had been accurate at every observation point.

She had sent the documentation and gone back to the fuel moisture analysis for the new district. The fuel moisture readings were coming in from the field technicians — 47 data points across three fuel types. She entered them into the spreadsheet. The entry took 90 minutes. She finished it and moved to the next task.

He had been the operations chief for eleven fire seasons. He had managed 47 incidents, submitted 19 after-action reports, and testified in two previous proceedings — both involving equipment failures, both on operational decisions, both areas where his authority and experience were directly relevant to the questions being asked.

The depositions in this case would be different. The plaintiffs’ expert would want to know about the 1400-hour drainage corridor warning — specifically, whether the warning was in the model before the burnover, whether the warning identified the specific drainage where the crew was trapped, and whether the warning had been communicated to the crew. The answers to those questions were in the FlamMap model, in the briefing printout, and in the radio communications record. Elena had the model. Anthony Redhawk had received the briefing printout. The radio communications were in the incident log.

He could not be deposed on the modeling methodology because he had not conducted the modeling. He could not explain why the drainage corridor was flagged in the model because he had not built the model. He knew the conclusion — the corridor was flagged, the warning was there — because he had seen the map. He had received the briefing. He had acted on it, ordering the crew withdrawal from the corridor that morning.

But the act of ordering the withdrawal was different from the act of identifying the risk. He had done the first. She had done the second.

He had described both as “our fire behavior predictions” because from his incident commander’s perspective, the analysis was in service of his operational decision. The model had informed the decision. He had made the decision. He had understood the model as a tool his team had deployed — a tool that had told him what his decision needed to know.

But a tool did not have a NWCG certification. The analyst had the certification. The analyst was the professional of record. The analyst was the person the court would need to question.

He had said “good work, Elena” at 1803 hours while she was updating the real-time spread path. The call had been thirty-three seconds. He had gone to direct the helicopter rescue. She had noted the time and continued entering the position data.

He opened the after-action report amendment form.

—

She was at her desk, the fire shelter on her belt — she was still in her field kit because she had come directly from a morning briefing on the new district assignment. She was preparing the discovery documentation: the FlamMap files, the timeline summary, the certification packet. She was working through the items methodically, checking each against the subpoena’s discovery list.

The fire shelter was on her belt. She was sitting at her desk. She did not take it off to sit at her desk, the way she had not taken it off during the 23 previous fire assignments when she had moved between the briefing tent and a vehicle or a command post. She had been wearing field gear all morning. The shelter was on her belt the way it was always on her belt when she was in field gear.

She finished the discovery package. She sent it to Jocelyne Vásquez. She went to get a coffee. When she came back, she sat down and touched the pull tab once before she opened the next task in her queue.

—

She was clear about what the discovery documentation showed. The 1400-hour corridor clearance warning was in the model at 1342 hours. The crew boss had received the printout with the annotation at approximately 0820 hours during the briefing. The pullback order had been issued at 1412. The corridor had been cleared by 1437. The burnover had occurred at 1730, in the drainage, with 14 firefighters who had re-entered the corridor after the initial pullback for equipment retrieval.

She had not known about the re-entry. She had seen the incident log entry after the fire was contained: three crew members had re-entered the drainage at 1702 to retrieve equipment, and the burnover had caught them and 11 others at varying positions in the corridor. The pullback had been executed correctly. The re-entry had not been.

She had no opinion on the legal question of liability. That was for the court. She had documentation of the fire spread prediction and the communication of the timeline warning. That was what the subpoena had asked for.

She had prepared it accurately. She had sent it. She had gone back to work.

The federal deposition was held in a federal building in the district where the wrongful death case had been filed. Elena was in a conference room with the plaintiffs’ attorney, the defense attorney, the court reporter, and Donovan on the defense side. Donovan’s introduction had been brief: “Dr. Reyes is the NWCG fire behavior analyst who produced the fire spread model. The modeling methodology questions are for her.” He had not spoken on modeling topics for the remainder of the deposition.

She was deposed for four hours.

Jocelyne Vásquez led the examination. The questions built from the foundation to the conclusion. First: the FlamMap software — what it was, what inputs it used, what outputs it produced, what its validation record was. She answered all of these. Then: the Arroyo Seco model specifically — when it had been built, what inputs had been used, how many times it had been saved, when the version that showed the drainage corridor red had been produced. Version 31. 1342 hours. 4 hours and 48 minutes before the burnover at 1730.

Then: the drainage corridor designation. What made the drainage a high-probability rate-of-spread corridor. The slope. The fuel loading. The wind alignment. She answered each question.

Then: the communication. The briefing tent on the morning of the burnover. The printout she had given Anthony Redhawk with the 1400-hour pullback annotation in grease pencil.

Jocelyne Vásquez said: “The annotation says 1400 — corridor clearance. You wrote that at approximately 0800.”

She said: “At the briefing, yes.”

Vásquez said: “The burnover was at 1730. The crew pullback from the corridor had been initiated at what time?”

She said: “According to the incident log, the pullback was ordered at 1412. The crew began withdrawal at 1415. The first crew member cleared the corridor entry at 1437.”

Vásquez said: “So the warning was in the model at 1342. The pullback was ordered at 1412. The corridor was cleared by 1437. And the burnover occurred at 1730.”

She said: “Yes. That’s the sequence.”

The defense attorney cross-examined. He asked about the confidence level of the rate-of-spread projections, the variability in the fuel moisture inputs, the potential for the model to have underestimated the time to burnover. She answered each challenge from the model documentation — the input validation steps, the confidence intervals for each variable, the cross-validation against the fire’s actual rate of spread in the first two hours. The model had been accurate. The drainage corridor prediction had been accurate. The timeline had been accurate.

The deposition record stated: “Fire Behavior Analyst: Dr. Elena Reyes, NWCG-FBAN-ER-7741. 1400-hour drainage corridor warning documented in model version 31, timestamp 1342 hours.”

—

She walked from the vehicle to the briefing tent with the fire shelter on her belt — the compact silver pack, 25cm × 12cm, the pull tab slightly worn from monthly testing over 11 years. She had never deployed it. She had carried it through 23 fire assignments, four states, and one federal deposition, where she had noticed it on her belt when she sat down and had not removed it. The FlamMap project was in the federal court discovery archive — “Fire Behavior Analyst: Dr. Elena Reyes, NWCG-FBAN-ER-7741, drainage corridor predicted 1400 hours” — alongside the amended after-action report and the NWCG attribution protocol. Anthony Redhawk was at the briefing tent entrance. He nodded when she arrived. She had the current fire’s FlamMap output on her tablet. She checked the drainage corridors on this terrain. She touched the pull tab once, the way she always did at the start of a new briefing. She went to the briefing tent.

—

Jocelyne Vásquez contacted her after the deposition was complete.

“Dr. Reyes — your FlamMap timeline is the clearest documented fire behavior prediction I have reviewed in a wrongful death case. The 1400-hour drainage warning is in your model. The record is clear.”

Anthony Redhawk had heard about the deposition through the fire management community.

He said: “You’re the analyst in the record.”

She said: “Yes.”

He said: “You warned them at 1400 hours.”

She said: “The model did.” She touched the pull tab.

Donovan called the evening after the deposition. He said: “Good outcome. Your model held up.” She said: “The model was accurate.” He said: “Yes. I’ve amended the after-action report and I’m updating incident reporting protocol — NWCG analysts will be named going forward.” She said: “Yes.” He said: “Good work, Elena.” She said: “Thank you.” She looked at the fire shelter on her belt. She was in the federal building. She was still wearing it.

The amended after-action report arrived by email — “Fire behavior analysis by Dr. Elena Reyes, NWCG Fire Behavior Analyst, NWCG-FBAN-ER-7741. FlamMap spread model, Arroyo Seco Fire.” She read it. She filed it in the Arroyo Seco project folder alongside the original after-action report with “fire behavior support.” She had not annotated either of them. Both were in the folder.

The NWCG incident reporting protocol arrived the same day from Donovan’s agency — a policy document requiring that certified fire behavior analysts be credited by name and NWCG certification in all after-action reports and incident documentation. She read it. She filed it.

The distribution confirmation showed 47 state and federal agencies had received the original after-action report. She had the confirmation record in the project folder. She did not know which of the 47 had updated their files. She did not contact them. The amendment had been issued to the same distribution list. The individual agency file management was their responsibility.

—

The new fire assignment was a grass fire in a different district — faster-moving, lower fuel loading, different terrain than the Arroyo Seco canyon environment. The NWCG attribution protocol meant that her name and certification would appear in the incident documentation from the first model run.

She had run the initial FlamMap model the previous afternoon, before the briefing. The model showed two candidate spread corridors — a northeast drainage and a ridge crest — with the northeast drainage rated higher probability based on the fuel loading and wind alignment. She had annotated the map with the spread timeline for each corridor. NWCG-FBAN-ER-7741 in the model header.

Anthony Redhawk was not assigned to this fire. She did not know the crew boss. She had briefed the incoming crew boss at the tent entry, handed him the annotated map, and watched him study the northeast corridor. He had asked one question: “This drainage — if the wind picks up after 1600, does the timeline change?”

She had said: “If the wind increases to the 1800–2100 forecast values, the timeline shortens by approximately 45 minutes. I’d want clearance from the corridor by 1315 instead of 1400.”

He had said: “We’ll clear by 1315.”

She had nodded. She had touched the pull tab. She had gone back to the FlamMap console to run the sensitivity analysis on the 1800-hour wind scenario.

She was at the console now — the fire shelter on her belt, the FlamMap display open, the sensitivity run in progress. The northeast drainage’s rate-of-spread at the higher wind values was tracking at 1.4 times the base case. She noted the multiplier. She updated the timeline annotation on the map.

She touched the pull tab once, the way she always did at the start of a new briefing. She went to the briefing tent.

The sensitivity run completed. The northeast drainage at the 1800-hour wind values would see a rate-of-spread increase of 1.4 times the base case — a 45-minute compression of the spread timeline. She had updated the map annotation: 1315 corridor clearance instead of 1400. She printed the updated map and took it to the crew boss.

He was at the tent entrance. He looked at the updated annotation. He said: “1315. I’ve already moved it.”

She said: “Good.” She gave him the printout.

She went back to the console. The model was in steady state — no significant input changes since the morning update. She checked the wind observation data from the two remote automated weather stations on the ridge. The 0900 readings were consistent with the forecast: wind speed holding at the base case values. The 1800-hour increase was still a forecast event. She noted the readings. She set a reminder to check the 1200-hour readings when they came in.

The fire shelter was on her belt. She had worn it for eleven years, through every type of fire assignment — initial attack, large incident, prescribed burn. She had never been in a position where she had needed to pull it. She checked the tab at the start of every assignment the way she checked the FlamMap inputs: verification protocol, before she began.

She touched the pull tab once. She went to the briefing tent.