Southwest flash floods / Operational test

Where to look first.

As moisture from Hurricane Polo brought flood watches to four states, the Resilitix team ran EmergenCITY AI as an operational test and briefed a federal emergency-management team on where to look first, eight times in four days.

Operational test

Sep 28 – Oct 1, 2026

Flash floodingRiverine floodingFour states

Prepared by the Resilitix team.
Briefed to a federal emergency-management team.

01 / Documented work

~3.5 h

from the first question to a completed brief

Sep 28: first run 7:40 AM MDT; first brief completion record 11:15 AM MDT (3 h 35 min)

8 briefs

in four days, across four states

Sep 28 (two), Sep 29 (four), Sep 30, Oct 1

6 corridors

named in the first brief, from a statewide watch

Pecos, Galisteo Creek, Jemez, Chaco River, NM-537 headwaters, Magdalena drainages

Preparation times are based on recorded run and brief-completion timestamps. They are not measures of partner response time.

The operational challenge

A broad threat.
Limited time to assess it.

The model interprets. Analytical tools compute. The system shows its work. Human teams validate and decide.

The watch covers everything.

Flood watches spanned four states. No team can check every river, town and road.

The evidence is scattered.

Forecasts, flood outlooks, alerts, floodplains and population sit in separate systems.

The forecasts keep moving.

A manual assessment can be out of date before it is finished.

Priorities must hold up.

A ranking without visible evidence is hard to act on or to share.

02 / IN ACTION · SEP 28, THE FIRST BRIEF

One morning: from a statewide watch to six river corridors

01
Statewide flood watch
virtually all of New Mexico
02
27 counties
with forecast river flows at or above a 20-year return-period threshold; 39,560 residents in flagged cells
03
18 counties
where Google Flood Hub also showed a flood event; 2,885 residents in flagged cells
04
6 river corridors
named in the first brief as where to look first

Counts refer to screened cells and named areas in the first brief; they describe potential exposure.

The first brief was completed about three and a half hours after the first question. The team’s post-event review found reported closures or washouts along four of the six corridors over the next three days.

Retrospective review of one event. These figures describe reported impacts within areas flagged beforehand; they do not establish predictive accuracy or operational decision outcomes. Non-impacted flagged areas and false-alarm burden were not assessed. Method and impact list available on request.

Map from the Sep 28 brief: New Mexico hotspot cells along six numbered river corridors — Sep 28 brief: 1 Pecos · 2 Galisteo Creek · 3 Jemez · 4 Chaco River · 5 NM-537 headwaters · 6 Magdalena drainages. Potential exposure, not confirmed flooding.

Explore another view from the brief

Map from the Sep 29 evening brief: where Google Flood Hub flood events overlapped forecast river flows at or above a 20-year return-period threshold across Arizona, New Mexico and Colorado, with nine numbered places — Sep 29 brief: Flood Hub flood events overlapping forecast flows at or above a 20-year return-period threshold. Numbers mark places listed in the brief. Potential exposure, not confirmed flooding.

03 / Briefing cadence

Priorities that moved with the forecast

Each brief re-ran the screens. On Sep 29, Safford–Duncan was ranked top priority (P1) on an NWS warning, then eased to P2; new Flood Hub signals lifted Albuquerque, Santa Fe, the upper Pecos and Roswell to P1.

8 briefs, Sep 28 – Oct 1, 2026. B1: Sep 28, Morning; B2: Sep 28, Afternoon–evening; B3: Sep 29, Morning; B4: Sep 29, Morning; B5: Sep 29, Afternoon; B6: Sep 29, Afternoon–evening; B7: Sep 30, Morning; B8: Oct 1, Morning. — Brief-completion sequence; B8 is placed by its morning run window. Enlarge the timeline to inspect timestamps; the archive below gives the date and scope of each brief.

Browse all 8 brief covers Dates, editions and coverage

B1, Sep 28, Morning brief cover: New Mexico — B1 · Sep 28
Morning
New Mexico

B2, Sep 28, Afternoon–evening brief cover: New Mexico — B2 · Sep 28
Afternoon–evening
New Mexico

B3, Sep 29, Morning brief cover: New Mexico — B3 · Sep 29
Morning
New Mexico

B4, Sep 29, Morning brief cover: Arizona, Colorado, Texas — B4 · Sep 29
Morning
Arizona, Colorado, Texas

B5, Sep 29, Afternoon brief cover: Southwest, four states — B5 · Sep 29
Afternoon
Southwest, four states

B6, Sep 29, Afternoon–evening brief cover: Southwest, four states — B6 · Sep 29
Afternoon–evening
Southwest, four states

B7, Sep 30, Morning brief cover: Southwest, four states — B7 · Sep 30
Morning
Southwest, four states

B8, Oct 1, Morning brief cover: Texas — B8 · Oct 1
Morning
Texas

These are cover images from the briefing archive. Full briefs and supporting methods can be requested from the team.

04 / Operational value

What the work helped make visible.

The examples below describe the assessments and briefing outputs prepared during this test.

Prioritize limited attention

Know where to look first.

In this event: the first brief narrowed a statewide New Mexico watch to six river corridors.

→ Focus the next round of local checks.

Translate threats into consequences

Know who and what could be affected.

In this event: 2,885 residents and 1,644 residential structures in hotspot cells, listed by county and ZIP code.

→ Connect a hazard signal to potential exposure.

Iterate rapidly

Keep priorities current as signals change.

In this event: eight briefs in four days; places rose and fell in priority as new forecasts arrived.

→ Priorities that keep pace with the forecast.

Communicate and defend

Show the evidence behind each priority.

In this event: each figure traced to a dated run, with its question, map layer and value range.

→ Give reviewers a clear path back to the evidence.

05 / POST-EVENT REVIEW · IMPACTS REPORTED THROUGH OCT 3

What a post-event review found

After the event, the Resilitix team compared the seven Arizona, New Mexico and Colorado briefs (Sep 28–30) with impacts reported through Oct 3, compiled from news, NWS and NHC products, agency statements and public posts. Of the 43 impacts reported after each state's first brief, 40 were in areas the briefs had flagged beforehand, 31 of them in areas ranked priority; the median lead from an area's first flag to a reported impact there was about one day.

31 of 43

reported impacts in areas already ranked priority

reported after each state's first brief; 9 more were in areas marked monitor

12 of 13

major reported impacts in areas flagged beforehand

severity 4–5: dam and levee emergencies, impassable highways, a record river crest

~1 day

median lead from first flag to reported impact

23.7 h across the 40 impacts flagged beforehand; measured from the first relevant brief’s recorded completion time to the impact time or first report

Impacts reported after each state's first brief, by how the briefs had flagged the area beforehand.

Single-event retrospective finding. False-alarm burden and decision outcomes were not assessed.

Map of flood impacts reported in Arizona, New Mexico and Colorado, Sep 28 to Oct 3, colored by how the briefs had flagged each area beforehand — Impacts, by how the briefs had flagged each area: priority (amber), monitor (pale), flagged after (gray), before the first brief (outlined). Reported impacts, not screening output.

How to interpret the comparison

The map overlays reported impacts with the areas named in the earlier screening. Spatial overlap describes what was covered in this event; it does not establish predictive accuracy.

The recorded briefing outputs and the reported-impact list are different evidence sources.

Review method and definitions

Impact list: 67 records (61 from the post-event impact dashboard, 6 Colorado items from NWS Pueblo products and local news), re-scored on Oct 9; 13 non-impacts removed and 3 duplicates merged, leaving 51 impacts. Of these, 43 came after the first brief for their state. Priority: P1, P2 or a Sep 28 hotspot county; monitor: P3 or the watch list. Lead time uses the recorded file completion time of the first brief that flagged the area and the event time where reported, otherwise the first report. These timestamps do not establish partner receipt or action. Texas was included in operational briefing but not in this review.

The retrospective measures use different denominators and timing references across these events. They should not be combined into a single accuracy score.

Request the method and impact list ↗

Bring your operational question

Apply the workflow
to your next question.

Explore an operational test for your jurisdiction or network with the Resilitix team.

Discuss an operational test ↗

Opens your email app · info@resilitix.ai

NEXT CASE STUDY

Hurricane Isaias

From regional warnings to local priorities.

Read the case study →

EmergenCITY AI is designed for screening-grade resilience intelligence, planning, preparedness, monitoring, and decision support. It does not replace official forecasts, emergency command decisions, engineering studies, regulatory determinations, or local expert judgment.

The system runs the first pass. Your team makes the call. Resilience intelligence on demand.

emergencity.ai · info@resilitix.ai · EmergenCITY AI · Geospatial Resilience Intelligence

The operational test at a glance

A broad threat.Limited time to assess it.

The watch covers everything.

The evidence is scattered.

The forecasts keep moving.

Priorities must hold up.

One morning: from a statewide watch to six river corridors

Statewide flood watch

27 counties

18 counties

6 river corridors

Priorities that moved with the forecast

B1 · Sep 28

B2 · Sep 28

B3 · Sep 29

B4 · Sep 29

B5 · Sep 29

B6 · Sep 29

B7 · Sep 30

B8 · Oct 1