From the Utility Management edition of October 31, 2026
Houston Public Works builds a living stormwater master plan, testing an AI surrogate model that cuts model runs from six hours to three seconds
City of Houston, Houston Public Works (reported by Autodesk) · Municipal stormwater and flood resilience · Houston, Texas, United States
- Publisher
- Autodesk One Water Blog
- Original publication
- Autodesk One Water Blog, 23 Oct 2026
- Source retrieved
- 2026-11-01
What happened
Houston Public Works chose InfoWorks ICM after a documented evaluation of 50 criteria and 20 providers. It built an $11 million citywide Stormwater Master Plan with an interactive public flood-risk map and a planned three-year update cycle. A managing engineer reports that the AI surrogate model returns results in three seconds versus six hours for a full hydraulic run. Autodesk's footnote says surrogate modeling is AI functionality still under development and being tested by customers such as Houston. A separate WEFTEC 2026 session abstract (29 Sep) co-authored by the same Houston engineer describes a wastewater data lake and a hydraulically informed machine-learning model ranking dry-weather sanitary sewer overflow risk to move cleaning from calendar to risk-based schedules, without reporting outcome figures.
Why it matters
One of the few current U.S. municipal stormwater AI examples: a large city using an AI emulator to make citywide flood modeling fast enough for frequent updates and operational use. Stormwater was a recorded gap in both earlier editions.
Evidence and measured results
Vendor-published customer story with direct quotes from the Director of Public Works, City Engineer and a Managing Engineer. The six-hours-to-three-seconds figure is a practitioner statement with no stated accuracy comparison against the full model. The feature is pre-release. The WEFTEC abstract is a conference summary without results.
Limitations and uncertainty
Vendor marketing with no independent check; surrogate accuracy versus full model is not reported; the feature is under development. Published 23 Oct, before the 2026-10-29 edition, and not archived; included now because U.S. stormwater AI evidence was missing from both prior editions and this is the most recent U.S. municipal example found.
Put this evidence to work
Lighthouse Advisory interpretation, grounded in this source. Enriched 2026-10-31; this does not change the original publication date. Each role follows the same checklist so takeaways compare across sources.
Sales
Role takeaway
- Customer problem
- Citywide hydraulic flood models take hours per run, limiting scenario testing and frequent updates.
- Stakeholders
- Director of public works
- City engineer
- Stormwater modeling manager
- Emergency management coordinator
- GIS manager
- Discovery questions
- Which decisions wait on hydraulic model runs today?
- How often is the stormwater model updated and recalibrated?
- How would you validate a faster approximate model?
- Value hypothesis
- An AI surrogate could make flood scenarios fast enough for frequent planning and storm-day use, if validated against the full model.
- Potential engagement
- A validation study comparing surrogate and full-model results for one watershed across design and historical storms.
- Unsupported claims
- Surrogate accuracy equal to the full model
- Production readiness
- Reduced flood damage
- Cost savings
Pre-sales engineering
Role takeaway
- Fit
- Fits cities with a calibrated, citywide 1D/2D hydraulic model and capacity to generate training runs.
- Architecture and integration
- Surrogate trained on full-model runs, results feeding asset management and public flood-risk maps.
- Prerequisites
- Calibrated citywide hydraulic model
- Library of full-model runs for training
- Asset and GIS integration
- Constraints
- Feature is pre-release
- Accuracy outside training range unknown
- Retraining needed after network changes
- Security
- Treat detailed flood and asset models as sensitive infrastructure data and control external sharing.
- Proof of value
- Compare surrogate and full-model depth and extent for at least five storms
- Test an out-of-range storm
- Measure run time and retraining effort
Delivery
Role takeaway
- Work
- Model maintenance, surrogate retraining each update cycle, validation reporting and public map uncertainty notes.
- Dependencies
- Modeling staff
- Vendor software roadmap
- GIS team
- Emergency management
- Ownership
- Stormwater modeling manager under the city engineer.
- Skills and adoption
- Modelers need skills in emulator validation; operations staff need guidance on when surrogate output is decision-grade.
- Governance checkpoints
- Validation sign-off before operational use
- Retraining after each base-model update
- Review of public map communications
- Acceptance criteria
- Proposed: surrogate depth error within an agreed tolerance for validated storms
- Proposed: retraining completed within each three-year update
- Proposed: uncertainty statement published with public results
- Risks
- Overreliance on fast output during storms
- Vendor feature changes
- Overstated precision in public maps
Implementation considerations
Lighthouse Advisory interpretation across the operating dimensions a public-sector buyer must settle before this evidence becomes a design. Each note answers the question under its heading for this specific source.
Architecture and integration
What must connect, and where does the AI sit in the workflow?
A surrogate model trained on many full hydraulic runs can stand in for the solver for rapid scenarios, but must be retrained when the network model changes. Houston feeds model results into an asset management platform and a public map.
Governance
Who approves, reviews and stays accountable for outcomes?
A documented 50-criterion software evaluation and a three-year update cycle provide a defensible procurement and maintenance record. Validating surrogate results against the full model should be part of model governance.
Security and privacy
What data, permissions and controls need testing?
Limited direct security content. Houston's City Engineer has elsewhere stressed keeping utility data in house; detailed flood and asset models are sensitive infrastructure information.
Accessibility and workforce
Who is affected, and what skills or accommodations follow?
Public flood-risk map is meant to let every resident see their own risk; modeling staff need new validation skills.
Procurement
What should contracts, pricing and exit terms secure?
Houston's documented 50-criterion, 20-provider evaluation is a model for defensible software selection; pre-release AI features should carry contract terms on validation and support.
Operating model
Which teams own the service once it runs?
Treats the master plan as a living model with a three-year update cycle rather than a one-time report.
Publication history
- 2026-10-31Utility Management · Issue 036 resources
Stable resource ID: houston-stormwater-master-plan-ai-surrogate-model-2026