TRB Annual Meeting Poster

Evaluating the Impact of Flood Model Selection on Roadway Vulnerability Assessments: An Example from Massachusetts

Courtney Marie King, M.S. · Rahul Soma, M.S. · Maximillian Petersen · Farshid Vahedifard, Ph.D. · Shan Jiang, Ph.D.

Transportation Research Board (TRB) · Tufts University

Transportation agencies rely on data-driven analyses to allocate funding for resiliency projects equitably, yet Massachusetts still depends primarily on FEMA flood maps while newer high-resolution models like the First Street Flood Model (FS-FM) remain underutilized. This study evaluates FS-FM's applicability for roadway vulnerability assessments by comparing it with FEMA maps and demonstrating how higher-resolution data improves risk precision and decision-making. Findings show that model choice significantly influences investment prioritization, underscoring the need for transparency and publicly accessible high-resolution flood data to support equitable and effective resiliency planning.

01  /  Introduction

The Problem with One-Size-Fits-All Flood Maps

As climate change intensifies flood risk across the United States, transportation agencies face increasing pressure to allocate limited infrastructure funding wisely. The data they use to make those decisions—flood hazard maps—can dramatically shape which communities and roadways are prioritized for investment.

Massachusetts continues to rely primarily on FEMA's National Flood Hazard Layer (NFHL), a watershed-scale model that, while widely adopted, was not designed for fine-grained infrastructure analysis. Newer models like the First Street Flood Model (FS-FM) offer parcel-level resolution and broader statewide coverage—but remain underutilized in practice.

This study asks a fundamental question: Does it matter which flood model you use? By comparing FEMA and FS-FM vulnerability rankings for every MassDOT roadway segment in the state, we show that the answer is a clear yes—with major implications for equity and resource allocation.

1-mi
Segment granularity
0–5
Standardized risk scale
41%
Of segments FEMA rates high-risk that FS-FM assigns no risk
2
Models compared statewide

02  /  Data & Methods

Two Flood Models, One Road Network

🗺️

FEMA National Flood Hazard Layer

The federal standard for flood zone mapping. Based on watershed-scale hydrological modeling, FEMA's polygons define broad flood zones (A, AE, VE, X, D) tied to 1-in-100-year return periods. Its coarse resolution can overestimate flood extents across entire neighborhoods rather than identifying specific at-risk segments.

📡

First Street Flood Model (FS-FM 3.0)

A high-resolution, parcel-level flood model developed by First Street Foundation. FS-FM provides expected inundation depths on a continuous scale rather than binary zones, and offers broader geographic coverage including areas FEMA has not yet mapped. Tiles were extracted and merged from Zillow's climate-risk mapping interface.

Processing Pipeline

MassDOT roadway centerlines were divided into one-mile segments, each buffered by lane width. These segments were then intersected with both flood models to calculate average inundation depth. Both datasets were standardized to a 0–5 flood-risk scale using the table below, enabling direct comparison.

FEMA Zone FS-FM Depth Rank
A, AH, AO, VE, AE
High Flood Hazard (>1 ft)		 ≥ 3 ft 5
— 2 ft ≤ x < 3 ft 4
D
Moderate Flood Hazard (~1 ft)		 1 ft ≤ x < 2 ft 3
— 0.5 ft ≤ x < 1 ft 2
X
Low Flood Hazard (<1 ft)		 x < 0.5 ft 1
Not within flood zone Outside FS-FM 0

Flood risk ranking table used to standardize FEMA and FS-FM data to a common 0–5 scale.

Flood risk data comparison visualization
Figure 1. Flood risk data layers intersected with MassDOT roadway segments for vulnerability analysis.

03  /  Calculating Vulnerability

A Multi-Factor Vulnerability Score

Flood depth alone does not capture full infrastructure risk. The vulnerability ranking integrates three factors—flood risk, culvert condition, and bridge presence—each weighted by a sensitivity analysis.

Vulnerability Rank Formula
Vulnerability Rank  =  2(F)  +  C  −  B
F  (×2)
Flood Rank
FS-FM or FEMA score (0–5)
doubled by sensitivity analysis
C
Culvert Rank
Based on bankfull width ÷ culvert width
assigned via 0.5-mi buffer intersect
−B
Bridge Presence
Subtracted as surrogate for
an elevated, less-vulnerable roadway

The flood risk variable received a coefficient of 2, reflecting its outsized influence on overall vulnerability as determined through sensitivity analysis. Culvert vulnerability was assessed using the ratio of bankfull width to culvert width, with each culvert assigned a 0.5-mile buffer intersected against roadway segments. A value of 1 was subtracted when a bridge was present, serving as a surrogate for an elevated roadway.

Vulnerability scores from FS-FM data
Figure 2. Vulnerability scores derived from the First Street Flood Model (FS-FM). Scores range from 1 (least vulnerable) to 13 (most vulnerable).
Vulnerability scores from FEMA data
Figure 3. Vulnerability scores derived from FEMA flood zone data. Note the concentration of high scores in eastern Massachusetts.

04  /  Results

Major Statewide Discrepancies

After mapping flood model outputs onto the full MassDOT road network, the statewide results reveal striking differences between the two approaches. FEMA identifies roadways in eastern and southern Massachusetts as most flood-prone, with minimal risk assigned to the northwest—likely reflecting its limited coverage in that region and its watershed-based methodology.

In contrast, FS-FM produces more geographically varied results statewide, reflecting its broader parcel-level modeling and wider coverage area.

Statewide roadway flood risks - FEMA (left) and FS-FM (right)
Figure 4. Statewide results of roadway flood risks. (Left) FEMA model. (Right) First Street Flood Model. Note the differing geographic distribution of high-risk segments.

Key finding: FEMA labels 41% of roadway segments as high-risk that FS-FM assigns no risk. This level of disagreement means that investment decisions made under one model could be systematically different—and potentially inequitable—compared to those made under the other.

Map of flood risk differences by roadway segment
Figure 5. Map of flood risk differences by roadway segment, comparing FEMA and FS-FM models. Segments where the two models agree most strongly are the best candidates for confident near-term prioritization.
Detailed results maps
Figure 6. Regional detail of roadway vulnerability assessment results, showing segment-level variation between the two flood models.

05  /  Ground Truth Example

Morrissey Boulevard, Boston

To ground these statewide findings in real-world conditions, we examined Morrissey Boulevard in Boston—a well-documented flood-prone corridor where real events allow us to test model accuracy. During king tide and storm events, specific low-lying roadway segments along Morrissey repeatedly flood, even while the adjacent bridge structure remains largely unaffected.

King Tide flooding on Morrissey Boulevard, Boston, 2023
Figure 7. King Tide flooding event on Morrissey Boulevard, Boston (Eagan, 2023). Low-lying roadway segments flood while the bridge deck remains passable.

This example illustrates a key limitation of FEMA's coarser polygon approach: it tends to apply flood risk to entire zones, which can include the bridge structure in the same high-risk category as the vulnerable roadway segments beside it.

FS-FM's higher resolution allows it to pinpoint the vulnerable low-lying segments more precisely, separating them from elevated structures that are not meaningfully at risk during typical flood events.

At the same time, FS-FM can underpredict flood extent in certain areas—meaning neither model alone is sufficient. The Morrissey case illustrates why combining both datasets yields the most accurate picture of true vulnerability.

MassDOT's ongoing Morrissey Boulevard redesign—released for public comment in March 2025—underscores the practical stakes of this analysis for real infrastructure investment decisions.

06  /  Conclusion

Model Choice Matters for Equity and Resilience

Finding 01

Significant Model Disagreement

FEMA's coarse polygon model often overpredicts flood risk at the segment level. FS-FM can underpredict in certain areas. Neither model alone is sufficient for equitable infrastructure prioritization.

Finding 02

Areas of Confident Agreement

Where both models agree on high-risk segments, agencies can prioritize interventions with greater confidence. These areas of convergence provide actionable starting points for investment.

Finding 03

Transparency Is Essential

The 41% discrepancy in high-risk classifications underscores the urgent need for publicly accessible, high-resolution flood data and clear documentation of which models inform funding decisions.

These findings underscore the need for publicly accessible, high-resolution flood data and improved modeling approaches that integrate both datasets to strengthen accuracy, equity, and long-term climate resilience planning. Transparency about model choice is not a technical footnote—it is a matter of environmental justice.

References & Acknowledgements

Sources

  1. Eagan, J. (2023, September 15). 'Conditions Similar to Nor'Easter': Boston Prepares for Lee. WCVB. wcvb.com
  2. Federal Emergency Management Agency & Commonwealth of Massachusetts. (2023, July). MassGIS Data: FEMA National Flood Hazard Layer. Mass.gov. mass.gov
  3. First Street. (2023, July 31). First Street Flood Model (3.0) [Technical Methodology]. firststreet.org
  4. First Street. (2025). Climate Modeling Data. First Street. firststreet.org
  5. Massachusetts Department of Transportation. (2024, October). Road Inventory Viewer. GeoDOT. gis.massdot.state.ma.us
  6. Massachusetts Department of Transportation. (2025, March 7). Draft Study on Morrissey Blvd Alternatives Released for Public Comment. mass.gov
  7. MassDOT. (2024, March 25). Bridges | MassDOT Open Data Portal. opendata.arcgis.com
  8. MassDOT. (2024, March 25). Roadway Culvert Inventory. GeoDOT. opendata.arcgis.com
Acknowledgements: Courtney acknowledges the D3M@Tufts fellowship for supporting this research. Shan Jiang acknowledges the Eileen Fox Aptman, J90, and Lowell Aptman endowment for supporting this research.