Left in the Flood

GIF
Flood damage by year
122 hard-hit, vulnerable counties got almost no flood-mitigation money
Annual Flood Event Records: Shows the number of records each year and their breakdown by the four flood types.
Positive-Loss Frequency and Typical Loss Size: Compares the percentage of records reporting positive losses and the median loss amount.
Record Share versus Loss Share: outer: each flood type’s share of records, inner: shows its share of total losses.
NRI Frequency and Historical Record Counts: Examines the association between NRI annualized flood frequency and retained historical records.
PR Charts

Inspiration

Floods are America's costliest disaster, but funds are distributed unevenly. Reactive relief fails to protect vulnerable communities, amplifying socioeconomic inequality. We built this to answer: Who suffers most, and does FEMA money reach those needing it most?

What it does

Left in the Flood is a spatial intelligence platform visualizing 27 years (1999 - 2025) of NOAA records. It also acts as a decision support system. For any county, our AI estimates federal recovery funding and recommends optimal mitigation measures like property acquisitions.

How we built it

Frontend: HTML, Mapbox, and deck.gl for an interactive 3D map.
Data: Python scripts clean NOAA files and join FEMA grants.
Machine Learning: Logistics and Regressions powers our Elastic Net funding model and Multi-label Logistic Regression recommender.
Scope: Synthesized 145,363 event records, machine learnt from 36,450 of them.

Challenges we ran into

Missing Data: Pre-2007 records lacked damage values, requiring heavy interpolation.
Long-tail Distributions: The costliest 1% of events cause 92.1% of recorded damage. Modeling this was statistically difficult.
Performance: Rendering massive 3D data required baking data directly into the frontend.

Accomplishments that we're proud of

Uncovering Gaps: Identified 122 vulnerable counties receiving under 1¢ of mitigation money per $1 of damage.
High-Performing Model: Our model catches 84.47% of funded mitigation measures.
Cohesive Storytelling: Unified scattered datasets into one clear narrative.

What's next

Broader Scope: Integrating Coastal floods due to higher median losses.
Dynamic Tracking: Adding year-over-year vulnerability metrics.
Causal Analysis: Linking mitigation projects to long-term economic resilience.

Why it matters

Vulnerable counties suffer more, yet rarely receive mitigation projects. Federal relief remains reactive rather than equity-driven. By quantifying these disparities, we turn disaster logs into actionable insights. Climate resilience must be accessible to everyone, not just the wealthiest.