Translating FEMA risk data and socioeconomic indicators into decision-ready resource allocation strategy for geographically distributed operations.
Organizations managing assets, operations, or populations across geographically distributed regions face a persistent challenge: allocating limited preparedness and mitigation resources under uncertainty. This project addressed two core questions drawn from FEMA's National Risk Index framework - how expected annual losses differ across communities with varying levels of social vulnerability, and how accurately county-level hurricane risk can be predicted using publicly available socioeconomic indicators.
The analysis integrated three data sources at the U.S. county level: FEMA's National Risk Index (hazard exposure, Expected Annual Loss, and Social Vulnerability Index ratings), U.S. Census Bureau American Community Survey data (demographic composition, housing, income, and labor force metrics), and USDA county-level income and poverty data spanning 2018–2022.
The analytical pipeline included:
- Data cleaning, FIPS key matching, and multi-source joins across NRI and Census datasets
- Distribution analysis of Expected Annual Loss across SOVI rating tiers, identifying non-normal skew patterns requiring non-parametric testing
- Kruskal-Wallis hypothesis testing to assess statistically significant differences in hurricane loss across social vulnerability levels (total loss p = 0.019, population loss p = 0.006, building loss p = 0.015)
- Pearson correlation analysis to quantify linear relationships between continuous SOVI scores and each loss type
- Predictive modeling of Expected Annual Loss using eleven candidate models - KNN, Decision Tree, Polynomial Regression, Interaction Models, Linear Regression, XGBoost, Random Forest, Elastic Net, SVM, Neural Net, and a Naive baseline - evaluated by RMSE and standard error
- Variable importance analysis identifying labor force participation, armed forces percentage, hurricane risk score, and total population as the strongest predictors of financial loss exposure
Counties rated Very High or Relatively High on social vulnerability consistently experienced significantly larger hurricane losses than lower-rated counties - a gap confirmed statistically rather than descriptively. Population loss showed the greatest variation across vulnerability tiers of all loss types examined, indicating the human cost of hurricanes is more sensitive to community vulnerability than either building or agricultural loss.
The KNN model ranked first across all eleven models with an RMSE of 6.58E-08 and near-zero average percentage error when predicting Expected Annual Loss for Hillsborough ($533M actual vs. $533M predicted), Miami-Dade ($959M vs. $959M), and Palm Beach ($905M vs. $905M). Miami-Dade emerged as the highest-risk county across both absolute loss and social vulnerability dimensions, compounded by a poverty rate of 14.1% and housing costs running 51% above the Florida state average.
Three costed recommendations were developed from model outputs and benchmarked against empirical industry sources:
Targeting the highest-SOVI counties identified by the model. The financial projection is derived directly from FEMA/NIBS "Mitigation Saves" research, which documents that every $1 invested in hazard mitigation returns $4–$11 in avoided losses. Applied to a centered investment of $30M: $30M × $4 = $120M (low estimate), $30M × $11 = $330M (high estimate). The investment range itself is sized to Miami-Dade's EAL profile (~$900M+ annually) and the marginal cost of flood-resistant retrofits, utility hardening, and emergency shelter expansion across the top three counties.
Derived from standard state-level analytics unit cost benchmarks: data infrastructure and cloud capacity ($1.5M), staffing of 10–12 FTE analysts, engineers, and GIS specialists ($2M), predictive modeling and dashboard development ($1M), and hardware/security ($0.5M). Projected displacement savings of $10M–$25M per major hurricane are calculated using HUD disaster assistance cost benchmarks of $18K–$28K per displaced household - avoiding 400 displaced households yields $10M in savings, avoiding 900 yields $25M. Procurement savings of $8M–$12M annually are based on Florida's estimated $55M–$60M annual emergency procurement spend; a 15–22% efficiency gain from pre-staged resources yields $8.25M–$12.1M.
Sized across four spending categories: job retraining and workforce programs ($20M), rental stabilization and emergency housing ($15M), employer payroll continuity incentives ($10M), and county-level economic mobility programs ($5M). The projected EAL reduction of $20M–$45M is grounded in FEMA/NRI elasticity estimates linking a 1% improvement in unemployment or poverty to a 2–5% reduction in annual losses. Applied to Florida's hurricane EAL baseline: 2% of $900M = $18M, 5% of $900M = $45M, rounded to $20M–$45M for presentation clarity.
All financial assumptions and disaster-cost benchmarks are supported by:
- FEMA / NIBS "Mitigation Saves" studies (latest edition, drawing on 2017–2022 data) - source for $4–$11 mitigation return ratio
- HUD Disaster Assistance Cost Benchmarks (2018–2022) - source for $18K–$28K per-household displacement cost
- NOAA Hurricane Impact Data (through 2022) - contextual hurricane loss benchmarks
- USDA Economic Research Income Data (2018–2022) - county income and labor force inputs
- World Bank Resilience & Poverty Reports (2018–2022) - socioeconomic-to-loss elasticity support
- U.S. Census Bureau ACS (2018–2022) - all demographic and housing inputs at county level
R · tidyverse · randomForest · caret · tidycensus · ggplot2 · Kruskal-Wallis · KNN · XGBoost · Elastic Net · SVM