Locating real-time water level sensors in coastal communities to assess flood risk by optimizing across multiple objectives

Climate change is increasing sea levels and the frequency and severity of storms, which, combined with environmental and social factors, elevates flood risk in coastal communities. While real-time water level sensors have become more capable and affordable, most deployments have relied on qualitative judgments or limited quantitative criteria. Traditional quantitative sensor placement emphasizes maximizing spatial coverage and minimizing uncertainty, but often overlooks flood hazard, exposure of critical infrastructure, and social vulnerability. The research question addressed here is how to optimally locate real-time water level sensors across a coastal community by simultaneously considering sensor-network performance, flood-specific hazards, and social vulnerability, while incorporating decision-maker expertise. The study proposes a multi-objective optimization and GIS-enabled workflow to identify a prioritized set of feasible sensor locations that support flood risk mitigation and emergency response, demonstrated on Chatham County, Georgia, USA.

Prior quantitative methods for sensor placement and network expansion typically focus on two objectives: increasing coverage and decreasing uncertainty. Some studies combine network utility parameters, but none integrate flood risk, social vulnerability, and critical infrastructure exposure concurrently in sensor network assessment. Previous water sensor methodologies often minimize uncertainty in unsensed regions, implicitly treating all regions equally and omitting social vulnerability. Qualitative placement based on local experience lacks a quantitative rationale and risks historical biases. The literature emphasizes the importance of protecting critical infrastructure and accounting for social vulnerability in resilience planning, but these are rarely included in sensor siting. This study addresses the gap by combining sensor network metrics with flood hazard and social vulnerability measures within a multi-objective optimization framework.

Study area and feasible solution space: The method defines a full solution space L for potential sensor locations by the geographic extent (bounded by local rivers and ocean), feasible waterways (rivers, lakes, wetlands; excluding ocean-facing locations and marshlands for maintenance accessibility), existing sensor coordinates, and a grid resolution chosen collaboratively (100 m cells in the Chatham County case). Each feasible location is represented by its grid cell center latitude–longitude. Parameters (objectives) calculated at each feasible location i (one-at-a-time sensor addition): 1) Network coverage (Ci): Using an objective mapping (optimal interpolation) algorithm with a Gaussian correlation and 5 km decay distance, a water level surface is interpolated from existing sensors plus a candidate sensor at location i. Subtracting a LiDAR DEM yields an inundation depth layer. Grid cells with >20% modeled error (from a Gaussian error function based on sensor proximity and correlation) are masked. Coverage gain is the increase in the number of valid grid cells (≤20% error) compared to the current network without i. Aim: maximize added coverage. 2) Network uncertainty (Ui): With the same inundation computation, after masking >20% error cells, compute the mean percent error over the remaining grid cells. Aim: minimize uncertainty (equivalently maximize reduction in mean error). 3) Critical infrastructure facilities density (Ii): Using locations of facilities prioritized by local emergency management (hospitals, police stations, power facilities, schools), compute a kernel density at i with a quartic decay to zero at radius r (1 km in the case study). The kernel sums over facilities within r with value 1 at a facility and (1 − (d/r)^4) decaying with distance d. Aim: maximize proximity/density to prioritize monitoring near critical assets. 4) Flood zone (Fi): Assign flood hazard levels using FEMA flood zones as a proxy for flood likelihood and coastal hazard exposure. Higher priorities are given to zones VE and A (100-year flood exposure, with VE additionally subject to wave hazards), followed by AE, then AH/A99, moderate risk X_500, and low risk X. Aim: prioritize locations with higher flood likelihood. 5) Social vulnerability (Vi): Use the county’s Damage Assessment Priority Index (DAPI), combining socioeconomic indicators (poverty, SNAP, unemployment), vulnerable residential indicators (renter-occupied and owned with no mortgage), and vulnerable housing unit indicators (mobile homes; multi-unit homes). Each component is ranked (1 = most vulnerable), and DAPI is the sum of ranks; lower DAPI indicates higher vulnerability and higher priority. Aim: prioritize locations with higher vulnerability (lower DAPI). Optimization and decision support: Each parameter is treated as an objective in a non-compensatory multi-objective optimization. The algorithm identifies the Pareto set P of non-dominated solutions (locations where no objective can be improved without degrading another). The problem is posed as an unconstrained maximization with sign conventions to reflect minimize/maximize aims, but solutions are restricted to the feasible set L. The Pareto set, typically a small fraction of L, is visualized in a GIS to communicate with decision-makers, who then select final sites within solution clusters leveraging local expertise and practical constraints (e.g., site access, permits). Sequential expansion: After installing selected sensors, recompute parameters including new sensors and rerun the optimization to obtain an updated Pareto set. Repeat as resources allow, supporting continuous network expansion aligned with evolving community goals. Implementation notes: Grid resolution and extent are collaboratively selected to balance detail and computational cost (100 m used). Gaussian decay distance (5 km) and kernel radius (1 km) can be tuned to local conditions and data density. FEMA zones provide flood hazard nationally (US), while DAPI is locally constructed but conceptually generalizable to other social vulnerability metrics. A GIS interface supports interactive exploration of solution clusters and metrics.

• The multi-objective optimization reduced the impractically large set of 28,890 feasible locations in Chatham County to 381 non-dominated solutions (≈1.3% of the total), substantially narrowing options for decision-makers. • Spatial maps of parameter values identified where new sensors most effectively increase coverage and decrease uncertainty. For example, added coverage ranged from 0 to 2,745 additional cells; mean network percent error across locations spanned approximately 7.02% to 7.81% in the mapped extent; critical infrastructure kernel density ranged from 0 to 5.08 facilities/km²; DAPI values in the mapped extent ranged from about 1,187 to 3,411 (countywide DAPI range reported: 1,009–3,411). • The Pareto set formed distinct geographic clusters, further simplifying decision-making by allowing selection within cluster regions rather than among hundreds of dispersed points. • Sequential expansion demonstration: After installing new sensors at selected clusters (including inland expansion and tributary-focused placements), rerunning the model produced a new Pareto set that removed improved areas (e.g., cluster A) and highlighted persistent priorities (e.g., cluster C) while revealing new candidate regions. This confirms the method’s support for iterative, goal-aligned growth of the network. • Comparison with traditional two-objective approaches (coverage and uncertainty only) showed markedly different prioritized locations. The five-objective method retained some traditional solutions but also surfaced additional sites that account for flood risk, critical infrastructure exposure, and social vulnerability, improving the network’s relevance for mitigation and response. • The workflow effectively integrates quantitative optimization with local expertise (e.g., accessibility via private docks, governance of flood-gated tributaries, permit constraints), leading to feasible, strategically valuable installations.

The findings demonstrate that incorporating flood hazard, critical infrastructure exposure, and social vulnerability alongside traditional network objectives yields a materially different and more societally relevant set of prioritized sensor locations. The reduced Pareto set (≈1.3% of feasible sites) enables efficient decision-making while preserving flexibility for local considerations. By identifying clusters rather than isolated points, the approach supports practical siting within feasible zones given access, maintenance, and permitting realities. The sequential expansion results show that the methodology adapts as the network evolves, focusing attention away from improved areas toward emerging priorities, thus aligning with long-term monitoring and resilience goals. Overall, the multi-objective framework addresses the initial research question by providing a rigorous, transparent, and adaptable process that better captures the multifaceted nature of flood risk and community needs, thereby enhancing both mitigation planning and real-time emergency response.

This study introduces and demonstrates a multi-objective optimization and GIS-based workflow to strategically locate real-time water level sensors in coastal communities. By jointly considering network coverage, uncertainty, flood hazard (FEMA zones), critical infrastructure exposure (kernel density), and social vulnerability (DAPI), the method systematically narrows the feasible solution space and produces a prioritized Pareto set that supports informed, equitable, and practical siting. Applied in Chatham County, GA, the approach reduced ~28,900 feasible locations to 381 non-dominated options (~1.3%), facilitated decisions through solution clustering, and supported sequential network expansion as new sensors were installed. The methodology provides a roadmap for other communities, is adaptable to local data and priorities, and integrates quantitative rigor with local expertise. Future research and development could extend the framework to simultaneous multi-sensor optimization (with computational advances), incorporate additional objectives (e.g., installation/maintenance cost, communications reliability), refine dynamic weighting based on evolving community priorities, and generalize social vulnerability proxies where DAPI is unavailable.

• The optimization places sensors one at a time; simultaneous multi-sensor placement is not addressed due to complexity and computational cost. Outcomes in isolated regions may differ when placing multiple sensors concurrently. • Parameterization choices (e.g., 5 km Gaussian decay, 20% error threshold, 1 km kernel radius) and 100 m grid resolution affect results and were selected for local context and computational feasibility; different settings may be preferable elsewhere. • The feasible solution space excluded ocean-facing locations and marshlands for serviceability, potentially omitting informative but hard-to-maintain sites. • Flood hazard uses FEMA flood zones (U.S.-specific proxy); applicability elsewhere requires alternative hazard datasets of comparable quality and resolution. • The DAPI metric is locally developed and data-dependent; transferring the approach requires analogous social vulnerability measures, potentially at coarser resolution. • Practical constraints (site access, permits, infrastructure ownership) influence final siting and may limit implementation of some Pareto-optimal locations.