The hidden consequences of sea level rise: assessing vulnerability of buried infrastructure to groundwater inundation in American Samoa

Over 625 million island residents reside in low-elevation coastal zones across the globe. Future sea level and flood model projections suggest that within a decade, over 30% of these people will live within the 100-year flood plain (Neumann et al. 2015). Many of these vulnerable populations are served by an extensive network of surface-based and subsurface infrastructure, which is particularly at risk to impacts of sea level rise (SLR). In the Western Pacific Ocean, regional sea levels have been rising at a rate of 3.2 mm/year since the 1990s (Hamlington et al. 2014). Like many tropical islands across the globe, the U.S. Territory of American Samoa (AS) is disproportionately vulnerable to SLR due to a high proportion of coastal development and limited land area (Nunn & Mimura 1997). However, Tutuila, which is the cultural and political center of AS, currently experiences an even higher rate of relative SLR due to the compounding effects of chronic geologic subsidence caused by viscoelastic relaxation following a disastrous 2009 earthquake. This subsidence, in addition to effects of global SLR, results in an effective SLR on Tutuila that is five times higher than the global average (Han et al. 2019). While the most noticeable effects of SLR typically manifest through surface flooding during acute or compounding events such as king tide or high wave events (Keener et al. 2021), a more insidious and chronic impact of SLR lies below, and occasionally above, the ground surface. Groundwater Inundation (GWI) has been identified around the world as an often-overlooked consequence of SLR (Rotzoll & Fletcher 2013; Masterson et al. 2014; Kane et al. 2015; Habel et al. 2017; Plane et al. 2019; Befus et al. 2020; Maliva 2021). This phenomenon not only causes inland flooding in areas disconnected from the sea, but can also raise the groundwater table, surreptitiously inundating buried infrastructure with salt water, which enhances corrosion and reduces lifespan through change in salinity and redox chemistry of the subsurface. Although there are few published case studies of GWI's effects on small tropical islands, its consequences when considered under future SLR projections suggest that GWI is likely to cause significant disruption to coastal and island communities in the future. For example, in the area of Waikiki on the Hawaiian Island of Oahu' Habel et al. (2017) used the MODFLOW model to determine that GWI is likely to put over $5 billion worth of real estate at risk of flooding with a SLR of 1 m. Knott et al. (2017) also used MODFLOW to determine that GWI reduced the service life of coastal roadway pavement in New Hampshire by 50–92%, and White et al. (2007) analyzed water table elevation data and found that GWI is the primary risk for the sustainability of drinking water supplies on small atolls such as Tarawa, Kiribati.

Water, power, and wastewater utilities in coastal regions are faced with significant climate adaptation challenges as they typically own and manage the majority of buried infrastructure. The American Samoa Power Authority (ASPA) is the sole provider of all three of the aforementioned services in AS and thus faces the unique challenges of managing a large number of critical assets in an isolated location that is especially vulnerable to climate change (Kumar & Taylor 2015; Han et al. 2019). While ASPA manages utilities across multiple islands in the territory, the economically and politically important Pago Pago Harbor area is one of the most vulnerable areas to GWI. The land fronting the harbor holds a large concentration of infrastructure built on or under coastal plains that lie just above sea level. These assets include approximately 91 km of subsurface water lines, 42 km of below ground wastewater lines, 20 km of buried power lines, and 76 km of roads. In addition, there are eight sewer system lift stations, five water main booster stations, 2,400 individual mapped buildings, and 1,330 water meters located in the inner harbor area as well. Staff members at ASPA have already reported instances of inundated sewer lines and corroded manholes, thought to be a consequence of currently rising sea levels. There are also many other subsurface assets in AS that are vulnerable to negative effects of GWI for which no data currently exists, such as On-Site Disposal Systems and underground fuel storage tanks. In order for policymakers and utility providers to plan resiliently and to adapt to SLR, decision makers need accurate information regarding the projected locations and impacts of SLR. Without a clear understanding of where GWI vulnerabilities are most acute, the utility cannot prioritize limited resources and focus adaptation activities in the most critical locations. Mapping high priority areas will yield a more effective adaptation response and significant cost-savings in the long term.

This study aims to address these issues by pursuing the following objectives: (1) assessing the impact of GWI on surface and subsurface based infrastructure by modeling inundation at various levels of SLR, (2) assisting planners and utility managers in AS by mapping those areas that are most vulnerable to SLR driven GWI, which will aid in prioritization of adaptation activities, and (3) quantifying the individual effects of both SLR and projected changes in future rainfall on GWI impacts in Pago Harbor area.

In order to assess the frequency and future risk of subsurface flooding on infrastructure, a MODFLOW based groundwater model was developed for Tutuila's Pago Harbor Region and included SLR data produced by the National Oceanic and Atmospheric Administration (NOAA) Office of Coastal Management and presented on NOAA's national SLR viewer (OCM 2016) to modify the base ocean elevation and the coastline position for multiple future climate and SLR scenarios. The range of possible SLR levels predicted by NOAA in the year 2090 was considered. Although the timing of when a given SLR value will occur remains uncertain, presenting a reasonable range of probable SLR heights ultimately gives present-day managers an understanding of risks that encompass the varying range of emissions trajectories possible during the next seven decades. Because climate drivers, such as precipitation, are also projected to be different in the future, our future scenarios also incorporated a dynamically downscaled groundwater recharge coverage of a Representative Concentration Pathway (RCP) 4.5 emissions pathway developed previously by Shuler et al. (2021) for the period 2080–2100 using the only available downscaled climate projections from Wang & Zhang (2016). The RCP4.5 pathway was selected as it represents the most extreme increase in precipitation, about 14% island-wide, of the available scenarios. Modeled water table elevations for present and future scenarios were then used to assess where subsurface and surface-based infrastructure would be permanently submerged below the water table, or affected by nuisance flooding, respectively.

The modeling framework used for this project was followed from the collaborative modeling framework developed by Shuler & Mariner (2020). Specifically, the model used for this study is derivative of three existing products: (1) a downscaled climate model produced by Wang & Zhang (2016) that was used as input to (2) a SWB-based water-budget model produced by Shuler (2019), which provided the present and future climate informed recharge coverages that were used as input to (3) an island-wide numerical groundwater model described in Shuler & Mariner (2020). Specific details about the aforementioned products are described in the respective publications and are omitted here for brevity. The modeling framework was produced in a reproducible open-source manner by employing the Python package FloPy to manage model development and to document all code-based model scripts via Jupiter Notebooks (Kluyver et al. 2016) in an open-source repository on GitHub (Shuler 2022). MODFLOW has been previously proven to be an effective model for simulation of coastal GWI in high-island settings (Masterson et al. 2014; Habel et al. 2017), and a newer study by Habel et al. (2019) suggests that using more computationally or time intensive density dependent models such as SEAWAT is not necessary to accurately simulate local groundwater gradients for successful GWI simulation.

The FloPy model was run using MODFLOW 2005 and the resulting water table surface was used to assess the depth to groundwater by geospatially subtracting the land surface height at each cell. This provided a spatially distributed estimate of the depth to groundwater throughout the entire model domain.

Stakeholder involvement was a key element and many of the critical input datasets were provided by ASPA, including water level observations, well extraction rates, and infrastructure locations. Two-dimensional as-built drawings of water, sewer, and buried electrical infrastructure (items 1–6 below) were obtained directly from ASPA as CAD or GIS-shapefiles. Shapefiles of roads and buildings were obtained from the AS Department of Commerce GIS-Portal (ASDOC 2015). Infrastructure considered in this assessment included:

• 1.

  Transmission and Distribution water mains

• 2.

  Sewer lines

• 3.

  Water system booster stations

• 4.

  Sewer system lift stations

• 5.

  Customer water meters

• 6.

  Buried electrical lines

• 7.

  Roadways

• 8.

  Buildings

Since the depth of buried assets is variable and none of the available data files contained vertical dimensionality, the average depth for all subsurface infrastructure was assumed to be 1 m. This estimate is considered to be conservative as staff members at ASPA have confirmed that many of their buried pipes are located even deeper, especially when placed under roads or other areas with vehicle traffic. Subsurface infrastructure included water, sewer, and buried electrical lines as well as sewer lift stations since most lift station components are located below grade. Surface-based infrastructure included water boosters, roads, buildings, and customer water meters. Roads and buildings shapefiles circa 2009 were obtained from the AS geographic information system (GIS) users' group. All shapefiles and as-builts were clipped to their extent within the inner harbor area shown on Figure 3. While the model boundary does include some outer villages on the north and south shores, infrastructure in these areas was not included due to boundary condition effects. To determine where infrastructure was flooded, all cells within the depth to water coverage(s) with a value less than 1 m for subsurface infrastructure and less than 0 m for surface infrastructure were aggregated into a single ‘flooded area’ shapefile, and geospatial locations of infrastructure were intersected with the flooded areas. This method was reproduced for each SLR and climate change scenario to calculate the total length or number of each infrastructure asset type that were modeled as flooded for each scenario.

Three different levels of SLR were selected for use in the model based on projections provided by the 2017 NOAA Sea level Viewer: 0.6 m (≈2 ft) SLR, 1.5 m (≈5 ft) SLR, and 2.4 (≈8 ft) m SLR. These sea levels are representative of different possible SLR projections at different times. Climate projections are often produced at the end-of-century resolution and managers often plan for these long-term timelines; thus, the year 2090 was selected as a key point in our assessment. Specifically in the year 2090, 2.4 m SLR is considered by the NOAA SLR viewer to be the ‘extreme’ scenario, 1.5 m SLR is the ‘intermediate high’ scenario, and 0.6 m SLR is the ‘intermediate low’ scenario (OCM 2016). These projected sea levels also correspond to other potentially useful points of reference; specifically, 1.5 m SLR is the ‘extreme’ scenario in the year 2070, and 0.6 m SLR is the ‘extreme’ scenario in the year 2040. The year 2040 is currently well within the engineering lifespan of most surface and subsurface infrastructure components, making this point of reference useful for informing present-day equipment procurement decisions. The year 2090 is also likely to have a different climate and thus a different groundwater recharge distribution than the present day. To account for this, the only available future climate informed groundwater recharge coverage was used (Shuler et al. 2021) as an input to the MODFLOW model. This future recharge coverage was derived from the only available downscaled climate model for AS (Wang & Zhang 2016), based on the RCP4.5 emissions projection for the period 2080 to 2100, which was selected as it represented the largest rainfall anomaly of Wang and Zhang's modeled emissions scenarios.

The model wrapper function was also applied in a sensitivity test of the model response to independent perturbations of each of the hK values and other key parameters including stream conductance, well pumping rates, and recharge rates. The final objective value, represented by the total length of flooded subsurface water and sewer lines, was calculated for each sensitivity test case and the resulting percent difference in this value between each test case and the base case scenario (no perturbations) was recorded. While it is not possible to directly quantify model uncertainty due to the iterative and non-linear nature of the model, sensitivity testing not only shows which parameters are most important, but also allows for an understanding of how much of the model uncertainty is derived from each parameter. Results of sensitivity tests are presented in Section 3.2.

Assessing impacts of GWI on surface and subsurface based infrastructure: For the present-day calibrated base case scenario, the model indicated that less than 1% of surface infrastructure, and less than 4% of subsurface infrastructure, is currently affected by GWI. This is a reasonable figure according to ASPA operations staff, and helps to validate the model's accuracy. For the different future SLR scenarios, GWI effects grade toward the most extreme case, where in the 2.4 m SLR scenario, which also includes the 2090 RCP4.5 climate regime, the model calculates that 30–90% of subsurface infrastructure, and 10–30% of surface infrastructure, may be affected by GWI or surface flooding, respectively. In this worst-case scenario, the model calculated that 21.4 km of roads would be flooded, 13 km of sewer lines, 7 of the 8 sewer lift stations flooded at their bases, 18.7 km of buried electrical lines, and 37.6 km of inundated water lines. For context, replacing 37.6 km of water lines would require approximately 6,200 individual 20-foot-long pipes, which if assuming each had a 12-inch outer diameter, would require 85 semi-truck loads to transport. The model assessed GWI impacts on eight individual infrastructure types, across a large region. While the exact locations of GWI impacts on each of these assets can be seen by plotting the associated shapefiles in the project repository (https://github.com/cshuler/Pago_GWI_Modeling), a higher-level synthesis is useful for managers and policy makers. This is shown in Table 1, which shows the percentage of the total number of point assets or the total length of linear assets affected by GWI for each scenario.

Quantifying the impacts of SLR vs. future climate on GWI: While the future simulations presented here included both future SLR and future downscaled recharge projections, isolated scenarios with only one of each of these drivers were also run to assess the contribution of each driver on the final results. For summarization, calculating the average cell-by-cell increase between the future-climate coverage (RCP4.5) and the base case climate coverage indicated that the future climate was on average simulated to be 12 ± 0.04% wetter than the base case. However, this 12% increase only translated into increases of 7.2, 1.4, and 0.42% of inundated subsurface water and sewer lines for the 0.6, 1.5 and 2.4 m SLR scenarios, respectively. This result suggests that the forcing from SLR, especially at the higher SLR levels, is the primary driver of increased nearshore groundwater levels rather than the increase in recharge at the magnitudes predicted by the Wang & Zhang (2016) projections.

The sensitivity of the model to each of its input parameters was assessed through an analysis that compared the extent of GWI impacts between different model runs each with an isolated perturbation in input parameter values. Four separate runs were conducted for each of the 15 tested input parameters, with each being individually decreased by a factor of 2× and 4×, and also increased by a factor of 2× and 4×. The reference output value used to quantify the comparison of each run was a domain-wide calculation of the total length of flooded subsurface water and sewer lines. Results indicated that the model was most sensitive to hK values, and specifically those in the alluvial geologic units that underlie villages containing a significant amount of development (Table 2). These were namely Pago Village (Vaipito unit) and Utulei Village (Utulei Unit). As expected, increases in hK reduced the amount and extent of GWI, and decreases in hK resulted in increases in GWI related flooding overall. Additionally, the model was quite sensitive to recharge and well pumping rates, which are both known to be major controls on groundwater heads.

The most direct application of this study is by managers and decision makers at utilities or agencies in AS who are faced daily with needing to make site-specific procurement and planning decisions for the future. The maps produced for this project have already been deemed ‘very useful’ to these ends by the executive director at ASPA. Since the catastrophic 2009 earthquake and Tutuila's acceleration in relative SLR (Han et al. 2019), utility managers in AS are already responding to the impacts of SLR. Workers at ASPA have noted instances of manholes corroding and heavy inflows and infiltration in sewer lines in low-lying areas. Our modeling suggests these problems will only be exacerbated by continued SLR and GWI, particularly in the lowest-lying areas of Pago Harbor. Another impact that is likely already occurring, but has not necessarily been attributed to GWI, is an increased depth of floodwaters in low-lying areas during rain events. GWI has been shown globally to be a major factor in exacerbating flooding from alluvial sources and heavy rainfall (Habel et al. 2017; Maliva 2021). All of the main population centers in the Pago Harbor area are located on alluvial plains at the base of mountain streams. Increases in the permanent soil saturation level underlying these plains reduces the infiltration capacity of these low-lying areas and can result in increased flooding or ponding when even smaller rain events occur.

In general, the presence of low-lying, inland sink areas identified by this work provides a clear caution against relying entirely on gray infrastructure such as seawalls as an adaptation measure. Even if a seawall or levee can hold back encroaching ocean waters on the surface, these structures will not be able to mitigate flooding or continuously saturated soil in areas that are hydrologically connected to the sea. In AS, seawalls have historically been the primary response to coastal erosion and rising seas. The issues noted herein only add to the many other commonly known negative side-effects of seawalls (Nunn et al. 2021), thereby reinforcing the idea that more advanced ‘green-engineered’ solutions or possibly even managed retreat from low-lying areas may be a more appropriate adaptation strategy in the long term.

There are many other islands throughout the world's oceans that are already or are very likely to experience similar consequences of GWI as Tutuila. However, each island hosts a unique set of vulnerabilities, partially controlled by development patterns, but also controlled by their geologic setting. To better understand the drivers of GWI vulnerability across different islands, this section examines the geology that controls the impacts and severity of GWI. Tutuila serves as a unique and interesting example, because although the Pago Harbor area is highly vulnerable to GWI, other parts of Tutuila are less vulnerable because of the island's diverse and varied geologic history. The southwestern part of the island, the Tafuna-Leone Plain (T-L Plain), is geologically younger and is characterized by an extensive lava delta fronted mostly by sea cliffs. Almost all of the T-L Plain is at or above 5 m in elevation, making SLR mostly inconsequential to the built environment. This shows the level of control that geologic processes have on vulnerability to SLR in volcanic islands; and how better understanding local geologic histories can be informative in region-wide adaptation planning efforts. Specifically, on Tutuila and other islands, two main factors control the conditions that lead to high vulnerability to GWI: (1) substrate age and (2) coastal geometry at the time of the last sea level high-stands. An island's age is one of the primary variables controlling the amount of developable land on an island, due to the effects of erosion and weathering that incise islands into a series of steep valleys and craggy peaks that preclude development. Generally, older islands are more eroded, and thus contain less developable land, which results in consolidation of development into the few flat areas, which are often thin coastal plains. Other examples of this type of terrain include much of the islands of Oahu (2.2–3.4 MYA) and Bora Bora (3.1–3.5 MYA). The eastern portion of Tutuila is included in this group because although relatively younger at 1.5 MYA, Tutuila's location within the South Pacific convergence zone creates very high rainfall rates that have enhanced erosion and weathering. Contrastingly, the Tafuna Leone Plain erupted within the last 10,000–20,000 years (Addison & Asaua 2006) and displays a flatter dome shaped geography characteristic of young shield volcanoes. Other young volcanic islands that share this terrain include Hawaii's Big Island (<0.5 MYA), Rapa Nui (<0.7 MYA), and Isla Isabella in the Galapagos (<0.7 MYA). These younger shields all generally have less eroded features, which make development away from the coast more common and cost-effective. Also, a lack of well-developed coral reefs, again due to younger geologic age, focuses wave energy directly on the coastline, thereby leading to development of sea cliffs that allow coastal development to be elevated above the sea, thereby simplifying adaptation to SLR.

Another geologic factor relating to an island's level of vulnerability to GWI is the coastal geometry at the time of the last sea level high-stands, which directly controls the prevalence of low-elevation coastal plains. These plains were formed in shallow depositional environments during major interglacial periods. Isotopic and marine records show that the Last Interglacial (130,000–120,000 years ago) sea level high-stand that ranged between 1–3 m created extensive coastal plains in the Hawaiian Islands (Ku et al. 1974; Szabo et al. 1994), which also manifest throughout the Northern and Southern Pacific (Muhs et al. 1994; Nunn & Mimura 1997). For example, Waikiki, the tourism-driven economic engine of Hawaii, is built entirely on one of these low-lying coastal plains, and in many places the ground sits around 2 m above MSL. Other developed areas on Oahu, such as the Eva Plain, were also deposited during these sea level high-stands, which caused shallow depositional environments to form, often behind barrier or fringing reefs. Thus, the presence of shallow lagoons and nearshore barrier or fringing reefs during the last interglacial high-stands is a major controlling factor in coastal plain development. Because these flat, nearshore areas are now desirable for residential and business development, and are easy to build upon, many of them have been extensively developed, creating a situation where some of the Pacific Island's most valuable real estate is situated in areas that were originally, and will likely soon again be, in a shallow depositional environment.

With any modeling study it is important to clearly understand the specific limitations and the conditions required to make use of results and to avoid overinterpretation. For this work, a key assumption relies on the accuracy of the modeled water table elevations, the DEM surface elevations, and the infrastructure locations relative to each other and to a common sea level-based datum. Lidar data for the 1 m DEM used in this study was originally collected in 2012 and represents accurate elevations only during a snapshot in time. It is well known that ongoing geologic subsidence is and has been changing AS's observed sea level baseline relative to the global datum; therefore, sea levels presented here should not be considered to be accurate in terms of absolute elevations. However, assuming that the subsidence induced land motion is primarily vertical in direction, the relative elevation differences between the base case and future scenarios used in this work should still remain accurate, as both the MODFLOW model and the geospatial analysis used the same 1 m DEM as their base datum.

The majority of uncertainty in the outputs likely arises from the MODFLOW calculated water table elevations due to the large number of difficult to constrain inputs used in the model. The model is limited by its most poorly constrained parameter, which, as shown by results of sensitivity testing, was the subsurface hK. Although the best available water level data was used to calibrate the model hK values, difficulty in constraining this parameter typically contributes the most to model uncertainties. Although this uncertainty cannot be propagated through the model (because it is run iteratively), the sensitivity testing revealed that uncertainties in the hK values of the most highly sensitive units, specifically Vaipito and Utulei, may be the most important components in the final synthesis of results. It should also be pointed out that the depth of infrastructure was assumed. Unfortunately, the complete lack of depth data not only precludes accurate vertical placement of features but also makes it impossible to assess the uncertainty associated with using a single depth. While a depth of 1 m was determined to be the best available assumption based on discussion with operations and field staff at ASPA, it is nonetheless a significant generalization. Finally, it should be noted that the horizontal accuracy of most utility-grade as-builts is subject to errors, as their locations are often poorly documented in records that are often quite old. Indeed, utility companies typically have to perform geophysical utility locates to avoid hitting buried pipes and lines while digging. Nonetheless, these datasets represent the best available infrastructure data, and with the groundwater model cell size of 5 m, it is likely that inaccuracies in the infrastructure locations do not significantly bias results.

Other notable assumptions include ignoring the dual-density effects of seawater, leading to our decision not to use more computationally expensive models such as SEAWAT or SUTRA that consider these effects. This decision was based on the findings of Habel et al. (2019), which used an even simpler approach and determined that a more complex modeling framework would not necessarily yield more accurate results. Although seawater recirculation will cause variation in water table elevations very near the coast, these are likely to be negligible at the region-wide scale where hydrodynamic lift in the water table will likely make up the bulk of effects of SLR on the aquifer. Additionally, because there are no dedicated monitoring wells in AS, water levels used for calibration were taken from shut off pumping wells as described in Shuler & Mariner (2020). The predominant assumptions in using this approach are that (1) the well recovery tests yielded fully recovered water levels, and (2) the effects of nearby pumping wells did not extend to the measurement point when the recovery tests were taking place. Therefore, the model was calibrated to a hypothetical state with no aquifer pumping. While this could lead to some inconsistencies if extraction wells were added into the model, the lack of other observation data precluded the use of a better method. Additionally, it was determined that the limited spatial extent of well drawdown cones, the variability of pumping rates, and the likely possibility of existing wells going offline and new wells being drilled by 2090 precluded the inclusion of well extraction in the model scenarios.

By intersecting a numerical groundwater model, future climate and SLR projections, and as-built locations of buried and surface-based infrastructure, this work produced estimates of the ‘hidden’ impacts of SLR, manifested as GWI in Pago Pago Harbor, home of the capital of AS. The primary future climate forcings applied in the model framework were NOAA's SLR projections at 0.6, 1.5, and 2.4 m, which bracket the uncertainty range for SLR in the region by the end of the century. Future climate forcing in the form of a modified groundwater recharge coverage was also included in our future modeled projections. Under these three SLR scenarios, the model predicted inundation of an average of 13, 44, and 63% of all buried infrastructure considered in the model, respectively. At higher SLR levels direct surface flooding will also be widespread, and our model predicts in the worst case of 2.4 m SLR that 18% of present-day surface-based infrastructure will be permanently flooded. Vulnerability to GWI impacts was greatest in areas of heavy development built upon the relatively low-elevation coastal plains at the valley mouths including Fogotogo Utulei Pago and Aua Villages. At 0.6 m of SLR, the projected increases in rainfall contributed somewhat to increases in GWI, but at higher levels of SLR impacts of GWI were almost completely due to SLR, and climate forcing was found to be an insignificant factor.

Overall, the results of this study have farther reaching implications beyond AS that can provide insight to other island communities likely to be facing similar challenges. Many island societies have historically been based around a paradigm of static sea levels, and it is clear that planning decisions can no longer be made within that framework. The maps produced by this model are by no means an exact prediction of the future, but they can assist managers in identifying the lowest-lying areas and zones where infrastructure intersects and may be vulnerable to both surface induced and groundwater caused SLR impacts. Recommendations based on the results of this work could include locating new utilities in areas not prone to GWI, or if existing utilities are already located in these areas, replacement timelines should consider reduction of service life and replacement with corrosion- and water-resistant materials. Further research to better document the actual depths of subsurface infrastructure would be useful to refine the estimates of impact presented herein. The conditions modeled in this study are often overlooked by or unknown to managers, and it is hoped this work helps to contextualize the risks associated with SLR in these types of terrains, and helps to prioritize where adaptation actions can have the greatest surface and subsurface benefit.

The models and inundation maps were co-produced with operations and engineering staff at ASPA, including Katrina Mariner, Wei Hua-Hsien, Will Spitzenberg, and Danielle Mauga, and the author is thankful to ASPA's leadership, especially Executive Director Wallon Young for the agency's support and openness to encouraging and applying the best available science. This work would not have been possible without the critical mentorship of modelers at the USGS Pacific Islands Water Science Center, including Scot Izuka and Kolja Rotzoll, as well as my mentors at the University of Hawaii, including Aly El-Kadi, Henrietta Dulai, and Craig Glenn. All shapefiles, data, model files, and maps are open-source and are available on GitHub (https://github.com/cshuler/Pago_GWI_Modeling). Funding for this work was provided by NOAA's Climate Program Office through the Pacific Research on Island Solutions for Adaptation (RISA) program, Grant number: NOAA-OAR – CPO – 2015 – 2004099.

All relevant data are available from an online repository or repositories: https://github.com/cshuler/Pago_Samoa_GWI.

The authors declare there is no conflict.