Using a ROMS reanalysis containing ocean level variability data, we define a reference surface that serves as a valid proxy for MHHW. We apply this proxy in producing SLR flood probability maps distinguishing direct marine overland flow from topographically isolated flooding. Our results show that when comparing our model products with flood maps produced using the TCARI method, significant differences are found under NOAA RSLR scenarios for 2050 and 2100. The flood maps reveal several features that are critical to appropriate interpretation and application of these products. These are discussed below.
Disagreement. Analyzing the TCARI and ORS model products highlights a difference pattern in the form of a “fringe” surrounding flood projections where the two methods otherwise agree. We find that the fringe zone is typically a TCARI product. That is, the ROMS method in general projects less flooding than the TCARI method.
As shown in Figure 4, this disagreement reflects a difference in the shape and location of the probability distributions of the ORS and TCARI products. The TCARI tidal distribution has a tighter spread and it is shifted right compared to the ROMS distribution. This shift represents a departure from the NOAA tide station observations of water level variability and creates an increase of at least 0.03 m in mean flood height. As a result, the TCARI method projects increased flooding compared to the ORS method. Figure 4 also reveals that the daily high-water values near the mean occur more often in the TCARI distribution than in the ROMS and NOAA distributions and consequently a tighter CDF after the convolution.
Figure 5 illustrates how the spread of the CDF will determine the inundation probability value for a given pixel. When selecting the value that corresponds to any flooding at all (>0 m flood-depth) the probability will be lower when the spread is larger (Note: (a) = 1-0.27 = 0.73 < (b) = 1-0.18 = 0.82). Thus, if we were to map flooding with a probability of at least 80%, CDF (a) would select the hypothetical pixel shown in Figure 5. Whereas CDF (b) would not, and would therefore generate a fringe.
Figure 4. Probability density functions of daily highest high-water variability as obtained from the NOAA tide station in Honolulu Harbor, the ROMS Reanalysis, and a normal distribution of the mean of daily higher high water from the TCARI tidal surface. Vertical lines correspond to the mean of the respective distributions.
Figure 5. Comparison of two CDFs with the same mean but different flood depth spread. The figure illustrates how a broader spread results in a lower probability for corresponding flood depths.
Flood Patterns. The distinction between marine overland flooding and groundwater flooding hinges on whether a flooded area is directly connected to the ocean. The difference may be determined by a single pixel that allows for a region identified by ROMS as topographically isolated, to be mapped by the TCARI method as flooded by marine overland flow. Given the difference in the probability distributions of the TCARI and ROMS data, there is significant opportunity for this to occur. Single pixels or a small group of pixels can connect otherwise isolated areas and may serve as tipping points or flood pathways that open areas to flooding by marine overland flow. The differences that result may actually be physically meaningless as these low-lying inland areas will likely flood anyway, if not by marine overland flow, then by groundwater inundation.
Roads. Roads may be flood conduits or barriers to flooding depending on engineering style. This needs to be recognized when considering adaptation plans. On one hand, when roads are located near the shoreline at low elevations, they may function as waterways. A good example of this is shown in Figure 3 where roads that branch out from the Ala Wai Canal are seen to channelize marine overland flow and promote flooding farther inland. In addition, low elevation roads may connect low-lying areas that might otherwise not experience direct marine flooding. On the other hand, raised embanked roads prevent overland flow, potentially creating areas of topographically isolated stagnant water. Additionally, they could interfere or eliminate tide and wave driven circulation that would eventually become important in maintaining water quality given the essentially permanent nature of sea level rise3. Embankments may also direct flooding to otherwise dry land parcels. It is important to note that topographic barriers to direct marine flooding such as embanked roads, sea walls, and other structures fail to prevent storm drain backflow and groundwater inundation. As such, models that depict the impacts of sea level rise must include all relevant flood sources30.
Topographically isolated locations. Topographically low-lying areas may be flooded by groundwater inundation or storm drain backflow. These areas need to be clearly identified in flood projections in order for engineers and planners to find solutions to groundwater-flooding-specific adaptations and design resilient future communities.
King tides. The hydrostatic projections presented in this study do not include dynamic ocean processes such as wave overtopping, coastal erosion, or other physical oceanographic processes. However, they are useful for visualization of extreme tide impacts. For instance, Thompson et al.25 projects that by midcentury, coastal sites will see a dramatic increase in king tide frequency. Additionally, under the NOAA Intermediate scenario, global mean sea level is projected to reach 0.3 m by 205023. A hydrostatic map depicting 0.6 m of flooding is useful for illustrating a 0.3 m king tide on top of the 0.3 m SLR projection.
Probability-based maps. Flood-depth and probability can both be used as thresholds depending on the user’s objective and the situation that is being analyzed. Some scientists tend to prefer probability values that correspond to standard deviations in a normal distribution (e.x., 68%, 95%, 99.7%). Similar values were used by Mastrandea et al.46 when developing the IPCC AR5 likelihood scale. Other professional fields might prefer different values as thresholds when determining risk, exposure, vulnerability, and other criteria. For instance, NOAA uses 20% and 80% as the confidence bounds of their flood mapping methodology47. However, the NOAA Sea Level Rise viewer main maps do not make the level of confidence immediately visible. In addition, the flooded areas are displayed with blue shades but there is no flood-depth value associated with the colors. Although there is uncertainty associated with this type of mapping, the lack of flood-depth values leaves users, especially the ones with publicly facing adaptation projects, with poor understanding of potential damage related to flood depth. For instance, a 15 cm flood, which can be associated with a king tide event, would be a critical threshold for transportation engineering as it is considered likely to stall small vehicles48. However, a 15 cm temporary flood in open spaces or recreational areas might not be a reason to trigger expensive and disruptive adaptation efforts.
Probability of flooding and flood depth values are essential to provide map users with a perspective of the severity of the flooding. User-defined probabilistic flood maps have the potential to open a new world of adaptation efforts. Therefore, although it may prove difficult to implement, interactive websites that allow users to access GIS layers by choosing probability or flood-depth values should be developed. This is especially enticing given that similar information is already available (e.g., NOAA SLR Viewer, PacIOOS SLR Viewer), it just needs to be reconfigured in a user-friendly form.