**3. Results**

Spatial data collected for each king tide event were aggregated through the Sea Level Rise mobile app and shared online using interactive web maps, so that volunteers with minimal GIS experience could visualize their own GPS observations populate the Tidewatch model's predictions in near-real time (Figure 6). This level of data accountability implemented an open interaction where users could conduct their own analyses and make their own mean difference calculations through ArcGIS Online's distance and area measuring tools from their data in a web browser while viewing the public map. This data interactivity spurred high engagemen<sup>t</sup> and participation for students involved in STEM research or related educational classes. Figure 6A shows an aggregated point map of 59,718 high water marks superposed with the Tidewatch Maps throughout the greater Hampton Roads region and

showcases the extent of areas not covered by automated sensors that were surveyed through *Catch the King* in 2017 (Table 1).

Comparatively, 2018 surveyed an even broader area than was monitored in 2017, but with less density (Figure 4). For example, Figure 6B shows Norfolk's Hague, where thousands of GPS data were collected in both years. In areas where a significant point density is reported, data points can form more than simple flood contours when combined with digital elevation survey data. With su fficient point density, one can form their own observation-driven interpolated flood model for comparison with hydrodynamic simulation results. In this case, the areal extent of The Hague shown in Figure 6B yielded a 94% match with the raster polygon built from the interpolated GPS points, with the Tidewatch model slightly erring on the side of over-prediction. The cursory overview of the Greater Hampton Roads Region shown in Figure 6 shows the Tidewatch Map in blue, GPS citizen science observations as aggregated blue dots, which render and dis-aggregate based upon the zoom level, and water level sensors from the Tidewatch Charts as red dots. This legend theme will persist through the next several spatial comparison figures, and was designated through the Sea Level Rise app for the observation data, and for the model via a meeting of the CCRFR with emergency managers [44].

**Table 1.** A tabulated table of quantities mapped by *Catch the King* volunteers in coastal Virginia, USA, where the SCHISM-Driven Tidewatch storm tide model provided inundation forecasts.


Given that flood impacts for king tides are simply tidal calibration data that are likely to be aligned along similar elevation contours without intervening atmospheric conditions, linear distance metrics can be useful to compute spatial di fferences in relatively flat areas. The standard distance formula may then be computed by GIS software to calculate the di fference between each GPS data point to the nearest predicted inundated space. To compute this, the modeled geospatial flood depths served through VIMS' Tidewatch Maps were converted into vector data polygons, with the maximum flood extent representing the 0 m flood contour. As the volunteers were instructed to map inundation in their communities by dropping time-stamped digital GPS breadcrumbs, the citizen scientists' data should ideally represent the observed GPS flood extents, and in most places, the model had an overwhelmingly favorable agreement. Figure 7 shows an example in Norfolk's Larchmont neighborhood adjacent to a dog park, where flooding is frequent, and the 112 points were used to compare with the light blue modeled flood extents as a linear distance, and averaged to form a mean horizontal distance deviation (MHDD) metric, which yielded an average deviation of 2.67 m for this site during the 2018 king tide.

Likewise, several other areas, ranging from residential, commercial, and industrial land uses, during the 2017 king tide are featured in Figure 8. Since Tidewatch Maps provide more than simply a maximum inundation extent, unlike tidal depths estimated from a bathtub model or a sea level rise topographic flood elevation viewer, temporal accuracy can also be assessed through the GPS timestamps reported on each user's measurements through the Sea Level Rise app. Figure 8A,C,E,G, shows a model distance comparison of forecasts and data from 13:00 to 13:59 UTC. Figure 8B,D,F,H show observation data and model forecasted depths for the same sites an hour later from 14:00 to 14:59 UTC. These figures are used to show varying flooding conditions during the king tide on 5 November 2017, which occurred at 14:30 UTC, similar to those noted in Figure 3 during 2011 Hurricane Irene. Figure 8 shows individual sites where monitoring e fforts took place in 2017 and ultimately contribute to the overall figure of +/−5.9 m in horizontal di fference between the maximum extents predicted via the Tidewatch Maps and the 59,718 high water marks measured through *Catch the King*.

**Figure 7.** Illustrative comparison of mean distance difference of 112 GPS data points collected in Larchmont, Norfolk (2.67 m), compared with the 2018 king tide peak inundation forecast in blue.

Figure 8A shows data from the 2017 king tide for the same area as Figure 7 in Larchmont, Norfolk for 44 high water marks collected by 2 citizen scientists from 13:32 to 13:52 UTC to yield a MHDD = 4.18 m. The same site from an hour later is shown in Figure 8B, comprised of 61 high water marks collected by 1 volunteer from 14:30 to 14:39 UTC to yield a less favorable MHDD = 6.92 m during the peak inundation period in *Catch the King* 2017. Figure 8C depicts inundation during the 2017 king tide along the south bank of the Lafayette River near the Haven Creek Boat Ramp in Norfolk by 73 high water marks collected by 2 citizen scientists from 13:35 to 13:59 UTC to yield an MHDD = 4.67 m. The same site an hour later is shown in Figure 8D, this time featuring 136 high water marks collected by 3 people from 14:30 to 14:47 UTC to yield a less satisfactory MHDD = 6.29 m during the peak inundation period on 5 November 2017. Figure 8E showcases GPS data from *Catch the King* 2017 for the Lafayette Shores neighborhood, nestled in the east bank of the Lafayette River, through six high water marks collected by a citizen scientist from 13:49 to 13:52 UTC to yield an MHDD = 2.15 m. The same site from an hour later is shown in Figure 8F, comprised of 29 high water marks collected by 2 people from 14:30 to 14:47 UTC to yield a less favorable MHDD = 4.70 m during the peak inundation period. Figure 8G depicts inundation during the 2017 king tide along the north bank of Little Creek in the East Ocean View neighborhood of Norfolk via 93 high water marks collected by 2 citizen scientists from 13:44 to 13:56 UTC to yield an MHDD = 9.81 m. The same site from one hour later is highlighted in Figure 8H, now featuring 68 high water marks collected by 1 person from 14:40 to 14:45 UTC to yield an improved MHDD = 4.06 m during the peak inundation period during *Catch the King* 2017.

While the areas shown in Figure 8 were surveyed by few citizen scientists, the area shown in Figure 6B is one of the most frequently monitored areas in the Sea Level Rise app's history. During 2017's *Catch the King*, the area featured in Figure 6B was monitored by 27 different volunteers at different times (not all during the flood peak period) to form 27 king tide inundation contours for The Hague. These were mosaicked into a composite maximum extent contour map comprised of 1134 GPS points stretching 2.17 km to form a maximum extent contour for VIMS to compare with its Tidewatch Maps modeled inundation. The total distance walked and recorded using the Sea Level Rise app by all 27 volunteers for The Hague alone in 2017's *Catch the King* was 22.58 km (Figure 6B). This is 10.39× the composite's distance at this king tide inundation site, meaning >10× the actual effort, or about a 10× greater distance was walked than represented by the composite 0 m flood depth contour. As a result, these distances along the waterways that were travelled as effort expended by volunteers was significantly greater than needed to efficiently validate the flooding extents (by 10×), and this is not counting a volunteer's travel to and from each reported flood site.

**Figure 8.** Model comparison of forecasts and data from 13:00 to 13:59 UTC at left, and 14:00 to 14:59 UTC used to show varying flooding conditions during the 2017 king tide at four sites ranging from residential, commercial, and industrial land uses throughout Norfolk, VA.

For perspective, the grand total for over 1000 volunteers to map inundation distances across both *Catch the King* years was 631.35 km, with the total number of unique king tide contours travelled in Hampton Roads across both years adding up to 173.65 km (Table 1). Therefore, significantly more effort was used than needed to e ffectively map the site, with Norfolk's Hague experiencing the greatest duplicated e ffort. This was also indicative of overlapping e fforts among other high-density monitoring areas at public beaches in Norfolk, Virginia Beach, and Hampton representing the other greatest density GPS data density areas with >6× overlap. As noted previously, the increased density of GPS data proved to be a boon towards supporting the development of data-driven area maps, which was useful in The Hague, but were less useful on public coastal beaches where the water was not surrounded by land, with transient sand elevations that may vary from those embedded in the hydrodynamic model via the latest lidar elevation surveys. Thus, in 2018 s *Catch the King*, greater emphasis was placed on coordinating volunteers at registration to commit to mapping unique locations when communicating with volunteers via training events, and through print and social media to best value their time commitment and most e fficiently validate the model.

Aside from the horizontal GPS surveys reported through *Catch the King*, Tidewatch is routinely validated through automated water level monitoring sites. An overview of the water level sensor data extracted from sensors through Tidewatch Charts during *Catch the King* 2018 across all data points revealed a favorable average vertical accuracy assessment of 3.7 cm via the root mean squared error (RMSE). This metric was drawn from 28 StormSense water level sensors, and 16 tidal USGS Sensors and 4 NOAA sensors. Six of these sensors, including three NOAA, two StormSense, and one USGS sensor, are shown in Figure 9 from VIMS' Tidewatch Charts, as an example comparison of hydrodynamic model performance during 2018 s *Catch the King*. These charts from 2018 are labeled in Figure 6A with their four-character station abbreviations, for spatial reference, with their time series data shown in Figure 9.

**Figure 9.** Time series of six Tidewatch Charts' water level sensors' observations (red), astronomical tidal estimations (blue), and computed residual di fference (green), compared with SCHISM model predictions (pink) from Tidewatch Maps during *Catch the King* at 10:30 on 27 October 2018; UTC-5.

The region had 16 less water level sensors in 2017, and *Catch the King* in 2017 took place during a king tide with no additional amplifying wind or rainfall e ffects. The aggregate RMSE comparison in 2017 across 32 sensors was 3.5 cm, resulting in a slightly better agreemen<sup>t</sup> with the model than the 3.7 cm RMSE value reported in 2018 [27,35]. As a result of the "blue sky" conditions, 722 citizen scientists collected data in 2017, but their data was less dynamically interesting than 2018, which had less volunteers (431), due in part to a mild nor'easter that occurred on the night before *Catch the King*, making the weather less favorable for volunteers. The nor'easter brought 11.17 m/s (25 mph) sustained winds for nearly 3 h from 03:00 to 06:00 UTC on 27 October 2018 (yet contributed negligible rainfall), as seen in the residual fluctuations represented by the green line of each automated monitoring gauge's measurements in Figure 9.
