**Degree**


The extent that duration of inundation was addressed and timing of when a flood event will arrive dictates the potential mitigating actions that may be taken. Tidal inundation events can easily be predicted through harmonic algorithms, and hydrodynamic models can improve upon this by informing citizen scientists, community planners, and emergency managers alike when the flood waters will arrive. This information is useful for personal preparation of one's home and assets that may be in low lying areas, route planning and guidance for personal and emergency response vehicles, and for scheduling road closures to minimize vehicular loss. Figures 4 and 8 illustrate the di fference an hour makes in terms of accuracy on model validation, and in the future, the recommendation for more frequent spatial mapping has already been recommended for the future development of Tidewatch Maps to eventually shift to 30 min time steps for the online time-aware layers for depicting more temporally-resolute flood mapping beyond hourly updates. However, presently, Tidewatch Maps are used to map all of Virginia's coastal floodplain via SCHISM's model outputs at a 1 to 5 m resolution (depending upon the accuracy of lidar point spacing for the model's underlying digital elevation assumptions). A 36-h Tidewatch forecast already consists of 37 state-wide coastal flood maps being automatically produced every 12 h. Thus, doubling that number to 72 iterative Tidewatch Maps per cycle is both computationally expensive for the model's post-processing, and impractical for users loading its flood forecasts via the web without newer technology to enhance loading times. Since users have most frequently accessed Tidewatch Maps using their smart phones to view its flood predictions during periods of significant power and internet outages, greater temporal resolution for 30-min update intervals is not likely to be implemented soon, as additional loading times are even more cumbersome for mobile devices.

**Figure 10.** Map of radial positional accuracy (in meters) reported by the Sea Level Rise mobile app in Norfolk's Hague during *Catch the King* 2017. Points in red were not included in the spatial comparison.

Aside from depths being directly validated via amplitude comparisons with automated water level sensors, surveyed points collected through *Catch the King* were merged with a DEM to translate the collected data to contours. While most modern smart phones have an altitude sensor, its error on accuracy is not sufficient to accurately report flood depths or meaningfully report heights above a reference datum. The Sea Level Rise app does display one's altitude in the app interface and records this with each point, but not all phone models share these data with the app or possess the internal hardware to report this. Thus, for the most reliable elevations, the GPS high water marks were merged with the topobathymetric DEM used to build the SCHISM model and the Tidewatch Maps, developed by the USGS [45]. Many citizen scientists were likely to test the app before venturing out to collect data, and several data points that were nowhere near water were collected. Instead, these points appeared in houses, apartment complexes, or traced around isolated puddles in parking lots that were non-contiguous with neighboring estuaries. These locations were flagged for use in storm water studies, and removed from this tidal inundation model validation analysis for any points >0.91 m (3 ft) elevation above NAVD88 were not included in the model comparison, as the king tide from neither year exceeded this height at any water level sensor in the region, and there was no significant rainfall

(>2 cm) accumulated during or preceding either *Catch the King* tidal flood mapping event. In other cases, users mapped tidal-connected drainage ditches that became inundated during the king tides and these points were included in the analysis (Figures 11 and 12).

The context through which the degree of inundation was monitored by citizen science data is made more useful through following proper training for data collection and appropriate data filtering. Data were collected by 722 volunteers in 2017 and 431 volunteers in 2018. These data were collected by over 20 di fferent smart phone models, which each vary in terms of relative accuracy due to the number of antennae included in each phone model to aid in triangulation of positional accuracy and for general clarity of cellular broadband communications through the device. As such, citizen science surveys are inherently less precise than those conducted by professional scientists using industry-standard GPS receivers capable of real time kinematics (RTKs). Since the high variation in phone models introduces variable accuracy, as does the number of GPS satellites in range, an estimated radial error metric is reported by the Sea Level Rise app for each GPS measurement by assessing the incoming signals from the global navigation satellite systems along with a correction stream. However, unlike professional survey equipment operated by trained professionals, smart phones are presently unable to achieve the 1 cm positional accuracy that RTK GPS tools can. Thus, points with a radial accuracy metric >10 m were not included in the spatial comparison (Figure 10).

**Figure 11.** Examples of obtaining aerial elevations from lidar and conducting hydro-correction to assure unobstructed ditch and creek channels persist for hydrologic transport of king tide inundation in the City of Hampton, VA. Red highlighted boxes correspond to areas depicted in Figure 12A,B.

**Figure 12.** Ditches extracted and represented at the sub-grid DEM pixel level for effective representation of drainage ditches leading to (**A**) the west edge of Tabbs Creek and (**B**) the south edge of the creek draining the fairways of Eaglewood Golf Course, both identified for hydro-correction in Hampton, VA, USA.

Upon filtering for these three things, it was found that the Tidewatch Map comparisons on 5 November 2017 during *Catch the King* 2017 had an overall MHDD of 5.9 m (19.3 ft). This statistic was calculated from 57,986 of the 59,718 total high water marks collected after less than 3% of the citizen scientists' measurements were filtered out for any of the six reasons previously noted for relative error on duration, depth, or degree of flooding. In a similar fashion, comparisons between the high water marks collected by citizen scientists during *Catch the King* 2018 observed a slightly less favorable overall MHDD of 6.2 m (24.6 ft), likely attributed to the winds from the mild nor'easter that occurred in the hours leading up to the event. This MHDD was calculated from 30,920 of the 33,847 total high water marks collected after 8.6% of the citizen scientists' measurements were filtered out of the surveyed data.

In the interest of improving future forecasts, it was found that less than 1% of the filtered GPS high water marks were still not within 50 m of the Tidewatch Map's predicted inundation raster. Further investigation into these sites identified two reasons for the discrepancy, both related to errors in hydrologic correction of the model's DEM calculated water depth assumptions. Figure 11 outlines a series of above-ground drainage ditches in Hampton VA, that occasionally become inundated when the water table rises with extra high tidal waters. Connection through these narrow drainage ditches can be obscured by thick canopied trees adjacent to the narrow tidal creeks and mostly non-tidal ditches that feed those creeks (Figure 12). The model's elevations are attributed to averaged digital elevations from aerial lidar surveys to source the DEM that the model uses to represent reality. Thus, the depths of the bottoms of these fine scale ditches (<1 m wide) were not likely to be correct unless the point spacing is extremely high. Naturally, this is acceptable, since the model was scaled to (at best) 1 m spatial resolution, and cannot accurately represent the slopes of such detailed drainage features without scaling to a 0.33 m resolution. Yet, these ditches were found to become tidal conduits for fluid movement capable of causing inundation far from the shoreline during king tides [46]. In other places, bridges over typically non-tidal creeks were not removed from the aerial survey data used to build the DEM, and removal of the occluding feature aided hydro-correction to correct the model's incorrect volume displacement in areas where entire creeks were shown to be dry due to the artificial dam imposed by a bridge, constricting proper fluid flow (Figure 13). Thus, one of the most important and immediately noticeable achievements that *Catch the King* accomplished for the hydrodynamic model's validation was the aid of hydro-correction for several small streams that were obscured in the aerial lidar surveys informing the Tidewatch Maps. In the case of several ephemeral creeks that temporarily became tidal during the king tide, the citizen scientists' survey identified locations where these ditches needed to be corrected (Figure 14) [47].

**Figure 13.** Conceptualization of macro-roughness features in an urban environment resolved within the street level inundation forecasts. In the example shown in (**A**), a bridge artificially obstructs flow from passing through to the hydrodynamic model grid cell below the one shown. With hydrologic correction provided by the citizen scientists via *Catch the King* (**B**) shows the grid opened, where a nearby flood contour can be extracted and applied to translate the inundation underneath the bridge and open flow to the opposite side, no longer impeding fluid flow.

**Figure 14.** Citizen Science flood extent observations aided in hydrologic correction for an ephemeral stream feeding Wolfsnare Creek in Virginia Beach, VA. (**A**) Citizen scientists mapped the tidal inundation extent approximately an hour before the king tide's peak. (**B**) Hydrologic correction fixed the lidar-derived DEM to permit flow through the small box culvert beneath the bridge to enhance the model's spatial accuracy via better estimation of cross-sectional flow and volume conservation.

For example, a typically non-tidal creek feeding Wolfsnare Creek in Virginia Beach was inundated during the king tide in 2017. *Catch the King* volunteers mapped the king tide approximately an hour before the king tide's peak, and the large initial mean horizontal distance difference from this cluster of points drew researchers' attention to investigate the hydrodynamic model's under-prediction of inundation. The error was traced back to a faulty elevation assumption attributed to obstructed flow underneath a bridge. VIMS researchers hydro-corrected the landscape to open flow using neighboring elevations from the DEM through the box culvert underneath the bridge and corrected ground elevations impacted by thick tree canopies surrounding a creek bed with low aerial lidar-point-spacing.
