**5. Conclusions**

A large-scale flood monitoring citizen science data collection effort was used to favorably validate an automated browser-based flood mapping service driven by a cross-scale hydrodynamic model predicting storm tide inundation in coastal Virginia, USA. The operational modeling effort for predicting tidal flooding can be mapped using multiple methods, ye<sup>t</sup> the most effective method was found to be the automated implementation of a street-level hydrodynamic model. The Tidewatch Maps implemented

by the Virginia Institute of Marine Science (VIMS) leveraged their SCHISM hydrodynamic model with inputs of: atmospheric wind and pressure data, tidal harmonic predictions at the open boundary, and prevailing ocean current inputs, such as the Gulf Stream. This information was successfully computed from a large scale model and translated to the street-level via SCHISM's computationally efficient non-linear solvers, and semi-implicit numerical formulations aided by a sub-grid geometric mesh with embedded lidar elevations.

Validation in the vertical scale found that the SCHISM model outputs via the Tidewatch web mapping platform compared well in Hampton Roads among the 32 extant water level sensors during the highest astronomical tide of the year on 5 November 2017, a king tide, yielding an aggregate RMSE of 3.5 cm. The region expanded its sensor base to 48 through an IoT sensor project, StormSense, to compare well again during the king tide on 27 October 2018, resulting in an RMSE of 3.7 cm. Horizontal validation was aided by time-stamped GPS flood extent data collected by citizen scientists through the world's largest environmental survey (in terms of the most contributions in the least amount of time), *Catch the King*. The citizen science flood mapping survey was established in Hampton Roads in 2017 and recruited volunteers through local, print, and social media outlets. The survey's organizers then trained the citizen scientists in the use of the free Sea Level Rise mobile flood mapping application at frequently inundated public spaces in the months leading up to each king tide event.

The citizen scientists' flood monitoring data formed time-indexed GPS breadcrumbs to form contours that were successfully aggregated and compared with the maximum inundation extents of the same time interval from VIMS' Tidewatch Maps. The data were filtered to minimize bias attributed to errors related to observing flooding duration, depth, and degree. Once the *Catch the King* survey data were filtered for these three things, it was found that the Tidewatch Map comparisons on 5 November 2017 had an overall mean horizontal distance di fference of 5.9 m (19.3 ft). The model comparison with the observations collected during the king tide on 27 October 2018 were found to be less favorable, yielding an average distance deviation 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. In each spatial validation e ffort, less than 9% of the surveyed data were excluded from the analysis.

Lessons learned from citizen science surveys have improved the model through cost-e ffective hydrologic correction of mission conduits for fluid flow. These were identified by filtered GPS observations that the model missed in its initial automated forecast, but were corrected in hindcast, in preparation for the next significant inundation event. Errors in hydro-correction did not relate to errors in friction parameterization of the model, but were more associated with flow pathways that were occluded from aerial lidar surveys. These areas included bridges, culverts, and stormwater drainage systems without tidal backflow prevention valves, which formed artificial dams in the digital surface model embedded in the forecasted Tidewatch Maps. Many of these identified areas have been corrected and have recently been used alongside the successful model validation in Hampton Roads to expand the forecast area of the Tidewatch Maps beyond southeast Virginia to include the entire coastal zone of Virginia in 2019.

As king tides are currently simply nuisance floods, which primarily inundate streets and driveways without significantly damaging infrastructural assets, the issues are presently geared towards tra ffic and transportation issues. Common concerns from citizen scientists involved in the *Catch the King* mapping events involved concerns regarding whether their vehicle could be safely street parked or if their vehicles needed to be safely moved into a garage during king tides. Others questioned whether they should take an alternate route to work or school or the store due to potential street flooding. As technology progresses, these questions will become more prevalent as we aim to ascertain whether modern route guidance mobile applications will be intelligent enough to account for intermittent inundation, or unintentionally lead vehicles down flooded streets simply because there is no tra ffic detected on them while an adjacent elevated street is congested. Some navigation applications, such as Waze, have aimed to crowdsource all road hazard data through their "Connected Citizens" program, but this method is only a temporary solution, as a model cannot currently automate road hazard flags for flooded locations where those particular app users are not or have not logged data.

Naturally, adeptly answering these questions becomes increasingly difficult once self-driving vehicles are involved. Thus, the outcomes of inundation modeling efforts for this tidal calibration effort will more significantly be realized once this trained citizen scientist army is deputized into post-hurricane surveys. Since 2011 Hurricane Irene was the last hurricane to significantly impact Virginia's Hampton Roads region, the Tidewatch automated mapping model has ye<sup>t</sup> to demonstrate widespread accuracy amidst a significant inundation event since the Sea Level Rise app's advent in 2014. The goal is to continue to improve the model with each *Catch the King* tidal calibration and train volunteers so they will be aware of where to find the latest flood forecast information, and how to collect meaningful flood validation data. Thus, this monitoring coordination approach with hydrodynamic modeling provided a novel procedural release of information to depict predicted maximum inundation extents for expediently effective model validation through the use of an overwhelming quantity of quality event data with relatively low risk to volunteer citizen scientists.

**Author Contributions:** Conceptualization, J.D.L., H.V.W., D.M., and W.A.S.; experiments conduction, J.D.L. and D.R.F.; formal analysis, J.D.L., D.S.; writing—original draft preparation, J.D.L. and M.M.; writing—review and editing, J.D.L.

**Funding:** VIMS' flood modeling research for this study was funded by the CCRFR. In 2018, *Catch the King* was financially sponsored by WHRO Public Media, the Hampton Roads Sanitation District, and school group involvement was encouraged through a gran<sup>t</sup> from the Hampton Roads Community Foundation, and the Batten Environmental Education Initiative. In 2017, *Catch the King* was funded by The Virginian-Pilot, The Daily Press, WHRO Public Media, and WVEC-TV. Funding for this article's publication was graciously furnished by the CCRFR and through FEMA gran<sup>t</sup> #FEMA-DR-4291-VA-011 to VIMS through the Virginia Department of Emergency Management.

**Acknowledgments:** The authors would like to thank the three reviewers whose kind suggestions and recommendations greatly improved the quality of the final published manuscript. *Catch the King* is supported by WHRO Public Media, The Virginian Pilot, the Daily Press, WVEC News 13, and the CCRFR. *Catch the King* also is made possible by the nonprofit groups, Wetlands Watch and Concursive Corp., creators and developers of the citizen-science 'Sea Level Rise' mobile app (freely available on iOS and Android platforms). *Catch the King's* adept volunteer coordinator is Q.J. Without her organization and the significant community support of volunteers, this effort would not nearly have reached the success that it has realized; a list of their names is available online: https://bit.ly/2ZLZwBS.

**Conflicts of Interest:** The authors declare no conflict of interest.
