**2. Methods**

#### *2.1. Citizen Science Flood Monitoring Data Collection*

*Catch the King* is a citizen science GPS data collection effort centered in Hampton Roads, VA, which aims to map the king tide's maximum inundation extents, with the goal of validating and improving predictive models for future forecasting of increasingly pervasive nuisance flooding [16]. The aptly-named effort is the world's largest simultaneous citizen science GPS flood data collection effort, and it is aimed at benchmarking the highest astronomical tide of the year, the king tide. Certified by Guinness World Records for having 'the most contributions to an environmental survey' on the planet, *Catch the King* was effectively publicized and promoted by the local news media, in Hampton Roads, VA [27,32]. High citizen engagemen<sup>t</sup> during a king tide inundation event resulted in an average of 572 GPS-reported high water marks being recorded per minute during the hour surrounding the king tide's peak [12]. Time-stamped GPS flood extent measurements and photographic evidence were reported via the free 'Sea Level Rise' mobile application, coinciding with the king tide, observed at 13:32 UTC (Universal Time, Coordinated) on 5 November 2017, in Hampton Roads, VA. Ultimately, *Catch the King* surveyed a total of 59,718 high water marks and 1582 photographs through 722 individual volunteers in its inaugural year [32,33].

Following 2017s success, *Catch the King* was repeated during the highest astronomical tide of 2018 on 27 October. This event was less attended as it immediately followed a strong nor'easter in a day prior, on 26 October. Still, significant response from the event's many volunteers, fueled by the local media partners' coverage leading up to the event, and 42 separate volunteer training events held all over Hampton Roads, resulted in 347 participants collecting 33,847 time-stamped GPS maximum flooding extent measurements and 458 geotagged photographs during the event [34]. There were 141 additional volunteers who collected 3881 GPS data points during the nor'easter's peak flood period coinciding with the high tide the night before the king tide. Combined, those totals are 488 participants and 37,728 GPS points during the 2018 *Catch the King* and nor'easter event, with 431 of those being unique volunteers and 57 having mapped the king tide on both the 26 and 27 October 2018 [35].

As the sea level rise and tidal flooding increasingly impact coastal Virginia, *Catch the King* offers residents a chance to crowdsource vital information about the tides' reach. Thus, time-stamped GPS data points and photographs were collected by volunteers to effectively breadcrumb/trace the high water line by pressing the 'Save Data' button in the 'Sea Level Rise' mobile app every few steps along the water's edge during the king tide's peak in 2017 and again in 2018. Initial responses collected by the free 'Sea Level Rise' mobile application were relayed to the event's dedicated volunteers via the local news media event's volunteer coordination Facebook page. In the days following each event, several people provided additional GPS data points they collected through Esri's Geographic Information System (GIS) Collector App and crowdsourcing geo-forms through ArcGIS Online that were missing from the initial cited statistics (Figure 2). This occurred as a few volunteers noticed their data initially missing from the interactive map as it was published in the Daily Press and The Virginian-Pilot among other media sources, and were later included in the final event totals and added to the final data map [27]. These updated totals were ultimately reported back to the community a month after each inundation event in 2017 and 2018 through a volunteer thank you and data review event [16,35].

**Figure 2.** Catch the King (CtK) citizen engagemen<sup>t</sup> time series chart corresponding to major news releases used to garner successful volunteer training participation and support leading up to the king 173 tide on November 5, from July 30–November 10, 2017. According to ArcGIS Online's data metrics, the volunteer invitation story map received 10,137 page views, in little over 3 months for an average of 93 page views per day. Story Map at: http://arcg.is/1f8W1q.

#### *2.2. Volunteer Coordination for Flood Extent Validation*

The volunteer coordination effort involved a hierarchical scheme led by an adept volunteer coordinator, Qaren Jacklich, from the Chesapeake Bay Foundation who has successfully led the 'Clean the Bay Day' litter and debris removal initiative in Chesapeake Bay for several years prior to *Catch the King*. Below the volunteer coordinator were over 120 volunteer "Tide Captains," who led smaller organized groups of volunteers. In many cases, these tide captains were knowledgeable school teachers, volunteer organization leaders, and enthusiastic users of the Sea Level Rise mobile app, who trained neighbors, friends, and family in their community over a series of separate volunteer training events held all over Hampton Roads, VA, USA [16]. Citizen scientists were trained in the use of the 'Sea Level Rise' mobile application to capture three types of flood data useful for model validation:


**Figure 3.** Comparison of GPS maximum flood extent observations (depicted as blue dots) in Hampton, VA, following debris lines remaining after Hurricane Irene 2011, used to geospatially verify model performance at (**A**) 00:00 UTC and (**B**) 01:00 UTC on 28 August 2011. GPS data collected by volunteers effectively illustrate the challenge with temporal matching for model comparison.

Training volunteers in proper, uniform data collection and appropriate geospatial filtering for these three data types is critical to recording an accurate location history for mapping tidal flooding and ensure the most effective approach for collecting error-free data for model validation. More than 45% of the volunteers were directly trained in a training session in 2017s event, while nearly 60% attended a training in 2017 [35]. However, in both years, many volunteers were still able to register as users and meaningfully map flooding in their community through Catch the King without any formal training, partially owing to the relatively intuitive interface of the app.

The 'Sea Level Rise' mobile app streamlined data collection and made model validation easier through a hierarchical internal quality assurance mechanism. This allowed the Catch the King event coordinator to limit participation to certain trusted registered users, and filter data permissions, such as photo uploads and GPS data collection, only to certain trained users (labeled as 'Champions' in the app interface). Post-event, tide captains and the volunteer coordinator could download their event data as csv files after the specified time window for their flood monitoring event expired, and even retroactively flag volunteer's data where consistently erroneous data points were measured. The resulting maps shown in the results and discussion sections present these dense data maps surveyed during the 2017 and 2018 king tide inundation events, and present the mean horizontal distance di fference (MHDD) comparative spatial calculations between the modeled maximum flood extent contours and citizen science flood validation data sets for each king tide flood event, followed by the lessons learned.

#### *2.3. Tidewatch Storm Tide Modeling*

Each year, prior to the king tide flood event, researchers at VIMS and the CCRFR design a web map to direct volunteers to public places that are forecasted to flood during the King Tide using VIMS' hydrodynamic models. This method proved impactful, as the 2017 *Catch the King* volunteer recruitment story map reached over 10,000 page views before the king tide in less than 3 months (Figure 2). Thus, the story map e ffectively conveyed the value of inundation forecasting by showcasing flooding impacts during the last major storm event in the region, 2011 Hurricane Irene, and the importance of time-stamped GPS data for tidal calibration and event calibration of models for improvement purposes (Figure 3). This was valuable to visually explain the value of accurate data collection both temporally and spatially to the citizen scientists.

As Figure 3A shows, on 28 August 2011 at 00:00 UTC after Hurricane Irene, 30 min before the storm surge peak was observed at the nearest sensor, the 14 GPS points that comprise this maximum extent contour compare well with an MHDD of 4 m to the nearest model grid cell center point. However, the model's prediction from an hour later, 30 min after the storm surge peak was observed at the nearest sensor shown in Figure 3B, the maximum flood extent compares less favorably, changing the MHDD to 6.5 m, and illustrating the burden of timing for reliable model comparison. Thus, during the king tides in 2017 and 2018, GPS data points were collected by each year's event's many volunteers to breadcrumb maximum inundation extents in public spaces and time-stamped (Figure 4). This approach was used to coordinate and validate the flooding extents across 17 coastal cities and counties in Virginia, USA, by enlisting the aid of over 1000 volunteers for approximately an hour once a year to walk outside and press the 'Save Data' button in the 'Sea Level Rise' app every few steps along the water's edge near them during the king tide.

The approach to presenting time series information and inundation areas for a flood model at the street-level can be a difficult task for development and comprehension. To simplify the approach, the open-source SCHISM model was developed at VIMS and used to compute Tidewatch's temporal-spatial inundation maps [36]. SCHISM is an open-source community-supported modeling system, designed for the effective simulation of 3D baroclinic circulation across ocean-to-creek scales. The model incorporates a wide range of physical and biological processes in a comprehensive modeling system that has been validated in many world-wide applications, ranging from general circulation [37], tsunami inundation [38], storm surge inundation [39], ecology [40], sediment transport [41], and oil spills [42]. The model is uniquely capable of accurately representing physical structures (both nature-based and engineered) in an inundated area in the model computations, not simply in the output displays. Furthermore, the outputs from this model can be nested with other hydrodynamic grids to provide street-level (1–5 m scale) urban inundation predictions for individual land parcels [43]. The results may be presented as high-resolution movies of flooding scenarios over multi-day periods, including tidal cycle variations, or translated and converted into GIS mapping applications, as seen here with Tidewatch for added utility.

**Figure 4.** Comparison map of *Catch the King* citizen science king tide data collection between 2017 (blue dots) and 2018 (green dots); points are aggregated by density. Yellow polygons illustrate a public spaces inventory, where volunteers were encouraged to map, while green-labelled park polygons were places where official *Catch the King* volunteer training events occurred. There were 35 such training events in 2017 and 43 in 2018: http://arcg.is/1vK8ru.

SCHISM is used to drive VIMS' Tidewatch's storm tide inundation maps, and the automated workflow to accomplish this is sub-divided into three tasks: (1) Preprocessing of the model grid and retrieval of the hydrodynamic model inputs, (2) SCHISM model simulation, and (3) post-processing retrieval of SCHISM inundation model outputs for GIS mapping, as illustrated in Figure 5:

	- a. Atmospheric data: The SCHISM hydrodynamic model is used to automate storm tide simulations based upon atmospheric wind and air pressure data available from the 05:00 UTC and 17:00 UTC updates of the US National Weather Service's NAM-nest 5 km atmospheric forecast model. For the 2017 and 2018 king tide flooding events, the 17:00 UTC, atmospheric forecast update from the previous night was used.
	- b. Flux boundaries: Riverine boundaries defined at waterfalls and key discharge points at the fall lines of Chesapeake Bay's major estuaries were driven by flow rates predicted by the national water model and obtained at intervals similar to the atmospheric data.
	- c. Open boundaries: Inputs for tidal open boundary grid nodes were harmonically computed and estimated for the amplitude, phase, and frequency for 16 tidal constituents.
	- a. Initiate simulation after archiving the previous run, and complete a successful check for all input files.
	- a. Extraction of source data from the station output from the 2D hydrodynamic forecast simulation.
	- b. Clear the previous run's data to archive, and import new simulation data to differentiate each day's morning and evening simulation updates,
	- c. Construct a web-enabled time-aware street-level GIS map from SCHISM grid outputs.
	- d. Publish the output map of 37 time-aware rasters overlaid with Tidewatch Station time series data.

**Figure 5.** Automation flowchart for SCHISM model initiation and simulation for every 36-h forecast. The Tidewatch model updates twice daily, outputting mapping outputs at AdaptVA.org at noon and midnight local time (05:00 and 17:00 UTC).

Upon completing these steps, model and visualization performance metrics indicated, on average, that step one requires 1 h to complete pre-processing, step two requires 1.5 to 2.5 h to run the simulation (variable based upon volume of inputs), using 72 nodes on VIMS' high performance computing cluster, and step three requires 2 to 3 h in post-processing to combine the model's binary outputs, and translate/index them to hourly geospatial outputs to a convenient interactive web-map presentation format. The SCHISM model consumes approximately 2.40 GB of input data to model the US East and Gulf Coasts for every simulation, twice daily (1.75 TB annually). Comparatively, the SCHISM model outputs the mosaicked results of over 200 combined sub-grid sub-basin rasters to form the Tidewatch Map's composite 36 h. time aware layer cache at 16 zoom-level pyramids for a 21.25 GB product after each simulation (15.48 TB annually). These steps allowed users to interact with flood data from the global scale to street-level in a single web map (Figure 6).

**Figure 6.** (**A**) Overview of study region in Hampton Roads, VA, featuring the 2017 king tide maximum inundation forecast from the Tidewatch Map in blue, GPS citizen science observations as blue dots, and water level sensors from the Tidewatch Charts as red dots. (**B**) Inset shows a high-density concentration of flood validation data in the historic Hague region of Norfolk, VA, USA, where the model had favorable agreemen<sup>t</sup> with the citizen science observations. Here, the model yielded a 94% spatial match, and a 4.2 m MHDD. See the map online: http://arcg.is/1HLOPS.
