**1. Introduction**

Inherently, hydrodynamic models are best validated with water level sensors, due to the precision afforded by defining the timing and depth of inundation at a location in an automated manner [1–4]. As a result of decreased technological costs, low-cost low-energy networks of water level sensors leveraging the Internet of Things (IoT) are beginning to dramatically densify the flood data available in urban environments in coastal areas throughout the world [5,6]. Hampton Roads, VA, USA, hosts one of these IoT networks called StormSense. The network functions as a flooding resiliency partnership between the Virginia Commonwealth Center for Recurrent Flooding Resiliency (CCRFR) and several coastal cities in Hampton Roads [7]. The network's primary goal is to monitor and transmit automated flooding alerts in real time when inundation occurs [8,9] However, an additional function of these sensors is the integration with federal sensor data from the US National Oceanic and Atmospheric Administration (NOAA) and the US Geological Survey (USGS) to validate and improve the Virginia Institute of Marine Science's (VIMS) flood forecast models [7–9].

However, when it comes to logistical considerations and the high cost of maintenance involved in deploying traditional *in situ* or remote water level observing systems, these factors can limit sensor density when even finer scale data are needed, and therefore impede these systems' ability to accurately monitor fine-scale environmental conditions [8,10]. In recent years, the combination of youth who are increasingly globally connected to the internet, and a growing population of retired professionals, poses an opportunity to create a wide-ranging and diverse network of citizen scientists with the capacity to span multiple societal themes [11,12]. Citizen science is public participation in conducting scientific research by non-professional scientists, typically following some form of informal training on data collection. While not a panacea for all inundation monitoring needs, citizen scientists can augmen<sup>t</sup> and enhance traditional research and monitoring. Their interest and engagemen<sup>t</sup> in flooding resiliency issues can markedly increase spatial and temporal frequency along with an e ffective duration of sampling. This can reduce time and labor costs, provide hands-on science, technology, engineering, and mathematics (STEM) learning related to real-world issues, and increase their public awareness and support for the scientific process. Naturally, a lack of su fficient professional oversight in citizen science endeavors can introduce caveats to overcome before wide-scale inclusion in an established coastal observing system, ye<sup>t</sup> progress in this underutilized resource is promising [4].

First seeing significant adoption in the US in the aftermath of 2012 Hurricane Sandy, citizen science flood monitoring e fforts first became useful through mobile phone pictures capturing inundation with a time-indexing landmark in view, such as a clock tower or local bank clock [13,14]. These pictures gave credence to the digital medium with the advent of enhanced-GPS, which leverages the Global Positioning System's (GPS) satellite constellation along with nearby cell towers to better triangulate a user's position on the ground. These tools have now begun to rival the utility of government-sponsored post-event flood monitoring e fforts, such as the USGS' high water marks [15]. While the latter approach a ffords confidence for model validation through a trusted agency for superior accuracy, the former possesses a greater capacity to document everywhere flooding occurs, with the inherent risk of potentially less accurate validation data. Regardless, collection of data at the local scale in public spaces where flooding is prevalent, such as streets, public right-of-way access spaces, and parks, can improve model prediction by properly resolving flow around small-scale features in the built environment [12]. Additionally, the model's predictive acumen can be enhanced via improved calibration of assumptions, such as: (1) Better friction parameterization of di fferent land cover types, (2) improved aerial elevation estimates of occluded roadway overpasses, and (3) identification of tidally-susceptible subterranean drainage infrastructure junctions (where tidal waters can enter city streets several blocks from the water's edge). Thus, quality assurance of flood validation data near these fine-scale features can become valuable model improvement assets through the proper training of a citizen scientist network [16].

Through technological progression, many e ffective methods for mapping inundation and flood depths have been developed using GPS, photo tagging, Augmented Reality (AR) image landmark recognition, and Quick Response (QR) codes [7,11,13]. Naturally, the emergence and growing necessity of smart phones in modern living has popularized the prominence of these recording methods. Additionally, the ease of access a fforded by mobile applications for making insurance claims, verifying flooding for municipal governmen<sup>t</sup> attention, and greater scientific aspirations has increased the intrinsic value of personal flood mapping [17,18]. Thus, flood-observing mobile applications, like "MyCoast" [19] and "Sea Level Rise" [12], or crowdsourcing web data geo-forms, like those implemented at the state [20,21], country [22,23], and international level [24], have emerged for myriad resiliency purposes. Typically, these applications exist to verify claims of flooding, validate flood forecast models, or inform long-term flood planning e fforts [19–24]. Mobile flood mapping platforms and applications have recently become information repositories that provide a living data archive of flood observation data with su fficient recording frequency and data density in urban areas where flooding is prevalent [25]. However, these tools have been shown to be of less utility in rural coastal areas, where statistically, less people are present and motivated to vigilantly monitor inundation, and where

enhanced-GPS signal strength is diminished due to less reliable cellular broadband coverage [26]. Yet, over time, these data sets can even become their own autonomous data-driven flood prediction models via sea level trend extrapolation when combined with Digital Elevation Models (DEM) [16]. Thus, high-resolution street-scale hydrodynamic models have recently found a new way to validate their predictions, and a cost-e ffective method for correcting erroneous elevation assumptions from aerial lidar surveys. This includes occluded areas in heavily canopied flood-prone areas and built infrastructure, such as box culverts, highway overpasses, and bridges that impact proper hydrologic drainage in flooding conditions [27].

A proactive and safe way to leverage these technological advances in citizen science flood monitoring methods without waiting for a major storm to elucidate inaccurate model assumptions is to map the incidence of "nuisance flooding." This approach takes advantage of mapping inundation in places where it frequently occurs with minimal danger to the reporter, and can identify issues with modeled flood forecasts without waiting for a major tropical or extratropical storm event to identify them first [12]. Hampton Roads, VA, USA experiences tidal nuisance flooding 12 to 18 times a year [28]. This is a frequency that amounts to no less than one cumulative week per year that low-lying streets in the region are inundated [29,30]. This chronic flooding fatigue can make it easy to forget that intermittent tidal flooding events cost cities and their residents time and money [30,31]. Of these tidal inundation events, the highest astronomical tide of the year has become known as the king tide [20]. While not a scientific term, a king tide is a name that refers to an exceptionally high tide, without the consideration of atmospheric amplification from wind or waves [21,23]. These predictable king tide events can be estimated far in advance and make coordinating and mobilizing a volunteer e ffort to track their inundation extent easy, while maximizing the opportunity for local weather impacts to potentially amplify the inundation observed [20].

This manuscript describes methods employed at VIMS to disseminate automated inundation forecasts called Tidewatch Maps. The forecasts function as an operational flood forecast model, which leverages the open-source Semi-implicit Cross-scale Hydro-science Integrated System Model (SCHISM) to automatically compute storm tide simulations throughout the entire US East and Gulf Coasts (Figure 1). SCHISM then translates those water level outputs to 220 localized lidar-derived sub-grid modeled sub-basins ranging from 1 to 5 m resolution to calculate 36 1-h geospatial flood depth layers covering all of Tidewater Virginia. SCHISM and the Tidewatch Map currently download inputs and update mapping outputs twice daily, every 12 h. Reliable inundation prediction depends upon accurate simulation of large-scale inundation of the tidal long wave during a king tide to successfully propagate from the ocean, through the continental shelf, estuarine systems, into creeks, and ultimately city streets, and rigorous conservation of fluid momentum and mass as flood waters permeate the built environment. These Tidewatch forecast maps were benchmarked in Hampton Roads by >100,000 GPS-reported high water marks collected by citizen scientists during two king tide flooding events occurring in 2017 and 2018.

The following sections highlight how coastal communities are being meaningfully engaged in coastal ocean observing mechanisms and the research e fforts they support. What follows is a description of: (1) A citizen science flood mapping project called *Catch the King* based in Hampton Roads, VA, (2) e ffective volunteer training methods using cell phones to provide meaningful GPS observations for effective model validation, (3) hydrodynamic modeling approaches used for expediently simulating and publicly mapping near-term inundation, along with (4) a summary of the results. The paper concludes with an identification of the modeling and monitoring challenges and potential solutions for modeling and citizen science e fforts in the future.

**Figure 1.** The SCHISM Hydrodynamic model grid used to drive the Tidewatch storm tide prediction model, and inset at right, are some example city-scale sub-grid model sub-basins for Norfolk (top) and Portsmouth (bottom) driven by SCHISM using Dirichlet boundary conditions.
