Analyses of policies that improve water quality often suggest that the costs far exceed the benefits. Keiser and Shapiro (2019a) suggests that this partially arises from the difficulty of accurately measuring the benefits. Measuring these benefits is often complicated by the lack of data on visitations. In this paper, we study the value of recreation amenities nationwide using data from mobile devices about aggregate visitor counts and dwell time at each water recreation site by home census block group. We combine the mobile movement data with data on water quality and weather to construct a comprehensive, novel, and detailed dataset of roughly 30k water-based recreational sites with linkage to recreation visits made by 22 million representative residents. Using these data, we construct aggregate share data of recreation visits from each census block group to each site. We develop a random coefficient logit model of site choice to estimate the welfare effects of water quality improvements in the US. Our results suggest recreators are willing to pay an average of $8.2 per trip for a 1-meter increase in Secchi depth in the sites they visited, ranging from $6.8 to $18.2 across census regions. We estimate that raising the water quality of all sites to match the cleanest site would yield $56.3 billion in benefits, while the annual welfare losses from the closure of the most popular and most polluted sites would amount to $18.6 billion and $2.6 billion, respectively. Our model provides a practical framework for water quality integrated assessment models, offering a spatially detailed and economically grounded approach to evaluate the benefits and costs of water quality policies.