Using a Bayesian network to understand the importance of coastal storms and undeveloped landscapes for the creation and maintenance of early successional habitat

Abstract

Coastal storms have consequences for human lives and infrastructure but also create important early successional habitats for myriad species. For example, storm-induced overwash creates nesting habitat for shorebirds like piping plovers ( Charadrius melodus ). We examined how piping plover habitat extent and location changed on barrier islands in New York, New Jersey, and Virginia after Hurricane Sandy made landfall following the 2012 breeding season. We modeled nesting habitat using a nest presence/absence dataset that included characterizations of coastal morphology and vegetation. Using a Bayesian network, we predicted nesting habitat for each study site for the years 2010/2011, 2012, and 2014/2015 based on remotely sensed spatial datasets (e.g., lidar, orthophotos). We found that Hurricane Sandy increased piping plover habitat by 9 to 300% at 4 of 5 study sites but that one site saw a decrease in habitat by 27%. The amount, location, and longevity of new habitat appeared to be influenced by the level of human development at each site. At three of the five sites, the amount of habitat created and the time new habitat persisted were inversely related to the amount of development. Furthermore, the proportion of new habitat created in high-quality overwash was inversely related to the level of development on study areas, from 17% of all new habitat in overwash at one of the most densely developed sites to 80% of all new habitat at an undeveloped site. We also show that piping plovers exploited new habitat after the storm, with 14–57% of all nests located in newly created habitat in the 2013 breeding season. Our results quantify the importance of storms in creating and maintaining coastal habitats for beach-nesting species like piping plovers, and these results suggest a negative correlation between human development and beneficial ecological impacts of these natural disturbances.

Piping plovers demonstrate regional differences in nesting habitat selection patterns along the U.S. Atlantic coast

“…On Long Island, New York, Piping Plovers nested and reared broods on all 1‐km beach segments with ephemeral pools or bay tidal flats but nested in less than half of segments without those foraging habitats (Elias et al 2000). Cohen et al (2009) documented rapid Piping Plover colonization of a Long Island site after a series of storms created new nesting habitat adjacent to bay‐side intertidal flats, while others (Maslo et al 2019, Zeigler et al 2019 a ) observed rapid colonization in washovers by nesting piping plovers throughout the Mid‐Atlantic region following Hurricane Sandy in 2012. In New Jersey, Piping Plover nests were split evenly between sites with and without non‐ocean foraging habitats, but birds initiated nests in washovers if those features were present (Maslo et al 2011).…”

“…This could be a major factor explaining the 1.40 breeding pairs of piping plovers per mile of sandy ocean beach in New England in 2015, compared with 0.81 and 0.75 breeding pairs per mile, respectively, in New York–New Jersey and the Southern recovery units (Appendix : Table S3). Furthermore, variable trends in the number of breeding pairs in these two recovery units (USFWS 2019) appear to respond strongly to availability of washover and accessibility of nest sites to MOSH (Schupp et al 2013, Zeigler et al 2019 a ). Cohen et al (2009) found that peak density of Piping Plovers in the portion of their Long Island, New York study area where chicks had access to both ocean‐ and bay‐side foraging habitat was more than double the density of Piping Plovers in adjacent habitat without accessible MOSH.…”

“…We assumed that a resource unit did not have access to MOSH if movement barriers completely blocked all paths between the resource unit and MOSH. See Appendix : Table S1 and Zeigler et al (2019 a , b ) for additional information.…”

“…We associated resource units in situ with the geomorphic setting, substrate type, vegetation type, and vegetation density that best described the 5 × 5 m area comprising the used or available resource unit (detailed variable definitions in Zeigler et al 2017Zeigler et al , 2019a; see also Appendix S1: Table S1). We associated all resource units with the federally recognized recovery unit within which they were located: (1) New England (Maine through Connecticut), (2) New York-New Jersey, and (3) Southern (Delaware through North Carolina; Fig.…”

EcoQBNs: First Application of Ecological Modeling with Quantum Bayesian Networks

“…BNs are directed acyclic graphs which link variables by probabilities based on first-order Markov relations. The uses of BNs are many, including the analysis of the impacts of storm events on coastal environments [ 1 ], the assessment of ecosystem services and avalanche protection [ 2 ], the applications of informing decision-making under uncertainty [ 3 ], and much more. The advantages of BNs over traditional, frequentist statistical approaches are in their flexibility to handle multiple kinds of data, capacity to integrate expert-elicited knowledge, robustness to missing data, availability of machine-learning algorithms for the structuring and parameterizing of the models, and especially in their explicit representation of uncertainty and its propagation in calculations of probability outcomes [ 4 , 5 , 6 ].…”