Building a predictive model of submerged aquatic vegetation for Atlantic brant using remote sensing and in-situ sampling

Abstract

Submerged aquatic vegetation (SAV), e.g. eelgrass (Zostera marina), and macroalgae (Ulva lactuca., Enteromorpha spp., Gracilaria tikvahiae) provides critical ecosystem functions. However, because of a changing coastal environment due to shifting climates and anthropogenic land alteration, it is becoming more important to protect this ecosystem. There is a need to create a predictive remotely sensed model to assess SAV abundance and thus potentially long-term degradation. Secondarily, because SAV has many benefits to organisms including the Atlantic brant (Branta bernicula hrota), which is a specialist on these food sources, building a predictive model will aid in assessing the potential energetic carrying capacity of wintering brant. To meet this need, I used Landsat 8 imagery to create a Normalized Difference Vegetation Index (NDVI) of potential SAV between the southern coast of Long Island, New York and intertidal zones of New Jersey, USA, December 2015-February 2016. To quantify the accuracy of this index, I assessed SAV presence at two hundred and fifty-six 900 m2 sample sites including 174 predicted SAV points and 82 null points. Further at each site, I collected SAV biomass within a 1 m2 quadrat, as well as two subsamples within the 900 m2 resolution. A series of microhabitat variables were collected at each sample site to better evaluate predicted presence including water depth, water temperature, NH3 – N, turbidity, bottom type, and salinity. The NDVI correctly identified presence of eelgrass with 46% accuracy and the Ulva/Enteromorpha and Rhodophyta with 61% accuracy. Using 22 a priori general linear models and AIC model averaging, Ulva/Enteromorpha was significantly predicted by water depth and NH3-N, Eelgrass was significantly predicted by NH3-N, and Rhodophyta was best predicted by water depth. Using the top averaged models, I produced predictive maps that show the potential presence/absence of SAV species throughout the study area and extrapolated presence to available biomass. Estimates of energetic carrying capacity for both study areas were similar to mid-winter survey population counts for 2015–2016, thus further validating the accuracy of the NDVI predictive modeling.