Salt marsh plant species have long been known to be divided into ‘zones’ based on the interactions of physical and biotic drivers. Physical determinants of plant zonation include salinity and flooding frequency, factors both known to stress plants. Biotic determinants are less straightforward, as previous studies have shown evidence of both competitive and facilitative interactions between marsh plants. Herein lies the challenge for marsh ecologists, as conducting field experiments to disentangle and quantify these effects can be difficult. However, by pairing analysis of marsh photographs with measurements of salinity and elevation, we can assess drivers of species zonation through modeling techniques. Photographic data collection offers advantages in terms of decreasing time spent in the field while also generating large amounts of data, allowing for stronger statistical analyses. 10,618 photographs and 139 measurements of salinity and elevation were taken at Dean Creek marsh on Sapelo Island, off the Georgia coast in 2014. 721 photographs were manually annotated to generate presence/absence data for six marsh plant species. Data was analyzed with the goal of identifying strong predictors for each species, and comparing these predictors to known determinants of zonation for well studied species such as Spartina alterniflora.
Results/Conclusions
Results from analysis of preliminary annotation data showed a negative relationship between Spartina and elevation, a result consistent with the species’ known preference for low marsh habitat. Environmental factors were strong drivers for other species as well, with increases in Batis, Sarcocornia, and Borrichia abundance as salinity and elevation increase. Interspecific interactions appear to play a role also, with distinct communities forming in the high and mid marsh. Accounting for elevation and salinity, Borrichia and Batis exhibit positive effects on one another which could be indicative of facilitative interaction, but more thorough examination of the relationship is required. These results, while preliminary, show promise for the predictive capability of models built on photographic data. The eventual goal of this project is to use deep learning image classifiers and modeling techniques to quickly and effectively characterize salt marsh plant communities and their drivers. Image classifiers can annotate thousands of images much faster than human eyes, and with comparable accuracy. Modeling can account for biotic and abiotic drivers in the marsh, and quantify their contributions to individual species distributions. Used together, these methods could be scaled up for use across entire marshes, and could be an invaluable management tool as salt marshes face an uncertain future.