Since the 1940s, anthropogenic nitrogen (N) inputs have grown to dominate global N cycles, particularly in fluvial systems. Negative impacts of this enrichment on downstream estuaries are well documented. Efforts at N reductions are increasingly successful but evaluating ecosystem response trajectories is difficult because of a lack of knowledge of historic conditions. To document continental-scale coastal food web N-dynamics prior to large increases in human N-loads, we sampled 208 fish from an archival collection, taken from coastal waters across the continental U.S., with a median collection year of 1904. The archival fish were compared with 526 samples collected in 2015 from 126 estuaries also along the U.S. coastline. We used stable isotopes of N (δ15N) and carbon (δ13C) as a proxy for human inputs and organic matter sources. Watershed attributes from 1910 and 2012, census data, fish life histories, and basic estuarine geography were used to develop random forest models that determined which variables were the best predictors of isotope values. State, latitude, and fish trophic level were consistently the most important predictors, while human impacts played a lesser role. When the fish were collected (~1914 vs 2015) was not an important predictor, rather where the fish was collected was the best predictor of N source. The model results illustrate the important role that geography plays in coastal food web dynamics and underscore the importance of offshore N-sources to coastal food webs.
Keywords: Anthropogenic; Carbon; Estuaries; Fish; Machine learning; Nitrogen; Random forest; Stable isotope; United States.