Quantile regression illuminates the successes and shortcomings of long-term eutrophication remediation efforts in an urban river system.

Despite massive financial investment in mitigation, eutrophication remains a major water quality problem and management priority. Eutrophication science-well established for lakes-is not as well developed for rivers, and scientific understanding of how rivers respond to eutrophication management is far more limited. Long-term data are required to evaluate progress, but such datasets are relatively rare for rivers. We analyzed 23 years of water quality data for the Charles River, a major urban river system in the northeastern U.S.A., to examine nutrient and phytoplankton biomass (chl-a) responses to decades of phosphorus (P) management. Using the more novel and robust approach of quantile regression, we identified statistically and ecologically significant declines in both total phosphorus (TP) and chl-a over time, but only for middle percentiles. Statistically high concentrations of TP and chl-a persist-the segments of the data of greatest concern to managers and the public-and yet this critical result is concealed by statistical tests often employed in eutrophication studies that only evaluate mean changes. TP, temperature, precipitation, and river segment jointly explain the most chl-a variation observed at the decadal scale. Spatial variation is also considerable: despite a significant decline in TP, the impounded lower river exhibits no long-term trend in chl-a and continues to experience annual blooms of harmful cyanobacteria-a lagging response comparable to that of a recovering eutrophic lake. Despite long-term successes in reducing P, chl-a, and cyanobacteria in the Charles River system, we did not detect any significant, long-term change in the attainment of statutory compliance, illustrating the protracted and complex nature of the river's response. Our analysis demonstrates the need for high-frequency, long-term water quality data to evaluate the progress of eutrophication management in urban rivers, and the utility of quantile regression for detecting critical trends in the occurrence of statistically low-frequency but ecologically high-impact events, including blooms of harmful cyanobacteria.