Declining Predictions of Net Ecosystem Production in US Rivers and Streams Throughout the 21st Century

Metabolism is an essential component of carbon cycling in river ecosystems, and understanding its response to climate change on a broad scale is imperative. Here we employ deep-learning models trained on an extensive data set to reconstruct daily metabolism in a total of 293 rivers and streams across the continental US from 1980 to 2020. Three key variables, gross primary production (GPP), ecosystem respiration (ER), and net ecosystem production (NEP), are examined to unveil long-term trends. Our analysis reveals that continental US rivers and streams experience an increase of 0.045 g O₂ m^-2 day^-1 decade^-1 in GPP from 1980 to 2020, largely driven by alterations in runoff and insolation, while ER declines more strongly at a rate of 0.078 g O₂ m^-2 day^-1 decade^-1, primarily attributed to the combined effects of discharge, thermal conditions, and temperature changes. Such changes have caused a slight decrease in the NEP over the past four decades. Moreover, our well-trained models project that NEP continues to decline at a rate of 0.017 ± 0.008 g O₂ m^-2 day^-1 decade^-1 under future climate scenarios, resulting from asymmetric and converse trends between GPP and ER. Such persistent net heterotrophy shifts would threaten aquatic biodiversity and weaken ecological resilience of flowing waters to climate change.