Seasonal carbon dioxide concentrations and fluxes throughout Denmark's
stream network
Abstract
Streams are important freshwater habitats in large-scale CO2 emissions
budgets because they are generally supersaturated with dissolved CO2.
High CO2 concentrations driven by terrestrial carbon inputs, groundwater
flow, and internal respiration vary greatly across space and time. We
compiled and used environmental monitoring data to calculate CO2
concentrations along with a wide range of predictor variables and
trained machine learning models to predict spatially distributed
seasonal CO2 concentrations in Danish streams. We included outputs from
a national hydrological model to investigate the influence of
hydrological processes. We found that CO2 concentrations in streams were
supersaturated (mean = 118 µM) and higher during autumn and winter than
during spring and summer. The best model, a Random Forest model, which
scored R2 = 0.46, MAE = 46.0 µM, and ⍴ = 0.72 on a test set, predicted
seasonal CO2 concentrations for the entire stream network. The most
important predictor variables were catchment slope, seasonality,
elevation relative to the nearest stream, and depth to groundwater,
which highlights the importance of landscape morphometry and
soil-groundwater-stream connectivity. Stream CO2 fluxes, calculated by
using empirical relationships, averaged 253 mmol m-2 d-1, and the annual
emissions were 513 Gg CO2 from the national stream network (area = 139
km2). Our analysis presents a framework for modeling seasonal CO2
concentrations and estimating fluxes at a national scale by means of
large-scale hydrological model outputs. Future efforts should consider
further improving the temporal resolution, direct measurements of fluxes
and gas transfer velocities, and seasonal variation in stream surface
area.