Understanding bicarbonate concentrations across the continental US using machine learning models

Streamwater transport of carbon is an integral part of the carbon cycle, and is influenced by a variety of geological, hydrological, soil and climate factors. This study focuses on identifying the most influential factors that regulate concentration levels and export patterns of bicarbonate, a dominant inorganic carbon species in natural waters at pH around 6.0-9.0. We used the recently developed CAMELS-Chem dataset which encompasses atmospheric and stream water chemistry from 493 headwater catchments. Of this data, 170 sites had sufficient pairs (> 6) of bicarbonate and discharge concentration. We train the data using the XGBOOST package in R, a machine learning approach employing decision trees and gradient boosting to examine feature importance. The model output shows that the geological composition underlying a watershed is a primary control on bicarbonate concentrations. The peak bicarbonate concentrations occur in areas dominated by carbonate rocks. These rocks have fractures that increase porosity and permeability, expediting weathering and groundwater flow. Bicarbonate also exhibits dilution patterns with decreasing concentrations in response to high discharge, except in a few (~10) sites that express chemostatic patterns with small concentration variations. This indicates that for most watersheds, bicarbonate concentrations are higher under low flow conditions when base flow (or old groundwater) dominates in steams. The data and model also show that bicarbonate concentrations unexpectedly negatively correlate with forest fractions, leaf area indices and green vegetation content. Typically, larger vegetation cover is associated with higher carbon levels in shallow soil water, as increased photosynthesis and leaf litter in soils enhance soil respiration and increase carbon dioxide concentrations. This may imply that vegetation is more closely tied with organic carbon concentrations while bicarbonate is more reflective of geological conditions.