Identifying Controlling Factors of Calcium Concentration in US Streams Using a Machine Learning Approach

Abstract:

Calcium is a weathering product and is an important water quality measure for water hardness. Base cations (Ca, Mg, Na, K) can buffer episodic acidification and maintain water pH. Existing studies on calcium mostly focus on individual catchments. It is however not well understood how stream calcium changes as a function of watershed characteristics (e.g., geology, vegetation cover) and climate forcing. This work uses a large dataset and a machine learning approach (XGBoost) to understand Ca concentration dependence on different variables at the continental scale. Here we show the main factors of controlling the mean concentration of Ca in 493 headwater, pristine catchments across the U.S. in the CAMELS-chem dataset. We show that vegetation is the most important factor, followed by climate and geology. Specifically, the higher fraction of forest canopy and the deeper root depth in the vegetation category produces a lower mean concentration of Ca in the stream water. Increasing mean precipitation dilutes the concentration of Ca. In addition, the larger fraction of carbonate rocks produces a higher Ca concentration. The model challenges the common perception that lithology is the most important control of Ca concentrations. These observations provide a direction for further mechanistic-based investigation and a way to predict and estimate Ca concentrations under different climate and watershed characteristics conditions.