Identifying Controls of Climate and Watershed Characteristics on Concentrations of Dissolved Silica (DSi)

Dissolved Silica (DSi) originates from chemical weathering of Si-containing minerals and is an important water quality measure in both terrestrial and oceanic water systems. DSi is an essential nutrient for the phytoplankton groups that create about half of the atmosphere's oxygen and make up the base of virtually every oceanic food web. Studies have shown that fluctuations in DSi concentration directly affect the nutrient composition of water bodies, but it is unknown how much of an effect increasing climactic and anthropogenic perturbation will have on their aqueous concentrations on land. This work aims to identify relationships between mean DSi concentrations and watershed hydroclimatic, geology and land cover predictors in 299 sites in the Catchment Attributes and Meteorology for Large-Sample Studies (CAMELS)-chem data set. By measuring the correlation coefficient (R) between DSi concentration and the remaining predictors, we find that DSi concentration is most dependent on mean daily precipitation, leaf area index difference and watershed area, suggesting that climate and vegetation can both influence DSi concentrations. Using the XGBoost machine learning algorithm, we are able to construct a model that accurately reflects 55.7% of data used for testing. The most influential predictors for this model were not previously recognized as having a strong correlation with DSi concentration. However, the relative importance of individual predictors is very low, suggesting that this model may need further refinement to identify the most influential factors. We suspect that highly correlated predictors play a large role in our unsatisfactory results. In a future work, we hope to remove these types of predictors that largely contribute redundant information leading to improved model performance.