Predicting Silicate Weathering Rates across the Continental United States

Silicate weathering provides a stabilizing negative feedback in the global carbon cycle and it is thought to have played a key role in keeping surface environments clement for the majority of Earth's history. Nevertheless, there is still intense debate concerning the factors that regulate silicate chemical weathering rates and large uncertainties still exist in the quantitative estimation of silicate weathering rates. Well-controlled lab experiments have provided insights into the kinetics of weathering reactions, but there has been endless debate about how to mesh lab and field data.

Although various numerical models have been built to calculate silicate weathering rates with the help of field data, these models suffer from two main shortages. First, only limited parameters (e.g., runoff and temperature) that might influence silicate weathering are considered. Second, when more parameters are taken into account, only relatively simple parameterizations are constructed to reproduce the complicated reaction rate of silicate weathering. Such traditional approaches do not utilize the advances in data science. In this study, we use a suite of machine learning approaches (a subset of artificial intelligence) to predict silicate weathering rates and delineate the impacts of different factors on silicate weathering rates. There are vast amounts of data (e.g., temperature, runoff, precipitation, lithology, vegetation and soil properties) from across the continental United States which we compiled to train our model. Using the relevant parameters, our machine learning model could predict silicate weathering rates in different watersheds with low errors, relative to existing reactive transport models. We find that runoff, catchment slope, primary productivity, and temperature are the most significant parameters controlling silicate weathering rates. Our results could be used with global biogeochemical models and GCMs to provide a more realistic and refined view of the long-term response of the whole Earth system during intervals of climate perturbation.