Author Date


Degree Name



Computer Science


Physical and Mathematical Sciences

Defense Date


Publication Date


First Faculty Advisor

Christophe Giraud-Carrier

First Faculty Reader

Ben Abbott

Honors Coordinator

Seth Holladay


Reservoir Computing, Machine Learning, Streamflow Dynamics, Hydrochemistry, Hydrological Modeling


With severe drought continuing in the western United States and the effects of climate change becoming more apparent across the world, it is becoming increasingly important to be able to predict the impact of extreme weather events like storms, droughts, and fires on streamflow dynamics. This includes flow regime as well as biogeochemical behavior of river systems and their watersheds. This project explores the use of Echo State Networks (ESN), a subset of Reservoir Computing, on modeling and predicting streamflow variability with a focus on biogeochemical patterns. In this project ESNs are tested and compared in the hope of creating more robust streamflow chemistry predictors that are applicable in broader scenarios than what are commonly needed for Machine Learning applications to Hydrological problems.

Reservoir Computing models are proven to be an effective model for multivariate time series problems like streamflow prediction, (problems with more than one time-dependent variable, where each variable depends on both its past readings, as well as its relation to other variables). ESNs have been in use since the late 1990's, but remain less well-known than more modern Deep Learning models. ESNs are an efficient Machine Learning model, and their inherent non-linearity makes them very dynamic and able to adapt to training quickly. This makes ESNs a good potential fit for large-scale environmental signal-processing and remote sensing problems. We also compare ESNs with a modern Long Short-term Memory (LSTM) model, which is frequently used for streamflow problems, and provide a template for when one model should be picked over the other.