PNNL

Abstract

As concerns about water availability increase due to a changing climate and extreme weather, there is a need for computational test beds to help explore water management challenges and elucidate interdependences within coupled human-natural systems. These test beds can range from process-based simulations that take advantage of domain-informed approaches for hydrologic cycling to more data-driven models that leverage the underlying statistical properties of the data. However, development of data-driven models for water availability can be inhibited by lack of access to model source code or data. For example, the Water Rights Analysis Package (WRAP) is used by the state of Texas to assess basin-level water availability, but it is closed source software and currently available data is insufficient for the development of data-driven models. We address these issues within the Colorado River Basin of Texas by: 1) simulating an ensemble of 1,000 streamflow realizations using Hidden Markov Models; 2) routing these realizations through WRAP to obtain water right-level shortage outputs; and 3) emulate WRAP’s water allocation processes by developing a long short-term memory (LSTM) neural network model. The trained LSTM emulates water shortages with high accuracy (