Determining the best functional form of reservoir operating policies to maximize the value of integrating seasonal streamflow forecasts into reservoir operations in the Colorado River Basin

Droughts are among the most severe natural disasters, but are difficult to prepare for because of their slow
onset. In absence of effective drought early warning systems and response plans, water users often
continue to pursue short-term economic gains that ultimately come at the expense of longer-term
economic and ecological sustainability. Fortunately, seasonal forecasts have shown promise for informing
adaptive water management policies that can reduce these impacts. One avenue of adaptation is through
reservoir operations, where seasonal streamflow forecasts can enable hedging of releases to favor more
frequent but less impactful water shortages over more devastating impacts down the line. This can be
achieved by optimizing reservoir operating policies that define how much water to release from a network
of reservoirs as a function of these seasonal forecasts. However, the best functional form for such policies
is an open research question.

The goal of this project is to compare alternative formulations of reservoir operating policies conditional
on seasonal climate forecasts to see which is most effective in reducing economic and ecological drought
impacts. We investigate this question in the Colorado River Basin (CRB). Commonly termed the
“lifeblood of the West,” the Colorado River provides irrigation water for over 5.5 million acres of
agricultural land, drinking water for more than 40 million people, and over 4000 MW of installed
hydropower capacity. Yet managing this system is becoming increasingly challenging due to ongoing
climatic and anthropogenic drought conditions that jeopardize water security and endanger the river’s
ecological health. To improve water security in the basin, this project aims to optimize reservoir operating
policies in Lake Mead, Lake Powell, and three upstream reservoirs by coupling a RiverWare reservoir
model of the CRB with Borg, a multi-objective optimization algorithm, to reduce the frequency and
severity of water shortages to the Upper Colorado River Basin (UCRB), Lower Colorado River Basin
(LCRB), and Mexico. In this study, we explore different ways of incorporating seasonal forecasts of the
inflows to these five reservoirs into their operating policies considering different functional forms of the
operating policies using a model-free, closed loop optimal control method called Direct Policy Search
(DPS). Specifically, we compare using logistic and Gaussian Radial Basis Functions (RBFs) for the
reservoir rules. Our work illustrates the value of integrating streamflow forecasts into reservoir operating
rules for drought management, while also providing insights into how to formulate the policies to
maximize that value.