Multi-Year Colorado River Flow Predictions Inferred from the Statistics of Tree Ring Data.

Abstract:

The Colorado river has been in a drought for the last twenty years. The mean river flow for the past twenty years has been 12.5 million acre-feet per year (natural streamflow at Lees Ferry, Arizona). This a significant decrease relative to the flows measured since 1906, which have a mean value of 14.9 million acre-feet per year. This discrepancy has resulted in a cumulative ~50 million acre-feet mismatch between supply and demand which has been partially met by substantially drawing on the reserves in the Powell and Mead reservoirs.

How long will the drought last? For this water year we are looking at Lees Ferry flows of less than fifty percent of average. It is very difficult currently to reliably predict the river flows more than one year in advance. The mean of the downscaled CMIP5 models suggest that the Colorado river flow will increase over the next two decades back to the long term mean measured since 1906 but there is a large scatter in individual realizations.

An extensive discussion of the state of the art is presented in a publication edited by Jeff Lukas and Elizabeth Payton; Colorado River Basin Climate and Hydrology: State of the Science. Western Water Assessment, University of Colorado Boulder, 2020. This extensive review however omits a technique which we present in this talk. Gaussian inference (or constrained random fields) can be used to predict river flows for the next one to three decades. We use measurements of the power spectrum of river flows inferred from tree-ring measurements. We compute the mean and variance of these flows as well as realizations of specific flows to illustrate the method and to predict flows for the next thirty years.

This can serve as a critical tool for risk analysis and planning on precisely the time scales that are relevant to long term mangement of the river. We give a proof of concept by giving a hindcast for one of the major droughts in the historical record of the Colorado River. The predictions from this method are considerably less optimistic than the current methods being used to predict river flows on these timescales.