Communicating Water Risks

The Colorado River has been dropping rapidly over recent months, leaving mudflats on its margins. Photo by Brian Richter

___________________________

In early January, I wrote an OpEd for the Denver Post newspaper warning of dire water shortages in the Colorado River Basin should we experience a recurrence of some of the bad runoff years we’ve seen in the recent past, such as in 2001-2004, or in 2012-2013.

I was quite surprised and dismayed at many of the reactions my editorial elicited. I was labeled as being an alarmist, as suggesting an unrealistic doomsday scenario, as playing Chicken Little.

I am sorry to say that my dire warnings appear to be coming to fruition. Lake Powell has dropped another 50 feet this year and its water level is expected to be at elevation 3,537 by year’s end. If the reservoir drops that much again next year (and multiple signs appear to be pointing in that direction, including widespread soil moisture deficits and persistent La Nina conditions), hydropower production from Glen Canyon Dam will cease, sending an electricity shockwave across the Southwestern US.

Already this year, all of the boat marinas on Lake Powell have closed due to low lake levels.

I was motivated to write the Denver Post editorial not because I’m an alarmist nor a pessimist, but because as a global water professional I have witnessed firsthand what can happen if water risks are not managed sufficiently. I felt morally obligated to communicate what was not being said nine months ago, i.e., that the risk of serious water scarcity was growing very rapidly and could quickly lead to a crisis situation. I’ve been advising governments, corporations, cities, and conservation organizations on water risk management for more than 30 years, and I felt compelled to sound an alarm in the hope that appropriate actions could be undertaken in a timely fashion to manage those risks.

Big Kudos to the US Bureau of Reclamation

Last week, the US Bureau of Reclamation made some very important changes in their procedures for communicating water risk in the Colorado River Basin. With these changes, I’m hoping I won’t feel compelled to be a bearer of bad news in the future because the Bureau will now be doing a much better job of analyzing and communicating risks in their own official reports.

The Bureau regularly updates and publishes two types of water risk forecasts – a 24-Month and a 5-Year forecast of projected reservoir levels throughout the Colorado River Basin, which are widely read by water managers and large water users. These forecasts are based on what has happened in the past, i.e., how would water levels in storage reservoirs change if we were to experience a sequence of years similar to what we’ve seen in the past?

Acknowledging Climate Change

When performing these forecasts of future reservoir levels, the Bureau of Reclamation uses a series of annual runoff measurements from past years, and then scrambles (‘resamples’) those annual flows to create new (potential) sequences of years to evaluate what might happen in the future. When conducting such a modeling exercise, it’s usually beneficial to use as many historical years as possible, so that a rich mixture of possible future sequences can be evaluated.

However, we are now learning that the past may not be a reliable predictor of the future, particularly when it comes to predicting river runoff that is being increasingly influenced by climate change. Importantly, the Bureau has decided to use a much shorter, and more recent, series of historical observations as the basis for their projections. In the past, their randomized modeling utilized a long period of 114 years (1906-2019) of historic measurements. Recognizing that earlier years may no longer represent the expected climate going forward, the Bureau is now using only the most recent 32 years (1988-2019), with the hope that this more recent period will be a more reliable predictor of future conditions.

In explaining their change of approach, the Bureau states, “Use of the (shorter time period) is supported by multiple research studies that identified a shifting temperature trend in the Colorado River Basin in the late 1980s that affected runoff efficiency and resulted in lower average flows for the same amount of precipitation.”

Of note, river flows in this recent period have been 10% drier than the longer-term past. This acknowledgment by the Bureau that the climate has changed – and their forecasts need to be adjusted to reflect that new reality – is a giant step forward in hydrologic forecasting for the Colorado River Basin.

However, the Bureau made another change in their methods that is just as important as shortening their historical reference period.

Improved Transparency About Worst-Case Scenarios

On numerous occasions during the past couple of years, I have expressed my concern with Bureau staff that they weren’t transparently communicating the true worst-case scenarios in their forecasts.

In the past, three basic outcomes have been highlighted in the Bureau’s forecasts: the “most probable” outcome (i.e., the most likely elevation of a particular reservoir such as Lake Powell at some time in the future), the “minimum probable,” and the “maximum probable.”

I’ve realized that even among the water professionals working in the Colorado River Basin, few really understand what these terms mean.

As described in the technical details of the Bureau forecasts, the “most probable” represents the 50^th percentile of model results, i.e., after running randomized sequences in their modeling of the future, they would take the middle value (median or 50^th percentile) to be the “most probable.” There is a fundamental problem with the Bureau’s use of this term, which I’ll touch on in a moment, but my bigger concern was their use of the term “minimum probable.”

You see, the Bureau’s “minimum probable” was actually NOT the worst-case scenario, as the term would imply to most readers.  Instead, it was their 10^th percentile model projection, meaning that in 10% of all modeled years, worse outcomes could be expected. But those worst-case scenarios were not being reported anywhere in the Bureau’s forecast reports.

My OpEd was intended to sound a warning about those scenarios with less than a 10% chance of occurring. I had been closely tracking reservoir levels and doing my own back-of-envelope modeling, and by early January of this year I could see that if the hydrologic system continued to produce very dry (<10^th percentile) runoff this year and next, disastrous consequences could emerge very quickly.

The graphs below reflect the two important changes in the Bureau’s reporting. This new forecast is based on the shorter 32-year historical period AND it includes the true minimum and maximum potential outcomes.

In these graphs you can see the difference in the apparent ‘worst case scenario’ when looking only at the spread between the 10th and 90th percentiles (dark yellow coloring) as compared with the true minimum, the true worst case scenario (bottom of the lighter yellow spread).

Those truer portrayals of the minimum forecast are scary as hell. If this worst case scenario were to unfold, Lake Powell would hit functional dead pool (i.e., no more water can be released into the Grand Canyon) by 2024, and Lake Mead would hit dead pool by 2026.

I’m praying that everyone involved in managing or using the river’s water — as well as those charged with protecting endangered species and other ecosystem conditions dependent on this river — will give due attention to the bottom (the minimum possible level) of these curves.

One More Change, Please!

There is one more important change that I hope the Bureau will make in its reporting of water risks.  Labelling the 50th percentile outcome (see dark yellow line in graphs above) as being the “most probable” scenario can be misleading to most readers of these forecast reports. While an argument could be made that referring to the central tendency (50th percentile) as “most probable” might be appropriate in a statistical sense, we need to be careful about how the general public and the media interpret this term.

When something is labelled as being “most probable,” that’s what most readers will expect to happen. Given how rapidly the climate is changing, causing shifts in the distribution of runoff probabilities, I suggest that we drop all references to “probable” — whether it’s most probable, minimum probable, or maximum probable — and simply report the possible outcomes as percentiles. For instance, change the language to say, “there’s a 50 percent chance that the future outcome will be higher or lower than this.” Or “there’s a 10% chance that the outcome could be less than this.”

My sincere thanks to the Bureau of Reclamation staff that has implemented important changes in their forecasts.