Taking climate change seriously: the Colorado River “stress test”

Courtesy Dave Kanzer and Eric Kuhn, from Eric’s 2013 Colorado River Water Users Association presentation


The Bureau of Reclamation Colorado River team did something remarkable in yesterday’s release of its new 5-year reservoir levels analysis – the “stress test”, a methodology pioneered a decade ago by an Upper Colorado River Basin technical team that included John Carron of Hydros and Eric Kuhn and Dave Kanzer of the Colorado River District is now the “new normal”, to borrow a terrible phrase. From the “5-year projections approach tab” here:

The method used to generate future inflows in the current projections includes resampling a subset of the historical natural flow record (1988-2019) using the Index Sequential Method (ISM), referred to here as “Stress Test” hydrology. In the past, the full historical record (1906-2019), known as the “Full” hydrology, was used to provide 5-year probabilistic projections. The Stress Tests hydrology scenario applies ISM to a shortened period of the natural flow record, 1988-2019, which removes the earlier portion of the natural flow record and focuses on the recent (approximately 30 years) hydrology. This period has a 10% drier average flow than the Full hydrology. Use of the Stress Test scenario 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 (McCabe et al. 2017Udall and Overpeck 2017Woodhouse et al. 2016).

The idea is that the traditional approach – using the entire period of record to model the probabilities of future river flows – is not longer valid because climate change is changing the river.

John, Eric, and Dave reasoned nearly a decade ago that using a shorter record, focused on our climate-changed Colorado, might better help managers think about and plan for what to expect next. (Dave also famously provided the memorable Homer Simpson image for Eric’s CRWUA presentation).

The “stress test” has been creeping into basin management discourse for a while, and Reclamation had already begun publishing stress test scenarios alongside. But the new 5-year flow and reservoir level estimates now are all in on the stress test.

The stress test may not be stressful enough, which was one of the implicit messages in the editorial Brad Udall and I published in Science magazine in May, and which Brad and I made more explicit here. But this use of the stress test is nevertheless hugely important, kudos to the Reclamation technical and management team for this important step.


  1. Maybe they should use estimates based in part on the Medieval Warm Period or some of those megadroughts of the last two millenia. If climate projections are considered somewhat valid, this would be a “worst case” estimate. IIRC, the eighties was a wet period and could that result in a more optimistic flow than is warranted?

  2. ISM is probably a good approach, since wet/dry periods tend to clump.

    Would it make sense to superimpose these sequences onto Udall/Overpeck’s “hot drought” projections? Same sequences, but with 20% less flow by mid-century, would be real instructive.

  3. Khalil J Spencer-

    Reclamation has used the Meko et al. (2007) tree-ring estimates of Colorado River flow, which cover all of the Medieval Warm Period (~900-1300), as a hydrology for long-range (~50-year) modeling in the 2007 Powell-Mead EIS and the 2012 Basin Study, see slides 10, 11, 21: https://www.usbr.gov/lc/region/programs/crbstudy/WebinarJan2013.pdf

    I don’t think these paleo-hydrologies have ever been used by Reclamation for the 5-year modeling but there are no technical barriers to doing so. It’s a good idea.

    Tres English-

    Agree that ISM (Index Sequential Method) is useful in that it preserves the historical wet-dry sequences. The main limitation is that the number of independent samples is very low. In the 32-year Stress Test hydrology, there are only *6* independent 5-year samples; the other 26 samples overlap with those. A stochastic generation method using the 13.3 maf mean of the Stress Test, and the wet-dry transition probabilities from the paleo, would probably be better. This would be similar to the “paleo-conditioned” long-term hydrology used in the Basin Study.

    And yes, it would make a lot of sense to blend historical sequences with a % flow reduction consistent with future warming scenarios. A Reclamation-commissioned study, not yet released, did something very much like that. The Basin Study, alternatively, used GCM-derived sequences directly as long-term planning hydrologies, without blending them with historical sequences–see the same 3 slides referenced above.

  4. Khalil J Spencer-

    The last year of the 1980s wet spell was 1987, and the Stress Test hydrology begins with the very dry 1988-1992 sequence, which as a 5-year drought was only second to 2000-2004 in the Lees Ferry natural flow record. But 1993-1999 was very wet again, so that 1988-1999 averaged 14.2 maf, vs. 12.6 maf for the 2000-2019 period that followed. Thus John’s and Brad Udall’s call to use only the period from 2000 onward as a “true” stress test.

  5. Jeff…
    I attended your workshop at Boulder City BuRec’s Regional Office years back. It was probably one of the best presentations that I got to attend while attached to BCOO. It made me think in the timeframe of decades verses the ‘short term’ when it came to data collection and perceptions that people have about climate change. The irony being using Tree Ring studies to correlate the climate events of the past. REAL SCIENCE filling in the blanks. For me, englightning. Thanks.

  6. Dave Gunderson-

    Thanks for the kind words, Dave. I remember the 2008 Boulder City tree-ring workshop well; it was my first and only time at the LC Regional Office. We still have info from that workshop archived on the TreeFlow website: https://www.treeflow.info/content/nevada2008

    One very relevant aspect of that workshop was Jim Prairie’s introduction of a novel methodology to blend the tree-ring record’s more complete expression of natural variability in precipitation and streamflows (i.e., Medieval megadroughts) with the likely long-term decreases in streamflow from human-caused warming, as seen in climate model projections. That methodology nicely captures how we need to look both farther back and farther ahead to fully assess long-term system risk for planning purposes.

Comments are closed.