Understanding the Hydrologic Spread

The models and methods scientists use to create projections of our future under climate change are all a little different. As a consequence, they frequently produce different results. To better understand these results, climate scientists like to use not just one model or method, but multiple models and methods together in what’s called an ensemble. But herein lies a problem.

Because ensembles are composed of multiple models and methods that are all a little different, ensembles produce spreads, or spectrums of results. So, how should we interpret an ensemble’s spread? As it happens, that is the subject of this blog.

Tackling the ensemble spread problem is a study published online last spring in the journal Earth’s Future. The study—led by CIRC Graduate Student and Research Scientist Oriana Chegwidden and including CIRC Researchers Bart Nijssen, David Rupp, and Philip Mote—is essentially an examination of the ensemble spread as it relates to the changing water supply of the United States Northwest under climate change.

The study is science about how science is done. For that reason, you are forewarned. Things in this post might get a bit nuanced for some readers. However, because Chegwidden and colleagues focus on hydrologic changes projected for our region, there is a real pay off for slogging through the details. Let’s start with what those hydrologic projections are currently saying about our region’s future.

Under projected future climate change, the Northwest will see less snow, and that snow will melt earlier. This climate change combo is expected to alter how water flows through our region’s rivers and streams, leading to lower summer flows and earlier, higher flows in the winter and spring.

This change to our water supply is likely to impact humans and wildlife alike. This is the big picture scientific consensus. This is where multiple ensemble spreads have led. This is also pretty much where the ensemble consensus ends. What’s lacking is a consensus on the details, specifically how those details might affect different regions of the Northwest.

This lack of consensus, you can probably imagine, poses a serious problem for our region’s water resource managers.

Be they working for fish and wildlife or residents of cities and farms, water resource managers all face the same question: what model and methods should I use when planning for future changes to the water supply?

This is where Chegwidden and colleagues’ paper comes into play.

The researchers looked at projected changes to Northwest United States for the years 2070–2099.

The study area for Chegwidden and colleagues’ paper is the Columbia River Basin. Which is to say the study area covers the most important water system in the Northwest United States, encompassing a considerable swath of the region (and a little bit of Canada), including parts of Oregon, Washington, and Idaho.

The Columbia River Basin was chosen, note the study’s authors, not only due to its importance to the Northwest, but also due to the diversity of the basin’s landscape.

If you’ve ever driven east through the Columbia River Gorge from the coast to Hood River, Oregon (or White Salmon, Washington) and beyond, you’ve probably noticed how the densely packed Douglas firs rather abruptly give way to sparsely spaced ponderosa pines. The reason is the Cascade Mountain Range.

The western slopes of the Cascade Range sees roughly 140 inches of precipitation a year, but farther east the interior of the Columbia Basin can receive less than 10 inches of precipitation a year. This is just one expression of the Columbia Basin’s hydrologic diversity. The basin’s hydrologic diversity is expressed in other ways as well.

Some sections of Columbia River basin receive most of their precipitation as rain, others as snow. The result: different sub-basins in the Columbia River Basin have drastically different hydrologic regimes that are expected to respond differently to climate change.

The Columbia River Basin’s hydrologic diversity is mentioned here because, basically, the researchers used the basin’s hydrologic diversity as a research tool, a kind of scalpel that allowed them to dissect, so to speak, the different models and methods that made up the ensemble they used in their study.

So, what did the researchers find?

To explain that, we need to talk about the ensemble Chegwidden and colleagues created for this study. First, the basics…

All future projections of climate change—sometimes called futures or projections—use numerical models run in a computer (or computers) to produce their results. These are the research tools Chegwidden and colleagues were interested in examining.

Imagine something like a large assembly line, but instead of manufacturing cars or widgets this assembly line creates projected futures, specifically projected hydrologic futures. In Chegwidden’s case these hydrologic projections are expressed as three different streamflow metrics for 396 sites along streams and rivers in the Columbia River Basin. This process starts, believe it or not, with guesswork—educated guesswork.

Climate researchers don’t know whether greenhouse gas emissions will continue at their current rate, and consequently if global warming will continue at its current rate, or if humanity will get its collective act together and cut those planet-warming emissions down to a reasonable level and limit the extent of global warming. Of course, to be fair to climate researchers, nobody knows this. However, we can make an educated guess based on the two possibilities we face: little to no action (high warming) or action (less warming). And the scientific community as a whole has generally settled on these two options.

To model future climate change, climate researchers generally use two emissions scenarios in tandem: the lower emissions scenario, RCP 4.5, which models a world in which humans do act to cut emissions so that while warming continues throughout this century it starts to level off after 2100; and the high emissions scenario, RCP 8.5, which models a world in which humanity does nothing (or next to nothing) to cut emissions and as result warming is allowed to continue at its current upward trajectory throughout this century and beyond.

(Note: RCP 4.5 isn’t the lowest emissions scenario. But it is lower than RCP 8.5. So, we use—as do most climate research outfits—the linguistically awkward lower and high, instead of lower and higher. Our apologies. We don’t like it either.)

Chegwidden and colleagues used both the lower emissions scenario (RCP 4.5) and the high emissions scenario (RCP 8.5) run through an ensemble made up of 10 global climate models (GCMs).

(We should note—for those who might care—that the researchers didn’t run the simulations themselves, but relied on data from GCM simulations, or runs, provided by the Couple Modeled Intercomparison Project, which acts as an international clearinghouse of vetted GCMs and the data resulting from their runs).

Okay. Let’s do the math. Ten GCMs runs twice for each RCP gives us 20 simulations. This is where the spread begins. It continues at the next stage in the hydrologic ensemble assembly line: downscaling.

GCMs—as their communal name suggests—model global climate, meaning the climate of the whole planet. GCMs are all about capturing the big picture. However, GCMs are a little bit like video games that lack good resolution. Try to zoom in too much from the global scale and you often find key local features of landscapes (you know little things like the Cascade mountain range) poorly rendered. To add local details to the map, climate researchers use a process called downscaling, which helps translate course GCM data into fine details at the local level.

For their study, Chegwidden and colleagues used two downscaling methods: the bias-correction spatial disaggregation (BCSD) method and the multivariate adaptive constructed analogs (MACA) method. (You don’t really need to know the names, but in case you’re interested.) If you’ve been counting, we now have a spread encompassing 40 results (2 RCPs x 10 GCMs x 2 downscaling methods).

Of course you can’t have hydrologic projections without hydrologic models. And here we come to the last stage of the assembly line. For their study, Chegwidden and colleagues used two of them:  the Precipitation Runoff Modeling System (PRMS) and the Variable Infiltration Capacity model (VIC), which was run under three different calibrations, meaning VIC basically comprised three hydrologic model choices rather than just one, making for a total of four hydrologic methods. This widened the spread to 160 results.

Here’s where things get interesting and where we can examine how the Columbia River Basin’s diversity relates to the choice of GCM, RCP, downscaling method, and hydrologic model. Let’s start with the Columbia River’s rain-driven sub-basins. To explain this, let’s look at rising temperatures. (Don’t worry, it will make sense.)

All scientifically vetted climate model projections show temperature goes up with rising CO2 emissions. However, there is some disagreement between models about the extent of this change. For instance, some GCMs show more rising temperatures given the same amount of emissions than others. Others show less given the same amount of emissions. This means you can take the same 10 GCMs, run them with the same RCP, and what you will find is that they all show warming, they just can’t agree on the extent of that warming. Precipitation, as Chegwidden and colleagues note, is another story.

Some GCMs in the researcher’s ensemble showed projected increases in annual precipitation for the late decades of this century, while others showed decreases. Logically then, if some GCMs showed an increase in annual precipitation and others did not, the choice of GCM should be an important methodological consideration that water managers in a rain-driven basin should consider when interpreting ensemble results. Put another way, if you receive most of your precipitation as rain, the disagreement among the models about whether it will likely rain or not is a disagreement you need to pay attention to.

As you’ve probably guessed, a similar pattern held for different methodological choses and other sub-basin types.

In a nutshell, Chegwidden and colleagues found that the relative importance of the methods used (i.e. which GCM, downscaling method, hydrologic model, etc.) depended on the characteristics of each sub-basin. For instance, which RCP was chosen (high or lower) was essential for basins that currently receive a large percentage of their water from high mountain snow. Whether emissions were cut or not and hence how warm things were projected to get greatly determined how much snow remained by the end of the century.

A similar pattern held for hydrologic models. The choice of hydrologic model greatly impacted streamflow projections, especially how the models created projections of low flows—which are exactly what they sound like, low flows in rivers and streams. This suggests that more attention should be paid to hydrologic model choice, especially if low flows are a concern in a sub-basin.

(Oddly enough, the downscaling methods used really didn’t factor in that much. Though the researchers point out they only used two methods. Had they used more, they note, things might have been different.)

Chegwidden and colleagues’ findings imply that while the spread in projections still remains large and varied, their study could help water managers better interpret an ensemble’s results and what it means for their sub-basin. More generally, the study shows how future modeling efforts could be focused to help us better capture the spread in climate change projections of any given area. And while that does not create a clear road path for water managers to follow, it does point the way.

Publication: Chegwidden, Oriana S., Bart Nijssen, David E. Rupp,  Jeffrey R. Arnold,  Martyn P. Clark,  Joseph J. Hamman,  Shih‐Chieh Kao,  Yixin Mao, Naoki Mizukami,  Philip W. Mote,  Ming Pan,  Erik Pytlak, and  Mu Xiao. “How Do Modeling Decisions Affect the Spread Among Hydrologic Climate Change Projections? Exploring a Large Ensemble of Simulations Across a Diversity of Hydroclimates.” Earth’s Future, 7 (2019). https://doi.org/10.1029/2018EF001047.

Featured Image: Caption: “Columbia River on the eastern side of Oregon.” Photo Credit: Bonnie Moreland, March 2018. This photo is in the public domain.


Key Findings:

  • Projected changes in hydrology can vary greatly depending on the modeling choices used.
  • The spread among projected hydrologic changes remains large for the Northwest United States.
  • However, while the spread remains large, this study points to potential ways to focus future modeling efforts to better understand the existing spread and could help water resources managers apply that understanding to their work.

Nathan Gilles is the managing editor of The Climate Circulator, and oversees CIRC’s social media accounts and website. When he’s not writing for CIRC, Nathan works as a freelance science writerOther posts by this Author. 

Screen Shot 2017-08-07 at 2.41.20 PM

Stay up to date on the latest climate science news for the Pacific Northwest, subscribe to the CIRCulator.


Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s