The Missouri River and Climate Change

Back in August, the USGS released a study on the "Trends in Annual, Seasonal, and Monthly Streamflow Characteristics at 227 Streamgages in the Missouri River Watershed, Water Years 1960-2011."  The Los Angeles Times wrote a story about the report.

The headline of the LA Times piece is "Climate change reflected in altered Missouri River flow, report says."

Sounds fascinating, given how this is what the USGS report itself said specifically about climate change and any trends the study found for flows in the Missouri River watershed:

This study did not examine forcing factors that may explain the observed streamflow trends, such as climate change, climate variability, land- and water-use changes, or groundwater pumping.

Seems to be a disconnect between what the source report examined and concluded and what appears in this LA Times article.

But even one of the authors of the report itself played up the climate change impacts when quoted by the LA Times:

'What is apparent is that the climate is changing and that it is being reflected in the stream flow conditions,' said Mark Anderson, director of the USGS Water Science Center in South Dakota and another of the report's authors.

Undoubtedly variations in climate (along with many other factors) over time within the study area will affect, and have affected, streamflow – but it should have been front and center in the LA Times article that, once again, "this study did not examine forcing factors that may explain the observed streamflow trends, such as climate change, climate variability, land- and water-use changes, or groundwater pumping."  To omit emphasizing this critical fact, and instead focus almost entirely on climate change in a study that did not even examine climate change, is a textbook example of poor science journalism.

This claim from the LA Times certainly attracted my attention:

Dennis Todey, a South Dakota University climatologist, points to precipitation changes.

'If you look at Wyoming and Montana, you see a decline in precipitation, but farther east into the Dakotas, there is a well-documented increase in precipitation,' Todey said. 'More water, more opportunities for flooding.'

Just recently, I discussed how there is not a significant trend in South Dakota's precipitation.  There are also no significant trends in North Dakota's annual precipitation since records began in 1895, nor since 1970 or over the last three decades.  Also no significant trend during the 1960-2011 period that covers the USGS study in question.

Here is Montana's annual precipitation since 1895. See a declining trend?

Not even the remotest hint of a significant trend since 1895, nor since 1970, nor since 1984.  Actually, over the past 30 years, the correlation is toward more precipitation – but the trend is highly non-significant (aka no trend).  Almost a perfect non-correlation between 1960 and 2011.

Same goes in Wyoming: no significant trend (near perfect non-correlation) since 1895, nor since 1970 or 1984.  As for the 1960-2011 period, this yields another near perfect non-correlation, with a slight positive slope toward more precipitation.

Overall, I simply cannot reproduce the claims made in the LA Times article regarding precipitation trends in the Dakotas, Montana, and Wyoming.

It is important to note that the USGS study relaxed the typical requirements for determining whether or not a trend is significant:

A trend was considered statistically significant for a probability value less than or equal to 0.10 that the Kendall's tau value equals zero. Significant trends in annual streamflow were indicated for 101 out of a total of 227 streamgages ...

Kendall's tau nonparametric test was used to determine statistical significance of trends in annual, seasonal, and monthly streamflow. A trend was considered statistically significant for a probability value (p-value) less than or equal to 0.10 that the Kendall's tau value equals zero. Although a more restrictive p-value would reduce the Type I error, the spatial distribution of the observed trends is under-scored with the relaxed p-value and is a reasonable compromise for a natural system. A Type I error occurs when the null hypothesis that Kendall's tau equals zero is rejected when true, or, otherwise stated, accepting a trend when, in fact, no trend exists. Relaxation of the Type I error probability also reduces the probability of a Type II error. A Type II error occurs when the null hypothesis that Kendall's tau equals zero is accepted when not true, or, otherwise stated, rejecting a trend when, in fact, a trend exists.

While this is fine, we must also note that the scientific convention on p-values is generally 0.05, not 0.10, and in some fields the significance cut-off can be 0.01.  I counted 19 of the 101 streamgages in the USGS study that had trend p-values of greater than 0.05 and less than or equal to 0.10.  Thus, perhaps it is more accurate to state that 82 of the 227 streamgages had significant trends, while another 19 streamgages had near-significant trends.  We should emphasize that almost two thirds of the streamgages exhibited no significant trend with the conventional criteria of a p-value less than or equal to 0.05.  That is a dominance of no net change – although this boring conclusion, of course, was left out of the alarmist media reporting.

Caution must be exercised in continuing "p-value creep," where we relax our standards for statistical significance to such an extent that too much poor science gets reported as fact.  High standards (i.e., p-values of 0.05 or 0.01) for achieving statistical significance exist for a reason – they weed out junk science that distorts a field of research and leads it down erroneous avenues into conclusions that are simply wrong and difficult to correct.  In general, natural science is skeptical and conservative.  If it weren't, it would start to approach whims, fancies, and largely unsubstantiated beliefs – otherwise known as the social sciences and humanities.

One of the key findings of the USGS report is as follows:

The mean annual streamflow contribution from the Missouri River watershed to the Mississippi River watershed was 89,400 cubic feet per second at streamgage 06934500. This streamgage had no significant trends in annual streamflow.

In other words, despite there being some stations in the watershed whose annual flows increased over time, and other stations whose flows decreased over time, overall, there was no significant trend in the water yield from the Missouri River watershed since 1960.

The question as to why the USGS study would start in 1960 is a valid one.  There is no apparent scientific explanation for a 1960 start date.  Why not also examine the post-1970 period and the last three decades – both periods over which climate scientists tell us the impacts of climate change should be most evident?

Then there is the most pressing question over how many of the stations that the USGS reported as having significant streamflow trends since 1960 don't have significant trends if we look at more recent decades.

For example, take this quote from the USGS report – which is one of only a very small number of times in the report the authors even speculate about possible climate change impacts:

A comprehensive analysis of cause of trends is outside the scope of this report. An increase in diversions or consumptive use of water during the study period, however, could not result in upward trends in annual streamflows over broad regions, such as WR3. All seven HCDN streamgages in WR3 have upward trends, which supports a climatological forcing for the upward trends.

The seven HCDN (Hydro-Climatic Data Network; "considered to have minimal human activities, such as artificial diversions, reservoir storage, or other anthropogenic works in the gaged watershed, and therefore are considered to more clearly express climate variability and change") streamgages in WR3 ("a mix of semiarid prairies in the west and temperate prairies in the east contributing streamflow to Lewis and Clark Lake") with upward trends in annual streamflow between 1960 and 2011 are as follows: Keya Paha River at Wewela, South Dakota [06464500]; James River near Scotland, South Dakota [06478500]; Big Sioux River near Brookings, South Dakota [06480000]; Rock River near Rock Valley, Iowa [06483500]; Big Sioux River at Akron, Iowa [06485500]; Floyd River at James, Iowa [06600500]; and the Little Sioux River at Correctionville, Iowa [06606600].

None of these stations has a significant trend in annual streamflow during the last three decades.  Yes, streamflows did increase between 1960 and the early 1980s, but then they stopped increasing, with most stations having near-perfect non-correlations over the past 30 years (some even with negative correlations toward less flow).  These details are important if one is speculating about climate change and the possible future impacts on nearby and downstream residents.  It also appears that many other hydrometric stations reported as having significant flow trend changes since 1960 also suffer from a loss – or even reversal – of trend significance in recent decades.

Two important conclusions come out of this examination: (1) the USGS report contains nowhere near the climate alarmism attributed to it by the media reports; and (2) in order to rigorously assess possible climate change impacts on streamflow in the Missouri River basin, future studies will need to look closely at what is going on in recent decades during which atmospheric greenhouse gas concentrations have increased most substantially.

Until then, we must keep focused on the fact that the vast majority of streamgage stations show no significant trends in streamflow, and the Missouri River watershed as a whole has seen no significant change in its water yield.