A Low Flow Conundrum - Part 2

So, building on last week's post...I'm currently dealing with a swampy area in southeastern Virginia - this is the first time I've modeled swampy/marshy areas. I took advantage of HSPF's high water table routines - new in version 12.2 of the model - to simulate these areas. Arriving at parameter values was an interesting experience, perhaps something to be discussed in a future post...a student working in the TMDL group at Virginia Tech is delving further into the sensitivity of these high water table parameters for his master's research.

One of the first oddities that struck me occurred when I generated the function tables for these watersheds. Ever since earlier research in the group demonstrated that the function table, as long as it is somewhat sound, has little effect on the overall hydrology predicted by HSPF, we have tended toward using an automated method to generate function tables based on the Natural Resources Conservation Service's hydraulic geometry curves and Digital Elevation Model (NED) information. It is preferable to gather one cross-section per modeled subwatershed, but in cases like the current one, where we have 78 subwatersheds to study, it becomes extremely costly to collect so many profiles.

So, moving forward with the NRCS data for the coastal plain region in Virginia, I noticed that the combination of bankfull depth, top width, and cross-sectional area did not yield a typical trapezoidal cross-section. Normally I use these three estimates to come up with a bottom width for the channel by assuming a trapezoidal channel geometry, but the calculations in this case yield a bottom width slightly larger than the top width. This didn't initially raise any flags for me, I made a mental note of the oddity and simply set the bottom width equal to the top width and moved on.

Unfortunately the studied streams did not have hydrology gauges, so I was unable to compare modeled hydrology with anything observed. We used a 'surrogate watershed' (that already had a TMDL completed) for which the function tables were calculated by another consulting firm (the methodology they used is not evident from the files they provided). During water quality calibration, I noticed that the streams went dry - a lot. This made no logical sense as we know the area we're studying is swampy. Further investigation showed that the free water surface evaporation from the reaches, nothing I had ever given much thought to before, was exceedingly high for the model of these watersheds. I traced the reason back to the difficulty in calculating bottom width - normally the bottom width is considerably smaller than the top width, so that while the flow is in the range of dry stream to bankfull (where it commonly stays), the surface area of the stream decreases as the water level falls, and evaporation decreases accordingly. Because I had set the bottom width equal to the top width for these swampy areas, evaporation continued at a high rate down to the last drop of water, causing the streams to go dry much faster than they should.

To solve this problem, I investigated the function tables from the surrogate watershed and adjusted ours to match their overall pattern. This involved a decrease in the surface area at near-zero flows - which makes logical sense, as when the flow is very small the water will start to move in small streams rather than spreading out across the full flat streambed. This solved the problem for 3 of the 4 study areas. In the fourth, however, it actually caused more problems. This goes back to what I mentioned previously about dealing with low flow issues - that is, setting a cutoff. When evaporation was high, the stream spent a considerable fraction of its time beneath the cutoff stages used for livestock and wildlife. That is, their contributions were removed from the stream a considerable amount of the time. When evaporation was set at a more reasonable level, the stream spent much more time above the cutoff, causing higher contributions from livestock and wildlife and thus increasing the various statistics we use to evaluate water quality calibrations.

This is a very interesting conundrum. Typically increasing flow (done in this case by decreasing evaporation) causes a decrease in bacteria concentrations (the old axiom "the solution to pollution is dilution" - outdated as we know it to be - comes to mind). This is the first time I've seen it actually INCREASE bacteria concentrations - and it is of course due to the way we use the stage cutoff to represent behavioral changes in animals.

I have used the neighboring watersheds as guides to help me set some reasonable parameters for this troublesome watershed. I am finishing up the modeling now and we'll see how well things go!