Closed rrknight-usgs closed 6 years ago
No full answer here, but consider again the trends in FDC using Kendall Tau that we have functional. Even with say an alpha=0.05 and perhaps finding all sites with like 50 years of data, if we have no trend on any ordinate in the FDC, does that me we can securely have a reference site?
:thinking:
These comments stem from the Kroll et al (2015) paper on quantifying hydrological alteration using FDCs. 💭 :thinking:
Kroll presents some interesting ideas on ways to use FDC comparisons to get at when a hydrograph is 'altered' or at least different than the comparison. Specifically, they demonstrate 3 approaches to defining the unaltered FDC (median of annual FDC quantiles, and then jackknifed mean and median FDCs using 5 yr samples from a 20 yr block - first 20 yrs of record assumed unaltered). They then define the test criteria (rejection region) by comparing samples of the FDC w/i the 20 yr block using 4 tests (K-S, Kuiper, CIs, and 'eco') for quantifying difference from the reference(...) effectively saying that anything outside what was seen in that re-sample would be considered rejection of the null (unaltered). I'm butchering this... but that is the idea. They then repeat the process (sort of) using the 'altered' period... test for power and so on. In short, they determine that defining the reference FDC is best done by mean annual Jackknifed method and the best (most powerful (1-beta)) test for alteration is using the CI (conf int). 😩 :face_with_head_bandage: 😕 :nerd_face:
In short... I like their approach. It makes sense to me, and jives well with things that we are already, for the most part, calculating / estimating / generating. Where they stop (b/c you can't do everything), is that they only test it against sites where there is significant regulation / alteration of the flow regime, using 117 basins in which there is a known POR w/o signif alteration, then a reservoir was dropped in... so easy to detect differences.
Questions ❓ / thoughts (just thinking out loud):