....

Sure, but that's a very big topic. How to narrow it down? As I've been with the USGS now for 15 years, you'll get very much a USGS-tinged view of things, which is focused on at-site flood frequency and regional flood frequency (under stationary conditions).

For at-site, as you may know, the general US federal government recommendation is to fit the log-Pearson type III distribution. Along those lines, the long-awaited update to Bulletin17B, called Bulletin17C, has been out for public comment, and has been revised in response to those comments. See:

http://acwi.gov/hydrology/Frequency/b17c/index.html

One big innovation in Bulletin 17C is "Expected Moments Analysis" (EMA) which allows moments for historic or other threshold measurement-based information to be included:

Cohn, T.A., Lane, W.L., and Baier, W.G., 1997, An algorithm for computing moments-based flood quantile estimates when historical flood information is available: Water Resources Research, v. 33, no. 9, p. 2089–2096.

Cohn, T.A., Lane, W.L., and Stedinger, J.R., 2001, Confidence intervals for expected moments algorithm flood quantile estimates: Water Resources Research, v. 37, no. 6, p. 1695–1706.

EMA is already implemented in PeakFQ, at:

http://water.usgs.gov/software/PeakFQ/

On regional flood frequency, the standard technology for USGS has been for ~30 years GLS regression on basin characteristics, computed one quantile at a time, going back to papers by Jery Stedinger and Gary Tasker in the 1980s, now implemented in WREG:

http://water.usgs.gov/software/WREG/

You could start with the manual for the program to see the theoretical background: http://pubs.usgs.gov/tm/tm4a8/.

A typical product of this type of study might be:

http://il.water.usgs.gov/projects/2004_flood_freq/

Bulletin 17B-based regional flood frequency work also brings you to the topic of regional skew. A former student of Jery Stedinger, Andrea Veilleux, has been working to implement a new method in the U.S. based on her Ph.D. work using Bayesian regional regression; see, e.g.,

http://pubs.usgs.gov/sir/2010/5260/

We're not totally ignorant of non-stationarity; you are probably aware of several important descriptive papers from recent years by USGS scientists such as:

http://onlinelibrary.wiley.com/doi/10.1029/2005GL024476/full ("Nature's style: naturally trendy")

and

http://science.sciencemag.org/content/319/5863/573 ("Stationarity Is Dead: Whither Water Management?"),

and we do have a new workgroup on the topic that is looking at when and how to address the problem. Since it's so new I can't predict very much about what we'll come up with.

But along the lines of non-stationarity, from a personal perspective, recently a pair of reports from a project that I have been working on for several years came out in which we used regional quantile regression to assess and adjust for the effect of urbanization on flood frequency. I think quantile regression holds a lot of promise for regional flood frequency because it allows you estimate the exceedance probability of a given flood event at a given site based on regional information, even in the presence of non-stationarity (though climatic non-stationary would require a modification of the particular approach we used), without going through the whole at-site flood frequency thing and THEN doing a regionalization analysis. These reports are at:

https://pubs.er.usgs.gov/publication/sir20165050

and

https://pubs.er.usgs.gov/publication/sir20165049

One other personal perspective is that I am still interested in scaling of floods (and streamflow statistics in general), a la my Ph.D. advisor, Vijay Gupta, (see, e.g, http://onlinelibrary.wiley.com/doi/10.1029/94WR01791/pdf).

A couple colleagues and I published a paper last year trying to adapt that thinking to scaling of flow-duration curves, taking into account the effects of omitted variable bias, and trying to make an approach to the unification of quantile versus moment analysis (not very successfully on the last point but perhaps the problem is better demonstrated than in the past): http://onlinelibrary.wiley.com/doi/10.1002/2014WR015924/pdf

So I guess that's what I know about, i.e., what the USGS does and what I have been working on recently. Hope it helps! Let me know if you have questions.

Best,

Tom

Sure, but that's a very big topic. How to narrow it down? As I've been with the USGS now for 15 years, you'll get very much a USGS-tinged view of things, which is focused on at-site flood frequency and regional flood frequency (under stationary conditions).

For at-site, as you may know, the general US federal government recommendation is to fit the log-Pearson type III distribution. Along those lines, the long-awaited update to Bulletin17B, called Bulletin17C, has been out for public comment, and has been revised in response to those comments. See:

http://acwi.gov/hydrology/Frequency/b17c/index.html

One big innovation in Bulletin 17C is "Expected Moments Analysis" (EMA) which allows moments for historic or other threshold measurement-based information to be included:

Cohn, T.A., Lane, W.L., and Baier, W.G., 1997, An algorithm for computing moments-based flood quantile estimates when historical flood information is available: Water Resources Research, v. 33, no. 9, p. 2089–2096.

Cohn, T.A., Lane, W.L., and Stedinger, J.R., 2001, Confidence intervals for expected moments algorithm flood quantile estimates: Water Resources Research, v. 37, no. 6, p. 1695–1706.

EMA is already implemented in PeakFQ, at:

http://water.usgs.gov/software/PeakFQ/

On regional flood frequency, the standard technology for USGS has been for ~30 years GLS regression on basin characteristics, computed one quantile at a time, going back to papers by Jery Stedinger and Gary Tasker in the 1980s, now implemented in WREG:

http://water.usgs.gov/software/WREG/

You could start with the manual for the program to see the theoretical background: http://pubs.usgs.gov/tm/tm4a8/.

A typical product of this type of study might be:

http://il.water.usgs.gov/projects/2004_flood_freq/

Bulletin 17B-based regional flood frequency work also brings you to the topic of regional skew. A former student of Jery Stedinger, Andrea Veilleux, has been working to implement a new method in the U.S. based on her Ph.D. work using Bayesian regional regression; see, e.g.,

http://pubs.usgs.gov/sir/2010/5260/

We're not totally ignorant of non-stationarity; you are probably aware of several important descriptive papers from recent years by USGS scientists such as:

http://onlinelibrary.wiley.com/doi/10.1029/2005GL024476/full ("Nature's style: naturally trendy")

and

http://science.sciencemag.org/content/319/5863/573 ("Stationarity Is Dead: Whither Water Management?"),

and we do have a new workgroup on the topic that is looking at when and how to address the problem. Since it's so new I can't predict very much about what we'll come up with.

But along the lines of non-stationarity, from a personal perspective, recently a pair of reports from a project that I have been working on for several years came out in which we used regional quantile regression to assess and adjust for the effect of urbanization on flood frequency. I think quantile regression holds a lot of promise for regional flood frequency because it allows you estimate the exceedance probability of a given flood event at a given site based on regional information, even in the presence of non-stationarity (though climatic non-stationary would require a modification of the particular approach we used), without going through the whole at-site flood frequency thing and THEN doing a regionalization analysis. These reports are at:

https://pubs.er.usgs.gov/publication/sir20165050

and

https://pubs.er.usgs.gov/publication/sir20165049

One other personal perspective is that I am still interested in scaling of floods (and streamflow statistics in general), a la my Ph.D. advisor, Vijay Gupta, (see, e.g, http://onlinelibrary.wiley.com/doi/10.1029/94WR01791/pdf).

A couple colleagues and I published a paper last year trying to adapt that thinking to scaling of flow-duration curves, taking into account the effects of omitted variable bias, and trying to make an approach to the unification of quantile versus moment analysis (not very successfully on the last point but perhaps the problem is better demonstrated than in the past): http://onlinelibrary.wiley.com/doi/10.1002/2014WR015924/pdf

So I guess that's what I know about, i.e., what the USGS does and what I have been working on recently. Hope it helps! Let me know if you have questions.

Best,

Tom

Nicely done, Tom. You do not like L-moments though.

ReplyDelete