Discussion: View Thread

The EWRI Currents magazine just presented a great article: "A Method to Estimate Return Period of Extreme
Events Applied to Hurricane Harvey Rainfall" by Christina Hughes and others. Lumped in with the historical
dataset of all kinds of precipitation in Houston, the Harvey accumulations have astronomically large return periods and thus
exceedingly small exceedance probabilities, and yet would we expect a landfalling and lingering hurricane to strike Houston
only every 90,000 to 100,000 years? I'd like to suggest that we think about one approach to the concerns they raise:
separating precipitation into unique storm types before statistical characterization.

The very large precipitation amounts recorded from Hurricane Harvey are making many rethink how
return periods and exceedance probabilities should be developed.  Kate Hirschboeck of the U of A Tree-Ring
Laboratory in Tucson wrote her dissertation on hydroclimatology, which may provide an answer. As I understand her
contribution, storms in the southwest were categorized by type and then characterized separately for each type. 
The premise is that, due to different atmospheric driving processes, different storm types are expected to have different
occurrence and magnitude frequency distributions.  In Arizona, that means that probabilistic characterization of precipitation
(and associated runoff) due to our three primary storm types may lead to unique characterizations and indeed that is what
we find: the probability distributions of our monsoon/convective, winter/frontal and recurvent tropical storms are different.

The tropical storms arriving in Arizona are essentially the tail ends of landfalling hurricanes and they have produced some of the
largest precipitation totals when viewed against records of all precipitation totals for all storm types.  Harvey's place in
Houston's (mixed storm type) long term precipitation records may be an especially notable example of this same situation.

Perhaps one approach is to develop conditional distributions: the first, conditioning probability is that of a storm type and the second,
conditional probability is that of precipitation amount for that specific  type of storm.  In Harvey's case, it would be the probability
of a hurricane striking Houston and then the probability of rainfall magnitude given that the storm is a hurricane (as opposed to just
another summer shower, for example).

Any thoughts?

------------------------------
Peter Mock Aff.M.ASCE
Hydrologist
PMGC
Paradise Valley AZ
------------------------------

 90,000 to 100,000 years is neither an outlier, nor sensical design value.  So I agree, that the type of storm matters and design data should reflect it.  A stalled hurricane is rare, but can happen and did happen.  Comparing it to geological timescales doe not provide any useful criteria.  Because the news has already moved on to this weeks scandal, I am not up to date on the recovery or overall impact on the Houston infrastructure/economy.  This event exceeded the envelope that can be designed for, like a tornado.  So the recovery and damage can serve only as a case study as to what infrastructure is the most vulnerable and how to protect it.  Not to design to handle the rainfall, but simply protect critical systems.

Many lessons can be learned from the Fukushima disaster.  Essentially, the backup generators were below sea level and subject to flooding. 

Can something happen?  Has it happened in the past?  Is there an easy way to mitigate the damage?  The return period does not come into play. 




------------------------------
Chad Morrison P.E., M.ASCE
Professional Engineer
Greenville RI
------------------------------

Original Message:
Sent: 11-01-2017 18:37
From: Peter Mock
Subject: Return Periods of (Mixed) Extreme Events

The EWRI Currents magazine just presented a great article: "A Method to Estimate Return Period of Extreme
Events Applied to Hurricane Harvey Rainfall" by Christina Hughes and others. Lumped in with the historical
dataset of all kinds of precipitation in Houston, the Harvey accumulations have astronomically large return periods and thus
exceedingly small exceedance probabilities, and yet would we expect a landfalling and lingering hurricane to strike Houston
only every 90,000 to 100,000 years? I'd like to suggest that we think about one approach to the concerns they raise:
separating precipitation into unique storm types before statistical characterization.

The very large precipitation amounts recorded from Hurricane Harvey are making many rethink how
return periods and exceedance probabilities should be developed.  Kate Hirschboeck of the U of A Tree-Ring
Laboratory in Tucson wrote her dissertation on hydroclimatology, which may provide an answer. As I understand her
contribution, storms in the southwest were categorized by type and then characterized separately for each type.
The premise is that, due to different atmospheric driving processes, different storm types are expected to have different
occurrence and magnitude frequency distributions.  In Arizona, that means that probabilistic characterization of precipitation
(and associated runoff) due to our three primary storm types may lead to unique characterizations and indeed that is what
we find: the probability distributions of our monsoon/convective, winter/frontal and recurvent tropical storms are different.

The tropical storms arriving in Arizona are essentially the tail ends of landfalling hurricanes and they have produced some of the
largest precipitation totals when viewed against records of all precipitation totals for all storm types.  Harvey's place in
Houston's (mixed storm type) long term precipitation records may be an especially notable example of this same situation.

Perhaps one approach is to develop conditional distributions: the first, conditioning probability is that of a storm type and the second,
conditional probability is that of precipitation amount for that specific  type of storm.  In Harvey's case, it would be the probability
of a hurricane striking Houston and then the probability of rainfall magnitude given that the storm is a hurricane (as opposed to just
another summer shower, for example).

Any thoughts?

------------------------------
Peter Mock Aff.M.ASCE
Hydrologist
PMGC
Paradise Valley AZ
------------------------------

While I agree to the approach described in the paper which I have yet not read, but only from the description presented in your article, I'd just like to call attention to a common mistake done in the analyses of extremes as well as yearly statistics. This refers to the use of calendar yearly data (January to December) to derive the largest or the few largest yearly events, aproach which is wrong because winters in most places start from say November to April and as such extreme events in the November to December months may be related to those in the January to April period. Hence, a correct extreme and annual analysis should consist of independent events and not from related events from a same atmospheric or climatic process occuring say in December and January. The correct division is to use hydrographic or hydrological years, which may differ in various globe regions in the start and end dates of the climatic yearly periods. November 03, 2017.

------------------------------
Dov Rosen P.E., M.ASCE
Sergiu Dov Rosen Sea Shore Rosen Engineering Consultants
Haifa, Israel
------------------------------

Original Message:
Sent: 11-01-2017 18:37
From: Peter Mock
Subject: Return Periods of (Mixed) Extreme Events

The EWRI Currents magazine just presented a great article: "A Method to Estimate Return Period of Extreme
Events Applied to Hurricane Harvey Rainfall" by Christina Hughes and others. Lumped in with the historical
dataset of all kinds of precipitation in Houston, the Harvey accumulations have astronomically large return periods and thus
exceedingly small exceedance probabilities, and yet would we expect a landfalling and lingering hurricane to strike Houston
only every 90,000 to 100,000 years? I'd like to suggest that we think about one approach to the concerns they raise:
separating precipitation into unique storm types before statistical characterization.

The very large precipitation amounts recorded from Hurricane Harvey are making many rethink how
return periods and exceedance probabilities should be developed.  Kate Hirschboeck of the U of A Tree-Ring
Laboratory in Tucson wrote her dissertation on hydroclimatology, which may provide an answer. As I understand her
contribution, storms in the southwest were categorized by type and then characterized separately for each type.
The premise is that, due to different atmospheric driving processes, different storm types are expected to have different
occurrence and magnitude frequency distributions.  In Arizona, that means that probabilistic characterization of precipitation
(and associated runoff) due to our three primary storm types may lead to unique characterizations and indeed that is what
we find: the probability distributions of our monsoon/convective, winter/frontal and recurvent tropical storms are different.

The tropical storms arriving in Arizona are essentially the tail ends of landfalling hurricanes and they have produced some of the
largest precipitation totals when viewed against records of all precipitation totals for all storm types.  Harvey's place in
Houston's (mixed storm type) long term precipitation records may be an especially notable example of this same situation.

Perhaps one approach is to develop conditional distributions: the first, conditioning probability is that of a storm type and the second,
conditional probability is that of precipitation amount for that specific  type of storm.  In Harvey's case, it would be the probability
of a hurricane striking Houston and then the probability of rainfall magnitude given that the storm is a hurricane (as opposed to just
another summer shower, for example).

Any thoughts?

------------------------------
Peter Mock Aff.M.ASCE
Hydrologist
PMGC
Paradise Valley AZ
------------------------------

Interesting thought.   For types what would be  the defining characterisics you might consider? You mention hurricanes  which by defintion are defined by there place of origin.   Up in the Northeast are  "Nor'easters which have different orgin and different.  I am just wondering because it would effect the data set.

------------------------------
Ronald Burns P.E., M.ASCE
Principal Engineer
Brookline MA
------------------------------

Original Message:
Sent: 11-01-2017 18:37
From: Peter Mock
Subject: Return Periods of (Mixed) Extreme Events

The EWRI Currents magazine just presented a great article: "A Method to Estimate Return Period of Extreme
Events Applied to Hurricane Harvey Rainfall" by Christina Hughes and others. Lumped in with the historical
dataset of all kinds of precipitation in Houston, the Harvey accumulations have astronomically large return periods and thus
exceedingly small exceedance probabilities, and yet would we expect a landfalling and lingering hurricane to strike Houston
only every 90,000 to 100,000 years? I'd like to suggest that we think about one approach to the concerns they raise:
separating precipitation into unique storm types before statistical characterization.

The very large precipitation amounts recorded from Hurricane Harvey are making many rethink how
return periods and exceedance probabilities should be developed.  Kate Hirschboeck of the U of A Tree-Ring
Laboratory in Tucson wrote her dissertation on hydroclimatology, which may provide an answer. As I understand her
contribution, storms in the southwest were categorized by type and then characterized separately for each type.
The premise is that, due to different atmospheric driving processes, different storm types are expected to have different
occurrence and magnitude frequency distributions.  In Arizona, that means that probabilistic characterization of precipitation
(and associated runoff) due to our three primary storm types may lead to unique characterizations and indeed that is what
we find: the probability distributions of our monsoon/convective, winter/frontal and recurvent tropical storms are different.

The tropical storms arriving in Arizona are essentially the tail ends of landfalling hurricanes and they have produced some of the
largest precipitation totals when viewed against records of all precipitation totals for all storm types.  Harvey's place in
Houston's (mixed storm type) long term precipitation records may be an especially notable example of this same situation.

Perhaps one approach is to develop conditional distributions: the first, conditioning probability is that of a storm type and the second,
conditional probability is that of precipitation amount for that specific  type of storm.  In Harvey's case, it would be the probability
of a hurricane striking Houston and then the probability of rainfall magnitude given that the storm is a hurricane (as opposed to just
another summer shower, for example).

Any thoughts?

------------------------------
Peter Mock Aff.M.ASCE
Hydrologist
PMGC
Paradise Valley AZ
------------------------------

This post was removed

​Would that change the approach to flood protection design, 100-year flood mapping and the like?

------------------------------
Michael Clough P.E., M.ASCE
Senior Project Manager
Humble TX
------------------------------

Original Message:
Sent: 11-01-2017 18:37
From: Peter Mock
Subject: Return Periods of (Mixed) Extreme Events

The EWRI Currents magazine just presented a great article: "A Method to Estimate Return Period of Extreme
Events Applied to Hurricane Harvey Rainfall" by Christina Hughes and others. Lumped in with the historical
dataset of all kinds of precipitation in Houston, the Harvey accumulations have astronomically large return periods and thus
exceedingly small exceedance probabilities, and yet would we expect a landfalling and lingering hurricane to strike Houston
only every 90,000 to 100,000 years? I'd like to suggest that we think about one approach to the concerns they raise:
separating precipitation into unique storm types before statistical characterization.

The very large precipitation amounts recorded from Hurricane Harvey are making many rethink how
return periods and exceedance probabilities should be developed.  Kate Hirschboeck of the U of A Tree-Ring
Laboratory in Tucson wrote her dissertation on hydroclimatology, which may provide an answer. As I understand her
contribution, storms in the southwest were categorized by type and then characterized separately for each type.
The premise is that, due to different atmospheric driving processes, different storm types are expected to have different
occurrence and magnitude frequency distributions.  In Arizona, that means that probabilistic characterization of precipitation
(and associated runoff) due to our three primary storm types may lead to unique characterizations and indeed that is what
we find: the probability distributions of our monsoon/convective, winter/frontal and recurvent tropical storms are different.

The tropical storms arriving in Arizona are essentially the tail ends of landfalling hurricanes and they have produced some of the
largest precipitation totals when viewed against records of all precipitation totals for all storm types.  Harvey's place in
Houston's (mixed storm type) long term precipitation records may be an especially notable example of this same situation.

Perhaps one approach is to develop conditional distributions: the first, conditioning probability is that of a storm type and the second,
conditional probability is that of precipitation amount for that specific  type of storm.  In Harvey's case, it would be the probability
of a hurricane striking Houston and then the probability of rainfall magnitude given that the storm is a hurricane (as opposed to just
another summer shower, for example).

Any thoughts?

------------------------------
Peter Mock Aff.M.ASCE
Hydrologist
PMGC
Paradise Valley AZ
------------------------------

Although I have not yet read the full article, the concept is intriguing.  Suggest fleshing it out by establishing consensus storm types for a given region.  Then run parallel projections of storm frequency to compare results of conventional vs. proposed methodologies to actual storms.

------------------------------
[John] [Casana] [PE. F. ASCE, D WRE]
[Senior Lead Engineer]
[Booz Allen Hamilton]
[Washington] [DC]
[ASCE Director][Region 2]
------------------------------

Original Message:
Sent: 11-03-2017 09:24
From: Michael Clough
Subject: Return Periods of (Mixed) Extreme Events

​Would that change the approach to flood protection design, 100-year flood mapping and the like?

------------------------------
Michael Clough P.E., M.ASCE
Senior Project Manager
Humble TX

Original Message:
Sent: 11-01-2017 18:37
From: Peter Mock
Subject: Return Periods of (Mixed) Extreme Events

The EWRI Currents magazine just presented a great article: "A Method to Estimate Return Period of Extreme
Events Applied to Hurricane Harvey Rainfall" by Christina Hughes and others. Lumped in with the historical
dataset of all kinds of precipitation in Houston, the Harvey accumulations have astronomically large return periods and thus
exceedingly small exceedance probabilities, and yet would we expect a landfalling and lingering hurricane to strike Houston
only every 90,000 to 100,000 years? I'd like to suggest that we think about one approach to the concerns they raise:
separating precipitation into unique storm types before statistical characterization.

The very large precipitation amounts recorded from Hurricane Harvey are making many rethink how
return periods and exceedance probabilities should be developed.  Kate Hirschboeck of the U of A Tree-Ring
Laboratory in Tucson wrote her dissertation on hydroclimatology, which may provide an answer. As I understand her
contribution, storms in the southwest were categorized by type and then characterized separately for each type.
The premise is that, due to different atmospheric driving processes, different storm types are expected to have different
occurrence and magnitude frequency distributions.  In Arizona, that means that probabilistic characterization of precipitation
(and associated runoff) due to our three primary storm types may lead to unique characterizations and indeed that is what
we find: the probability distributions of our monsoon/convective, winter/frontal and recurvent tropical storms are different.

The tropical storms arriving in Arizona are essentially the tail ends of landfalling hurricanes and they have produced some of the
largest precipitation totals when viewed against records of all precipitation totals for all storm types.  Harvey's place in
Houston's (mixed storm type) long term precipitation records may be an especially notable example of this same situation.

Perhaps one approach is to develop conditional distributions: the first, conditioning probability is that of a storm type and the second,
conditional probability is that of precipitation amount for that specific  type of storm.  In Harvey's case, it would be the probability
of a hurricane striking Houston and then the probability of rainfall magnitude given that the storm is a hurricane (as opposed to just
another summer shower, for example).

Any thoughts?

------------------------------
Peter Mock Aff.M.ASCE
Hydrologist
PMGC
Paradise Valley AZ
------------------------------

A quick reaction to your summary is that this seems to me very similar to the approach taken in defining probable maximum precipitation.  Pulling my old college hydrology textbook ("Hydrology for Engineers", by Linsley, Kohler, and Paulus, 1958 edition) I note that for the Houston area, the probable maximum appears to be about 30-inches of in 24-hours over a 1000 square mile area.

------------------------------
Stephen Nelson P.E., M.ASCE
Consultant
Coal Creek Envtl Assoc
Bellevue WA
------------------------------

Original Message:
Sent: 11-01-2017 18:37
From: Peter Mock
Subject: Return Periods of (Mixed) Extreme Events

The EWRI Currents magazine just presented a great article: "A Method to Estimate Return Period of Extreme
Events Applied to Hurricane Harvey Rainfall" by Christina Hughes and others. Lumped in with the historical
dataset of all kinds of precipitation in Houston, the Harvey accumulations have astronomically large return periods and thus
exceedingly small exceedance probabilities, and yet would we expect a landfalling and lingering hurricane to strike Houston
only every 90,000 to 100,000 years? I'd like to suggest that we think about one approach to the concerns they raise:
separating precipitation into unique storm types before statistical characterization.

The very large precipitation amounts recorded from Hurricane Harvey are making many rethink how
return periods and exceedance probabilities should be developed.  Kate Hirschboeck of the U of A Tree-Ring
Laboratory in Tucson wrote her dissertation on hydroclimatology, which may provide an answer. As I understand her
contribution, storms in the southwest were categorized by type and then characterized separately for each type.
The premise is that, due to different atmospheric driving processes, different storm types are expected to have different
occurrence and magnitude frequency distributions.  In Arizona, that means that probabilistic characterization of precipitation
(and associated runoff) due to our three primary storm types may lead to unique characterizations and indeed that is what
we find: the probability distributions of our monsoon/convective, winter/frontal and recurvent tropical storms are different.

The tropical storms arriving in Arizona are essentially the tail ends of landfalling hurricanes and they have produced some of the
largest precipitation totals when viewed against records of all precipitation totals for all storm types.  Harvey's place in
Houston's (mixed storm type) long term precipitation records may be an especially notable example of this same situation.

Perhaps one approach is to develop conditional distributions: the first, conditioning probability is that of a storm type and the second,
conditional probability is that of precipitation amount for that specific  type of storm.  In Harvey's case, it would be the probability
of a hurricane striking Houston and then the probability of rainfall magnitude given that the storm is a hurricane (as opposed to just
another summer shower, for example).

Any thoughts?

------------------------------
Peter Mock Aff.M.ASCE
Hydrologist
PMGC
Paradise Valley AZ
------------------------------

Considering it was a 4 or 3 day event depending on whether the side of town was east of I-45 or west it.  The event also impacted more than 10,000 square miles.  The 38-50 inches of rainfall is probably not a maximum probable event, but I would bet it is close to it.  Another way of looking at it is that Houston's current 100-year 24-hour rainfall is about 13 inches of rain.  Most of Houston had approximately back to back 100-year storms for 4 or 3 days in a row.  Not much of Houston area got any day in excess of the 19-inch 500-year rainfall.

Some thoughts on how to design to combat back to back improbable events.  
1.  Detention storage is not the answer.  In order to design for detention storage, it is necessary to select the total volume of rainfall.  Obviously, even the Corps of Engineers system in the reservoirs, which are the only systems in Houston designed for a 500-year event with freeboard couldn't provide adequate storage, even though they started out as empty.
2. Given Houston's proximity to the ocean, conveyance may be an answer.  If all the major drainage systems in Houston area were capable of handling the flows from the 100-year 24-hour rainfall event, very little of the City would have flooded.  Keep in mind, to design all the systems in Houston to the 100-year capacity would require trillions of dollars.  Note this system was designed in the 1930s by the Corps of Engineers, and was meant to supplement the storage reservoirs.  This is one of the reasons the Corps did not purchase the extra land for the freeboard storage, as their plan included significant conveyance improvements that were not constructed.
Potential problems to implementing a conveyance systems.
1. Channels would have to get bigger.  This means takings of some of the most expensive land in Houston, Bayou Frontage.  Possible 2-3 rows of houses along channels would have to be removed in some watersheds.  Major pieces of downtown would have to be purchased or bypassed.
2. Preservationist groups have been arguing that the creation of channels disrupts the natural flows and environmental conditions of the channels.  They are not wrong.  But so does high velocity floods, and temporarily full channels.  Both problems can be solved if the systems were redesigned properly.  These systems could create very wide, gently sloping channels that could become habitat to humans, vegetation, and wildlife while protecting the development that remains.  Higher areas could be vegetated to recreate the forest areas.  Lower areas could be kept mowed to provide the majority of the conveyance system requirements.  Trails could be added to both.
3.  There is not enough money in Texas to pay for these improvements quickly.  But we could start by purchasing the homes in Brays Bayou that flooded 3 years in a row in an involuntary buyout program, and stop the insanity of allowing homes to be rebuilt, and resources spent in areas that flood regularly.  We need to start admitting that some of our most populated Bayous are not currently designed for 100-year flood protection, and probably never will be.
4. We could start requiring flood insurance on all homes that the federal government provides mortgage protection for, instead of only the ones that are in the area mapped as 100-year floodplain. 
5, We could start describing risk in terms of percent chance of flooding over a 10-year period instead of a 1-year period.  Most people would accept a 1% chance per year, but would they accept a 9% chance over 10-years, or a 27% chance over 30-years?   These all represent approximately the same storm.


------------------------------
Dwayne Culp, Ph.D., P.E., P.Eng, M.ASCE
Culp Engineering, LLC
Richmond TX
------------------------------

Original Message:
Sent: 11-07-2017 02:11
From: Stephen Nelson
Subject: Return Periods of (Mixed) Extreme Events

A quick reaction to your summary is that this seems to me very similar to the approach taken in defining probable maximum precipitation.  Pulling my old college hydrology textbook ("Hydrology for Engineers", by Linsley, Kohler, and Paulus, 1958 edition) I note that for the Houston area, the probable maximum appears to be about 30-inches of in 24-hours over a 1000 square mile area.

------------------------------
Stephen Nelson P.E., M.ASCE
Consultant
Coal Creek Envtl Assoc
Bellevue WA

Original Message:
Sent: 11-01-2017 18:37
From: Peter Mock
Subject: Return Periods of (Mixed) Extreme Events

The EWRI Currents magazine just presented a great article: "A Method to Estimate Return Period of Extreme
Events Applied to Hurricane Harvey Rainfall" by Christina Hughes and others. Lumped in with the historical
dataset of all kinds of precipitation in Houston, the Harvey accumulations have astronomically large return periods and thus
exceedingly small exceedance probabilities, and yet would we expect a landfalling and lingering hurricane to strike Houston
only every 90,000 to 100,000 years? I'd like to suggest that we think about one approach to the concerns they raise:
separating precipitation into unique storm types before statistical characterization.

The very large precipitation amounts recorded from Hurricane Harvey are making many rethink how
return periods and exceedance probabilities should be developed.  Kate Hirschboeck of the U of A Tree-Ring
Laboratory in Tucson wrote her dissertation on hydroclimatology, which may provide an answer. As I understand her
contribution, storms in the southwest were categorized by type and then characterized separately for each type.
The premise is that, due to different atmospheric driving processes, different storm types are expected to have different
occurrence and magnitude frequency distributions.  In Arizona, that means that probabilistic characterization of precipitation
(and associated runoff) due to our three primary storm types may lead to unique characterizations and indeed that is what
we find: the probability distributions of our monsoon/convective, winter/frontal and recurvent tropical storms are different.

The tropical storms arriving in Arizona are essentially the tail ends of landfalling hurricanes and they have produced some of the
largest precipitation totals when viewed against records of all precipitation totals for all storm types.  Harvey's place in
Houston's (mixed storm type) long term precipitation records may be an especially notable example of this same situation.

Perhaps one approach is to develop conditional distributions: the first, conditioning probability is that of a storm type and the second,
conditional probability is that of precipitation amount for that specific  type of storm.  In Harvey's case, it would be the probability
of a hurricane striking Houston and then the probability of rainfall magnitude given that the storm is a hurricane (as opposed to just
another summer shower, for example).

Any thoughts?

------------------------------
Peter Mock Aff.M.ASCE
Hydrologist
PMGC
Paradise Valley AZ
------------------------------

Maybe I'm misinterpreting, but it appears from this discussion thread that we're looking at Hurricane Harvey as an extreme event, which, if we looked hard enough, could have been predicted.  I don't think that's a worthwhile exercise.  For centuries, engineers have based their designs on an obscure but fundamental assumption of stationarity:  the statistical properties of engineering design parameters will be the same in the future as they have been in the past.  Under conditions of stationarity, the averages and variances of engineering design parameters are statistically constant and therefore predictable.  That assumption enables engineers to design and build long-lived infrastructure projects using the body of knowledge of historical conditions and past performance of similar projects.   

Unfortunately, the assumption of stationarity is no longer reliable.  Increasing atmospheric concentrations of so-called greenhouse gases, emissions mainly from the burning of fossil fuels, is increasing the energy retained in the Earth's climate system.  This energy, in the form of heat, continues to destabilize the climate system, resulting in increasing ambient temperatures and extreme weather events.  Consequently, environmental operating conditions, the conditions under which our infrastructure is supposed to operate, have changed and continue to change significantly, in ways that are not readily predictable.  Non-stationarity is now the new normal and brings with it new and substantial and risks to infrastructure and the public that uses it.  

Last month at ASCE's International Conference on Sustainable Infrastructure, I presented a paper, "Delivering Safe, Cost-Effective, Sustainable Civil Infrastructure Projects Under Conditions of Non-Stationarity."  My colleagues and I proposed new methodology for the design and delivery of infrastructure projects, one that incorporates combinations of robustness, resilience, redundancy and adaptability based on risk and cost.  We think the methodology is a start for what will be a fundamental change in the way infrastructure projects are designed, constructed and operated.

------------------------------
Bill Wallace ENV SP, F.ASCE
Wilsonville OR
------------------------------

Original Message:
Sent: 11-01-2017 18:37
From: Peter Mock
Subject: Return Periods of (Mixed) Extreme Events

The EWRI Currents magazine just presented a great article: "A Method to Estimate Return Period of Extreme
Events Applied to Hurricane Harvey Rainfall" by Christina Hughes and others. Lumped in with the historical
dataset of all kinds of precipitation in Houston, the Harvey accumulations have astronomically large return periods and thus
exceedingly small exceedance probabilities, and yet would we expect a landfalling and lingering hurricane to strike Houston
only every 90,000 to 100,000 years? I'd like to suggest that we think about one approach to the concerns they raise:
separating precipitation into unique storm types before statistical characterization.

The very large precipitation amounts recorded from Hurricane Harvey are making many rethink how
return periods and exceedance probabilities should be developed.  Kate Hirschboeck of the U of A Tree-Ring
Laboratory in Tucson wrote her dissertation on hydroclimatology, which may provide an answer. As I understand her
contribution, storms in the southwest were categorized by type and then characterized separately for each type.
The premise is that, due to different atmospheric driving processes, different storm types are expected to have different
occurrence and magnitude frequency distributions.  In Arizona, that means that probabilistic characterization of precipitation
(and associated runoff) due to our three primary storm types may lead to unique characterizations and indeed that is what
we find: the probability distributions of our monsoon/convective, winter/frontal and recurvent tropical storms are different.

The tropical storms arriving in Arizona are essentially the tail ends of landfalling hurricanes and they have produced some of the
largest precipitation totals when viewed against records of all precipitation totals for all storm types.  Harvey's place in
Houston's (mixed storm type) long term precipitation records may be an especially notable example of this same situation.

Perhaps one approach is to develop conditional distributions: the first, conditioning probability is that of a storm type and the second,
conditional probability is that of precipitation amount for that specific  type of storm.  In Harvey's case, it would be the probability
of a hurricane striking Houston and then the probability of rainfall magnitude given that the storm is a hurricane (as opposed to just
another summer shower, for example).

Any thoughts?

------------------------------
Peter Mock Aff.M.ASCE
Hydrologist
PMGC
Paradise Valley AZ
------------------------------

We faced the problem when Hurricane and then Tropical Storm Agnus hit PA and NY in June 1972.  The Corps' Hydrologic Engineering Center suggested that if prevalent (southern states), hurricanes/tropical storms be separated. They developed a procedure for the analysis and than putting them back together.

For rare circumstances of either (NY, PA and Northern states throughout the US), you don't read the exceedence interval directly from a Log-Pearson Type 3 frequency curve derived from a period of record between 50- and 100-years, but use the discharge with the period of record you have and see if you can find out if the storm is the greatest since, say 1800 or 1825 and than use the discharge in the analysis, only extend the period of record to the 1800's, plot the curve and then read the excellence frequency from the curve.  The only problem is that the confidence limits shoot off to the moon and who knows what is even remotely real.

Regardless, you do the best you can and if you need an answer, lean the curve a bit to the right.

Brad Price, PE
F. ASCE

------------------------------
Bradford Price P.E., F.ASCE
RETIRED
Buena Vista VA
------------------------------

Original Message:
Sent: 11-01-2017 18:37
From: Peter Mock
Subject: Return Periods of (Mixed) Extreme Events

The EWRI Currents magazine just presented a great article: "A Method to Estimate Return Period of Extreme
Events Applied to Hurricane Harvey Rainfall" by Christina Hughes and others. Lumped in with the historical
dataset of all kinds of precipitation in Houston, the Harvey accumulations have astronomically large return periods and thus
exceedingly small exceedance probabilities, and yet would we expect a landfalling and lingering hurricane to strike Houston
only every 90,000 to 100,000 years? I'd like to suggest that we think about one approach to the concerns they raise:
separating precipitation into unique storm types before statistical characterization.

The very large precipitation amounts recorded from Hurricane Harvey are making many rethink how
return periods and exceedance probabilities should be developed.  Kate Hirschboeck of the U of A Tree-Ring
Laboratory in Tucson wrote her dissertation on hydroclimatology, which may provide an answer. As I understand her
contribution, storms in the southwest were categorized by type and then characterized separately for each type.
The premise is that, due to different atmospheric driving processes, different storm types are expected to have different
occurrence and magnitude frequency distributions.  In Arizona, that means that probabilistic characterization of precipitation
(and associated runoff) due to our three primary storm types may lead to unique characterizations and indeed that is what
we find: the probability distributions of our monsoon/convective, winter/frontal and recurvent tropical storms are different.

The tropical storms arriving in Arizona are essentially the tail ends of landfalling hurricanes and they have produced some of the
largest precipitation totals when viewed against records of all precipitation totals for all storm types.  Harvey's place in
Houston's (mixed storm type) long term precipitation records may be an especially notable example of this same situation.

Perhaps one approach is to develop conditional distributions: the first, conditioning probability is that of a storm type and the second,
conditional probability is that of precipitation amount for that specific  type of storm.  In Harvey's case, it would be the probability
of a hurricane striking Houston and then the probability of rainfall magnitude given that the storm is a hurricane (as opposed to just
another summer shower, for example).

Any thoughts?

------------------------------
Peter Mock Aff.M.ASCE
Hydrologist
PMGC
Paradise Valley AZ
------------------------------

I don't have experience with hurricanes being from the Pacific Northwest but I'd recommend checking out the US Army Corps of Engineers EM 1110-2-1415, Chapter 10, on mixed population frequency analysis.

------------------------------
Keith Hume P.E., M.ASCE
Engineer Drainage and WQ Infrastructure
Lynnwood WA
------------------------------

Original Message:
Sent: 11-08-2017 13:31
From: Bradford Price
Subject: Return Periods of (Mixed) Extreme Events

We faced the problem when Hurricane and then Tropical Storm Agnus hit PA and NY in June 1972.  The Corps' Hydrologic Engineering Center suggested that if prevalent (southern states), hurricanes/tropical storms be separated. They developed a procedure for the analysis and than putting them back together.

For rare circumstances of either (NY, PA and Northern states throughout the US), you don't read the exceedence interval directly from a Log-Pearson Type 3 frequency curve derived from a period of record between 50- and 100-years, but use the discharge with the period of record you have and see if you can find out if the storm is the greatest since, say 1800 or 1825 and than use the discharge in the analysis, only extend the period of record to the 1800's, plot the curve and then read the excellence frequency from the curve.  The only problem is that the confidence limits shoot off to the moon and who knows what is even remotely real.

Regardless, you do the best you can and if you need an answer, lean the curve a bit to the right.

Brad Price, PE
F. ASCE

------------------------------
Bradford Price P.E., F.ASCE
RETIRED
Buena Vista VA

Original Message:
Sent: 11-01-2017 18:37
From: Peter Mock
Subject: Return Periods of (Mixed) Extreme Events

The EWRI Currents magazine just presented a great article: "A Method to Estimate Return Period of Extreme
Events Applied to Hurricane Harvey Rainfall" by Christina Hughes and others. Lumped in with the historical
dataset of all kinds of precipitation in Houston, the Harvey accumulations have astronomically large return periods and thus
exceedingly small exceedance probabilities, and yet would we expect a landfalling and lingering hurricane to strike Houston
only every 90,000 to 100,000 years? I'd like to suggest that we think about one approach to the concerns they raise:
separating precipitation into unique storm types before statistical characterization.

The very large precipitation amounts recorded from Hurricane Harvey are making many rethink how
return periods and exceedance probabilities should be developed.  Kate Hirschboeck of the U of A Tree-Ring
Laboratory in Tucson wrote her dissertation on hydroclimatology, which may provide an answer. As I understand her
contribution, storms in the southwest were categorized by type and then characterized separately for each type.
The premise is that, due to different atmospheric driving processes, different storm types are expected to have different
occurrence and magnitude frequency distributions.  In Arizona, that means that probabilistic characterization of precipitation
(and associated runoff) due to our three primary storm types may lead to unique characterizations and indeed that is what
we find: the probability distributions of our monsoon/convective, winter/frontal and recurvent tropical storms are different.

The tropical storms arriving in Arizona are essentially the tail ends of landfalling hurricanes and they have produced some of the
largest precipitation totals when viewed against records of all precipitation totals for all storm types.  Harvey's place in
Houston's (mixed storm type) long term precipitation records may be an especially notable example of this same situation.

Perhaps one approach is to develop conditional distributions: the first, conditioning probability is that of a storm type and the second,
conditional probability is that of precipitation amount for that specific  type of storm.  In Harvey's case, it would be the probability
of a hurricane striking Houston and then the probability of rainfall magnitude given that the storm is a hurricane (as opposed to just
another summer shower, for example).

Any thoughts?

------------------------------
Peter Mock Aff.M.ASCE
Hydrologist
PMGC
Paradise Valley AZ
------------------------------

Thank you, everyone, for your postings as I learned or was reminded of several good points. Included in the postings are thoughts on reconsidering the statistical characterization of precipitation events in the Houston area, whether characterization has any value or not, what goes into engineering planning and designing for Houston, and that the statistical characterization may change with time.  All are valuable.

I think there is value to a conditional probability characterization of point rainfall and that many areas of the country, Houston included, have relatively distinct precipitation generating processes that are more prevalent within certain ranges of time within each year.  It may be that hurricanes are a particularly unique type of process that "intervene" in the more common and dependable seasonal storm patterns. I see this as an opportunity to inquire as to how these systems work from a refined perspective (this isn't new as one of our participants posted that this concept was brought up for storms in the NY area in the 1970s - thank you).

Some next steps in the inquiry appear to be identifying distinctive classes of atmospheric processes that bring precipitation to Houston, classifying storms into these classes, and then seeing what the data indicate.  I'll leave that to those with local expertise and experience.  Let us know what you find!

Thanks again,
Peter

------------------------------
Peter Mock Aff.M.ASCE
President
PMGC
Paradise Valley AZ
------------------------------

Original Message:
Sent: 11-01-2017 18:37
From: Peter Mock
Subject: Return Periods of (Mixed) Extreme Events

The EWRI Currents magazine just presented a great article: "A Method to Estimate Return Period of Extreme
Events Applied to Hurricane Harvey Rainfall" by Christina Hughes and others. Lumped in with the historical
dataset of all kinds of precipitation in Houston, the Harvey accumulations have astronomically large return periods and thus
exceedingly small exceedance probabilities, and yet would we expect a landfalling and lingering hurricane to strike Houston
only every 90,000 to 100,000 years? I'd like to suggest that we think about one approach to the concerns they raise:
separating precipitation into unique storm types before statistical characterization.

The very large precipitation amounts recorded from Hurricane Harvey are making many rethink how
return periods and exceedance probabilities should be developed.  Kate Hirschboeck of the U of A Tree-Ring
Laboratory in Tucson wrote her dissertation on hydroclimatology, which may provide an answer. As I understand her
contribution, storms in the southwest were categorized by type and then characterized separately for each type.
The premise is that, due to different atmospheric driving processes, different storm types are expected to have different
occurrence and magnitude frequency distributions.  In Arizona, that means that probabilistic characterization of precipitation
(and associated runoff) due to our three primary storm types may lead to unique characterizations and indeed that is what
we find: the probability distributions of our monsoon/convective, winter/frontal and recurvent tropical storms are different.

The tropical storms arriving in Arizona are essentially the tail ends of landfalling hurricanes and they have produced some of the
largest precipitation totals when viewed against records of all precipitation totals for all storm types.  Harvey's place in
Houston's (mixed storm type) long term precipitation records may be an especially notable example of this same situation.

Perhaps one approach is to develop conditional distributions: the first, conditioning probability is that of a storm type and the second,
conditional probability is that of precipitation amount for that specific  type of storm.  In Harvey's case, it would be the probability
of a hurricane striking Houston and then the probability of rainfall magnitude given that the storm is a hurricane (as opposed to just
another summer shower, for example).

Any thoughts?

------------------------------
Peter Mock Aff.M.ASCE
Hydrologist
PMGC
Paradise Valley AZ
------------------------------