Download presentation
Presentation is loading. Please wait.
Published byAnabel Andrews Modified over 8 years ago
1
Glenn E. Moglen Department of Civil & Environmental Engineering Virginia Tech Flood Frequency: Peak Flow Regionalization CEE 5324 –Advanced Hydrology – Lecture 20
2
Questions? Announcement: Exam 2 on April 9 (take home) Flood Frequency PeakFQ Flowchart comment Begin Regression Equation Regionalization US Wide Documents Virginia Documents Regression equation Fundamental Terms Regression equation Goodness of Fit Today’s Agenda
3
Flowchart in PeakFQ document Suggested Bulletin 17B analysis sequence.
4
What does “regionalization” mean? Ans. Regionalization is the extension of the results from flood frequency analyses at a set of gaged locations in a region to all ungaged locations within a region. How is regionalization typically done? Ans. Typically, flood frequency analysis results are related to watershed characteristics through multiple regression. What are some examples of this regression approach? Begin Regionalization
5
http://water.usgs.gov/osw/programs/nss/pubs.html National Streamflow Statistics (NSS) Program
6
These equations are from Miller (1978). The Bisese (1995) equations supersede the ones shown here. “Old” Virginia equations
7
Current rural peak flow report for Virginia
8
Northern Valley and Ridge indicated Peak discharge regions identified by Bisese (1995)
9
Regional peak flow equations in Virginia
10
Regional peak flow equations in Virginia – “NV” region only
11
See pages 8-10 of Jennings (1993) report for summary http://pubs.usgs.gov/wri/1994/4002/report.pdf Urban Equations
12
Shameless promotion… Moglen & Shivers (2006) Urban Equations – 2 nd Try http://pubs.usgs.gov/sir/2006/5270/
13
Classic linear model structure: Basic power law structure: Log-Transform of power law structure: Regression Equation Fundamentals: Definitions and Structure Criterion Variable Predictor Variables
14
In log-transform, note that if x 2 (or any criterion variable) is zero then logarithm is undefined. For this reason, an arbitrary constant is often added: So power law structure changes mildly: Regression Equation Fundamentals: Model Structure
15
Calibration: using a known set of observations of both predictor and criterion variables and determining (through regression or other methods) the coefficients and exponents in a given model structure. Prediction: using a calibrated equation where predictor variables have been determined to estimate the criterion variable. Regression Equation Fundamentals: Calibration vs. Prediction
16
Consider the model: Does the sign of the exponents on drainage area ( A, in mi 2 ) and forest cover ( F, in % ) make sense? If so, then these are “rational”. L is main channel length (in miles). What should the sign of c 3 be? Regression Equation Fundamentals: Rationality
17
The quality of regression equation in predicting a criterion variable is quantified through measures of GOF. S e /S y Bias USGS measures: ( S e,percent, N r ) Note: The correlation coefficient ( R or R 2 ) is probably what you are most familiar with. But is not truly appropriate here because we are NOT dealing with linear relationships. Goodness-of-Fit (GOF)
18
Standard error ( S e ) is appropriate: where = n – p – 1, and p is the number of predictor variables. Standard error is analogous to the standard deviation ( S y ). Both measure variability about a central tendency. Goodness-of-Fit (GOF)
19
Relative standard error is simply the ratio of standard error to the standard deviation. S e /S y less than 1 is good. The closer to zero the better. In the context of this course, please note that in the above equation, the “ y ” values are discharges ( Q ’s). Important: NOT log( Q )’s (unless otherwise indicated). Goodness-of-Fit (GOF)
20
Standard Error of Prediction (note: S e,USGS is in “log units”) USGS Goodness-of-fit values
21
Equivalent Years of Record USGS Goodness-of-fit values
22
Comments here are limited to power model functional form favored by the USGS Classic (and quick) method is log-log transform of data, followed by linear regression Another, slightly more complicated method, is non-linear numerical optimization. Methods give different results. Let’s explore… Methods of Regression Equation Calibration
23
We’ve seen this before in the context of the Horton Ratio calculations. Recipe: 1. Take log of predictor and criterion variables. 2. Use linear regression on log transformed data 3. Note that: Methods of Regression Equation Calibration: Log-log transform / linear regression
Similar presentations
© 2024 SlidePlayer.com. Inc.
All rights reserved.