Term project on "Flood Frequency Analysis" for "Water Resources in Changing Environment" class at University of Central Florida.
An interactive web application was built using the Shiny package in R for showing the results of the analysis: Shiny App
- U.S. Geological Survey's (USGS) National Water Information System.
- Peak discharge data for 8 gauge stations on Pearl river, Mississippi were selected for the flood frequency analysis.
Table: USGS gauge stations
| Station name | USGS Code |
|---|---|
| Jackson | "USGS02486000" |
| Edinburg | "USGS02482000" |
| Carthage | "USGS02482550" |
| Lena | "USGS02483500" |
| Rockport | "USGS02488000" |
| Monticello | "USGS02488500" |
| Columbia | "USGS02489000" |
| Bogalusa | "USGS02489500" |
We used following probability distributions for modelling annual maxima streamflow time series:
- Normal distribution
- Lognormal distribution
- Gamma distribution
- Pearson type 3 distribution
- Log-Pearson type 3 distribution
- Gumbel distribution
- Weibull distribution
- Exponential distribution
We selected following methods for estimating parameters of the distributions for this study:
- Maximum Likelihood Estimation (MLE)
- Method of Moments (MOM)
- Probability Weighted Moments (PWM)
Goodness-of-fit tests are used to summarise the discrepancy between a statistical model and the observed data. They are useful for comparing the observed values with either the values fitted by a model of interest or theoretical quantiles of a known sampling distribution. We used following metrics for determining whether a fit is satisfactory or not.
- Root-Mean-Square Error (RMSE)
- Kolmogorov-Smirnov test (K-S)
- Anderson-Darling test (A-D)
- Akaike Information Criterion (AIC)
- Bayesian Information Criterion (BIC)
- L-moment ratio diagram