Flood Frequency Confidence Bounds — Art, Science, or Guess!

The current (Bulletin 17B) procedure for estimating confidence bounds for frequency curves assumes that the underlying distribution of flood flows is the Log Pearson type III (LP3) probability density function (pdf). The confidence bounds are estimated using a sampling error procedure. This can result in wild estimates of the confidence bounds which often exceed expected physical basin limits. A test of existing procedures was conducted. For this test, graphical frequency curves were developed from historical annual flood data for three (3) rivers in California. These curves were based on a minimum of 100 years of recorded data and incorporated additional information such as Probable Maximum Flood and Standard Project Flood estimates to ensure the graphical curves were grounded with physical hydrologic and meteorologic limits. These graphically estimated frequency curves were assumed to be the true underlying distribution of flood flows for each river basin. Synthetic records were generated by randomly selecting 100 sets of 100 annual maximum flows from the graphical curves (10,000 record years for each river). Bulletin 17B procedures utilizing a LP3 pdf were employed to calculate the best estimate and 90% confidence bounds for the 100-year, 200-year and 500-year floods. This was done for all of the one hundred 100 year data sets and for an accumulative estimate (100 years, 200 years, 300 years, ..., 10,000 years). For comparison, the same procedure was completed using the Pearson type III (P3) pdf. The results of the study show that estimating the extreme events (100-year, 200-year and 500-year floods) using a predetermined pdf is subject to tremendous error and variability. Utilization of either LP3 or P3 distributions proved to be unreliable and ineffective. An alternate method based on a graphical frequency curve should be considered. Uncertainty in flood flows during the next 100 years should be assessed and a method based on sampling the graphical curve should be considered.