An Example of the Impact that Filled-In Peakflow Data Can Have on Flood Frequency Analysis

Missing peakflow data can be caused by a variety of reasons, for example, changes in funding priorities, relocation of stations, equipment failure, and, severe and unknown channel shape changes. Missing annual peakflow data for five gauging stations in the Santa Clara River Watershed (Ventura and Los Angeles Counties, CA) were filled-in using artificial neural network techniques. The most common record period for stations in this watershed was 73 yrs (1933–2005). The least missing peakflow data for a studied station was 16 and the most was 42. Flood frequency curves were developed using the Log-Pearson Type III method accounting for the weighted skew. A comparison between flood frequency curves using the existing data, and the existing data plus the filled-in data was performed. The smallest and largest changes caused by the filled-in data on predicted peakflows with return periods of 2, 5, 10, 25, 50, 100 and 200 yr on average were 0% and 92%, respectfully. The gauging station with the largest watershed (Santa Clara River at Montalvo, just upstream of the Pacific Ocean) had 18 missing peakflow values. Seventeen of the missing values were close together and early in the record (between 1933 and 1955). Including the filled-in data consistently reduced the flood frequency values entirely across the return period range from 2 to 200 yr. Filling in peakflow data can have a considerable impact on flood frequency curves.