Partial Duration Series of Wet and Dry Years Can Improve Flood Estimates in the Context of a Nonstationary Climate and Anthropogenic Disturbances

Accurately estimating flood levels is essential for effective infrastructure design, reservoir management, and flood risk mapping. Traditional methods for predicting these levels often rely on annual maximum flood (AMF) data, which may not always fit well to statistical models. To improve these estimates, we tested an approach that considers floods in relation to annual climate conditions—specifically, average, wet, and dry years—using daily streamflow data. We examined how well the Log Pearson Type III (LP3) distribution, a commonly used statistical model in flood frequency analysis, estimates flood levels when applied to these customized datasets instead of standard AMF data. Our study included over 70 years of data from 2028 basins across the United States, with drainage areas ranging from small (4.0 km²) to large (50,362 km²). We found that in some regions, LP3 better estimated frequent floods (recurrence interval of 2 to 25 years) when applied to AMF data. However, for less frequent, larger floods (recurrence interval of 50 to 200 years), the LP3 model worked better when applied to datasets representing wet or dry years. This approach could lead to more reliable flood predictions, which would benefit infrastructure planning and flood preparedness efforts.