Process-Based Flood Frequency Analysis in an Agricultural Watershed Exhibiting Nonstationary Flood Seasonality

Floods are the product of complex interactions of processes including rainfall, soil moisture, and watershed morphology. Conventional flood frequency analysis (FFA) methods such as design storms and discharge-based statistical methods offer few insights into process interactions and how they shape the probability distributions of floods. Understanding and projecting flood frequency in conditions of nonstationary hydroclimate and land use requires deeper understanding of these processes, some or all of which may be changing in ways that will be undersampled in observational records. This study presents an alternative process-based FFA approach that uses stochastic storm transposition to generate large numbers of realistic rainstorm scenarios based on relatively short rainfall remote sensing records. Long-term continuous hydrologic model simulations are used to derive seasonally varying distributions of watershed antecedent conditions. We couple rainstorm scenarios with seasonally appropriate antecedent conditions to simulate flood frequency. The methodology is applied in Turkey River in the Midwestern United States, a watershed that is undergoing significant climatic and hydrologic change. We show that using only 15 years of rainfall records, our methodology can produce more accurate estimates of present-day flood frequency than is possible using longer discharge or rainfall records. We found that shifts in the seasonality of soil moisture conditions and extreme rainfall in Turkey River exert important controls on flood frequency. We also demonstrate that process-based techniques may be prone to errors due to inadequate representation of specific seasonal processes within hydrologic models. Such mistakes are avoidable, however, and our approach may provide a clearer pathway toward understanding current and future flood frequency in nonstationary conditions compared with more conventional methods.