A New Framework for Generating Stochastic and Representative River Hydrographs for Long-term Simulations

Understanding the effects of climate change, particularly from a fluvial geomorphic perspective requires simulations of river behavior over hundreds or thousands of years. A fundamental input for such simulations is long hydrographs, which are elusive to the research community due a limited period of observed data. To overcome this, researchers instead rely on a variety of artificial hydrographs, including constant or various shapes of cyclic hydrographs. However, this approach does not faithfully reproduce the statistical characteristics of a natural hydrograph. In this study, we overcome this obstacle by generating a long series hydrograph, developed to conserve the statistical and stochastic characteristics of the observed hydrograph. This is accomplished through a combination of a daily precipitation generator with a finer-scale disaggregation technique. A stochastic precipitation generator (the Weagets Matlab toolbox), combined with a modification step to increase the generating accuracy, especially for flashy events, was tested to produce a long series of daily precipitation events. This series is disaggregated from daily to hourly values (using the HyetosMinute R package), conserving the sub-daily ratios of each month. The calibration of disaggregated process was improved through implementing additional criterion to effectively capture the hourly peaks. Finally, the Soil Moisture Accounting method, a continuous rainfall-runoff transformation method, is used to generate the corresponding hydrograph (using HEC-HMS). Observed streamflow series and satellite precipitation data are used to build the transformation model, where the validation results show high performance in predicting the flow duration curve of the hourly observed runoff. We use this framework to generate a three-hundred-year hydrograph for Ninnescah River in Kansas under steady state climatic conditions. The overall accuracy of the model is evaluated based on the comparison between the generated series (which is split into ten samples) against the observed at various percentiles, which showed high performance. By accurately generating long hydrographs we can better account for climate change induced alterations to hydrographs at longer time scales, aiding in predicting future modifications to fluvial landscapes.