Improving post-fire hydrologic predictions using geochemical data and coupled model-optimization systems in Devil Creek, CA

Wildfires and their hydrologic consequences pose serious hazards across the western United States. Post-fire flash flooding and debris flows are destructive to property and cost the lives of downstream populations. Field studies, geochemical analyses, and hydrologic modeling were used to investigate the hydrologic response following fire in the San Bernardino Mountains in Southern California. Field studies show the development of a hydrophobic layer which increased the overland flow component in the burned Devil Creek. The physical and chemical transformations also affect the hydrologic model predictions increases the parameter uncertainty and model error. Watershed chemistry is utilized to characterize the subsequent change in hydrologic processes and to explore the mixture of representative streamwater components through endmember mixing analysis (EMMA). Results from the EMMA-based hydrograph separation are used to optimize parameters within a conceptual model. This is achieved by isolating model flow components and calibrating related parameters to contributing portions of the hydrograph. Current operational forecasting by the National Weather Services includes the use of the Sacramento Soil Moisture Accounting (SACSMA) model for predictions of streamflow under post-fire conditions. The SACSMA model is coupled with commonly-used optimization techniques, including the Multi-step Automatic Calibration Scheme (MACS), the Shuffled Complex Evolution Metropolis (SCEM) and the Generalized Likelihood Uncertainty Estimation (GLUE) algorithms. The coupled model framework was integrated with estimates of geochemically-derived flow components. Success is determined not only by the accurate prediction of total discharge, but also by the prediction of flow from contributing sources (i.e. overland, lateral and baseflow components). Post-fire hydrologic modeling also faces issues of limited data: short-tern data and data gaps. The most efficient sampling strategy associated with commonly-used optimization techniques must be known in order to improve sampling campaigns in the future. The most efficient sampling strategy is determined using semi-synthetic data. Various sampling strategies were explored in order to evaluate sufficient model validation data to reduce parameter uncertainty. The semi-synthetic data was based on real data collected in Devil Canyon watershed from both pre- and post-fire periods, and interpolated into daily, weekly, event base sampling, random sampling at 50% and 25% temporal resolutions.