Following wildfires, the probability of flooding and debris flows increase, posing risks to human lives, downstream communities and infrastructure, and ecosystems. In Southern California (USA), the Rowe, Countryman, and Storey (RCS) 1949 methodology is often used to rapidly estimate post-fire peak streamflows. RCS is an empirical method that predicts peak flow and is based on rainfall-runoff data collected prior to 1949 and thus does not reflect current hydrogeomorphic conditions and extreme weather events due to climate change in Southern California. This research estimated post-fire peak streamflows for 28 watersheds in Southern California. To evaluate model performance, observed streamflow data were compared to RCS predictions before and after wildfire. Where data were available, RCS was used to predict peak flows up to 70 years after fire, indicating a full recovery of the watershed and a return to pre-fire conditions. RCS pre-fire results yielded mean absolute percent error of 167% for the 2-year event and 41% error for the 10-year event. RCS post-fire results, without the use of any bulking factors, under predicted post-fire peak streamflows for 2-, 5-, and 10-year events for 16 of the 28 watersheds during the post-wildfire recovery period (4 of which occured in the Santa Ynez Mountains). Local physiography, land cover, geology, slope, aspect, rainfall intensity, and soil burn severity were incorporated into a calibrated random forest (machine learning) algorithm to characterize model performance across various watersheds. Analyses demonstrate that the RCS model may be overgeneralizing watershed processes and may not represent the spatial and temporal variability present in systems affected by wildfire. We propose that machine learning may be a helpful technique in characterizing post-fire response of small watersheds. For future work, we will use lessons learned from this study to select and test an existing process-based hydrologic model that can accurately predict post-fire peak streamflow.
Learn how machine learning can be used together with post-fire prediction tools
Understand benefits and limitations of Rowe, Countryman, and Storey (RCS) 1949 methodology
Learn important parameters that influence large post-fire flood events and some of the tools that can be used to model these events