UTILIZING REPEAT AIRBORNE LIDAR FOR PREDICTION OF POST-FIRE SEDIMENT YIELD FROM SMALL STEEP DRAINAGES

In southern California, where debris flows following wildfires are a common occurrence, models have been developed by the USGS to predict sediment volumes issued by runoff-generated debris flows from recent burn areas. However, recent work indicates that these models may be less accurate in very steep terrain in this region that is susceptible to post-fire dry ravel loading, necessitating further work. To assess these models and develop additional predictive models, we utilized repeat airborne lidar of two steep burn areas in southern California timed to capture dry sediment loading of steep low-order channels and subsequent channel scour by debris flows and floods available for one of the sites. We used these data to estimate yields for 454 small catchments to train a random forest sediment yield model using 9 metrics related to slope, sediment supply, vegetation, burn severity, and rainfall. Watershed slope, pre-runoff dry ravel, and fire history were ranked as the three most important predictors. Using a validation dataset of 112 catchments, we found that this random forest model more accurately predicted sediment yields than the existing USGS volume model. Additionally, multitemporal lidar differencing revealed widespread reductions in sediment supply in channels following initial erosion events by two rain storms. However, soil erosion and channel-adjacent mass wasting continued to fuel small debris flows and floods observed later in the wet season. This indicates continued hazards of sediment-bulked flows despite reductions in channel sediment availability in the first year following fire, as has been shown in other studies. With increasing coverage of baseline lidar topography in the western US, we recommend similar hydrologic and geomorphic change detection monitoring be employed in mountainous regions experiencing large increases in fire where predictive models are needed for hazard management.