Physical and maximum entropy models applied to inventories of hillslope sediment sources

Purpose The purpose of our study was to identify major hillslope sediment sources in a partially urbanized coastal watershed supporting salmonid habitat and to evaluate the use of physical and maximum entropy models in predicting sites of greatest concern. Questions include when and where increased runoff from trail and unpaved road surfaces has influenced patterns of landslides and gullies to a greater degree than what would be expected from background processes and controls, such as precipitation intensity, vegetation, soils, and slope characteristics. Materials and methods San Pedro Creek Watershed, USA, provides habitat for Oncorhynchus mykiss despite 33% of the watershed being urbanized. The watershed drains steep hillslopes with a median slope of 21°, with the steepest slopes on the 578-m North Peak of Montara Mountain. We inventoried hillslope sediment sources based on field surveys and aerial photographic interpretation in 1941, 1955, 1975, 1983, and 1997. We interpreted causative factors using precipitation records, geologic and soil mapping, digital elevation derivatives, land cover, and road/trail network changes and applied a physical landslide susceptibility model (Stability Index Approach to Terrain Stability Hazard Mapping (SINMAP)) for hillslope stability and a maximum entropy model for assessing gully and landslide centroids. Results and discussion Maps of landslide and gullies reveal an association with land use changes over time. Agricultural land uses led to the development of extensive gullies in parts of the watershed, and some of these continue to contribute significant sediment to the stream system; others were built-over in residential developments. The most significant remaining gullies result from impervious runoff from roads built into steep hillslopes. Although the best single predictor of landslide susceptibility is physically modelled hillslope stability (SINMAP), slope equally contributed to multivariate MAXENT models (area under the receiver operator characteristic curve (AUC) = 0.74 in 1941, 0.65 in 1975, and 0.79 in 1983). Other covariates in the maximum entropy models include plan curvature, trail distance in 1975, geology in 1983 (favoring colluvium), and vegetation. Conclusions Combining physical hillslope stability with a maximum entropy model appears promising, although overall slope angle also contributed equally. Landslides are episodic and linked to major precipitation/runoff events, such as ENSO events in 1962, 1