First, let us describe the processes we are attempting to model and its geomorphic setting in the landscape. Much of the hilly lands of mountainous landscapes is covered with a loose soil mantle of variable thickness. Typically the boundary between the soil (the solum or O, A, and B horizons) and the underlying variably weathered bedrock is abrupt. The soils are commonly rocky, have low bulk density, and lack significant soil cohesion when high in water content, whereas the underlying bedrock is commonly fractured and has considerable cohesive as well as frictional strength. Although the soil is generally more conductive than the bedrock, we and others have found that the underlying bedrock is commonly highly fractured and may conduct large amounts of storm flow (e.g. Wilson and Dietrich, 1987; Montgomery et al., 1997, Johnson and Sitar, 1989). Overland flow is absent on the very steep highly conductive soils, but saturation overland flow can be significant in the lower gradient unchanneled valleys (Wilson and Dietrich, 1987; Dietrich et al., 1993; Montgomery and Dietrich, 1995). Horton overland flow may develop after intense fires in the more arid parts of the West due to fire induced hydrophobicity.

In the absence of overland flow, downslope soil transport is largely due to slope dependent processes of biogenic transport, creep and ravel. Except in those areas where glacial and periglacial processes are active or have dominated during times of glacial advances, the spatial pattern of soil thickness largely reflects the local balance of production from underlying bedrock and the net erosion or deposition due to downslope transport of debris (e.g. Dietrich, et al., 1995; Heimsath et al., 1997). Because of the slope dependent transport, any valley, hollow, swale, or other even subtle indentation into the hillside will be a site where soil transport will converge and cause net soil accumulation if there is no channel present to remove the converging soil. Ridges and noses are the mirror image of this, with what some call "divergent" transport on the ridges -transport in all directions away form the ridge- tending to keep this soils thin there. We have proposed a model that combines a soil production model with a transport model to predict the spatial pattern of soil thickness (Dietrich et al., 1995). This model correctly predicts the observed tendency for soils to be thick in the unchanneled valleys and thin on ridges. Hence, the soil mantle is a mobile, highly conductive layer of colluvium which varies in thickness in a relatively systematic way across the landscape.

Hilly landscapes are dissected into a branching network of valleys along which runoff and sediment transport is concentrated (Figure 1). Not all valleys have channels: typically the steepest branches, which appear as subtle swales into the hillside, do not. Figure 2 illustrates the relationship between valleys and channel networks (mapped in the field) in four different landscapes. The tips of the channel network almost always terminate at the downstream end of an unchanneled valley (or source area). Convergent slope-dependent soil transport will tend to cause the soil mantle to progressively thicken in the unchanneled valleys. This topography also focuses shallow subsurface storm flow towards the axis of unchanneled valleys. Even if storm runoff travels in the underlying bedrock (which it commonly does), the rapid decrease in conductivity with depth into the bedrock will perch runoff in the near surface and the head gradient driving water off the landscape will largely be determined by the elevation potential, i.e. the hillslope gradient. Hence, the surface topography gives a good indication of where storm water will concentrate, and unchanneled valleys (the axes of which have been referred to as hollows) will be sites of elevated water levels due to convergent subsurface flow. Similarly the topographic ridges will be places of divergent subsurface flow and elevated water in the soil is only likely to occur due either to local very intense rainfall or local bedrock heterogeneities that force subsurface flow in the bedrock up into the overlying soil mantle.

In such a landscape of soil mantled ridge and valley topography, shallow landslides typically only involve the soil mantle and commonly occur at or near the soil-bedrock boundary. These landslides may mobilize and travel a short distance downslope before coming to rest either still on the hillside or in a nearby channel. Other landslides may mobilize into a debris flow and enter a channel at sufficiently high momentum and on a sufficiently steep slope that they travel a great distance down the channel network, commonly scouring the channel to bedrock and depositing a massive amount of sediment downstream (e.g. Dietrich and Dunne, 1978; Pierson, 1977; Benda and Dunne, 1997). These debris flows typically originate in unchanneled valleys, at the tips of the channel network. Landslide maps in such environments commonly show that the majority of shallow landslides occur in steep, convergent (unchanneled valleys) parts of the landscape (e.g. Reneau and Dietrich, 1987) (Figure 2b).

Dietrich and Dunne (1978) and Dietrich et al. (1982) proposed that steep unchanneled valleys undergo a cycle of colluvium accumulation punctuated by periodic discharge due to landsliding (Figure 3). In effect, that is how the steep unchanneled valleys "work". They argued that even with a constant forested canopy, as the soil progressively thickened in the hollow axis, the effectiveness of root strength would diminish and eventually make the site much more susceptible to failure during intense storms. Reneau et al. (1984, 1991, 1993) and Reneau and Dietrich (1989, 1990) reported radiocarbon dates from basal colluvium at sites throughout the Pacific Coast Range which suggest that colluvium in hollows may accumulate for thousands of years and that the timing of landsliding in the natural landscape may be influenced by vegetation change induced by climatic variability (including climatically-induced changes in the frequency of forest fires). Benda and Dunne (1997) have emphasized the importance of stand replacing fires in reducing root strength and in making these sites vulnerable to storms. Other important contributions regarding slope stability processes and modeling associated with ridge and valley topography can be found in Dunne (1990), Sidle (1992), Wu and Sidle (1995), Duan (1996), Hsu (1994), Okimura and Ichikawa (1985), and Okimura and Nakagawa (1988), to mention a few most relevant works.

The picture that emerges from this work on shallow landslides is that surface topography has a great bearing on the location and frequency of shallow landsliding. Importantly, it is not just the local slope that matters, but also the curvature of the topography and how it focuses or spreads runoff downslope. A physically-based model that quantifies the influence of surface topography on pore pressure in a shallow slope stability model may effectively capture the essential linkage between topography and slope instability. With this linkage, a general digital terrain based model can then be built which takes advantage of digital elevation data and fast computing that is now available. Such a model would predict that flat areas are stable, that ridges (with divergent subsurface flow) may be steep enough to fail but require unusually large storms to generate instability, and that steep unchanneled valley axes (hollows) require the smallest rainstorms to fail (because of the convergent subsurface flow) and are therefore most susceptible to increased instability due to environmental change (such as clear cutting).