Rainfall-induced landslides are pervasive in hillslope environments around the world and among the most costly and deadly natural hazards. However, capturing their occurrence with scientific instrumentation in a natural setting is extremely rare. The prevailing thinking on landslide initiation, particularly for those landslides that occur under intense precipitation, is that the failure surface is saturated and has positive pore-water pressures acting on it. Most analytic methods used for landslide hazard assessment are based on the above perception and assume that the failure surface is located beneath a water table. By monitoring the pore water and soil suction response to rainfall, we observed shallow landslide occurrence under partially saturated conditions for the first time in a natural setting. We show that the partially saturated shallow landslide at this site is predictable using measured soil suction and water content and a novel unified effective stress concept for partially saturated earth materials.
 Shallow landslides, typically translational slope failures a few meters thick of unlithified soil mantle or regolith, can dominate mass-movement processes in hillslope environments [e.g., Caine and Swanson, 1989; Hovius et al., 1997; Trustrum et al., 1999]. Because these slides often mobilize into rapidly moving debris flows [Iverson et al., 1997], they present a hazard to human life, property, and activities [e.g., Sidle and Ochiai, 2006; Keefer and Larsen, 2007]. Regional operational forecasting of landslides and debris flows has generally relied on empirical correlations between rainfall and landslide occurrence [Keefer et al., 1987]. The accuracy of such predictions is in part a function of how well the initial moisture and stress conditions of hillside soils are known. Because the states of moisture and stress in a hillsope at a given time are the result of variably saturated and often hysteretic hydrologic processes acting over time scales from seasons to hours their quantitative estimation from precipitation measurements alone is often poor [Godt et al., 2006]. Analysis and prediction of stress conditions leading to landslides is a cornerstone of soil mechanics largely grounded on Terzaghi's  effective stress principle for saturated materials. This principle states that the controlling variable for the mechanical behaviour of earth materials is effective stress, defined as the difference between total stress and positive pore water pressure. However, this definition may not be appropriate for assessing the state of stress in hillslopes nor completely describe shallow failure of hillside materials under partially saturated conditions since it neglects the cohesive forces present in wet, but not saturated, soils and granular materials [e.g., Hornbaker et al., 1997; Mitarai and Nori, 2006].
 Instrumental monitoring of the pore-water response to natural and applied rainfall has been used to identify several hydrologic mechanisms that can lead to shallow landsliding. Formation of a shallow perched water table above a permeability contrast [Sidle and Swanston, 1982; Reid et al., 1988; Johnson and Sitar, 1990; Reid et al., 1997] and exfiltration of groundwater from bedrock [Montgomery et al., 1997] can initiate shallow landslides. Theoretical and analytical results from a variety of geologic and climatic settings have also advanced the hypothesis of shallow slope failure in partially saturated materials [Morgenstern and de Matos, 1975; Wolle and Hachich, 1989; Rahardjo et al., 1996; Collins and Znidarcic, 2004]. To test this hypothesis, Springman et al.  applied artificial rainfall to a steep slope in the Swiss Alps to induce a landslide and showed that the glacial moraine materials were nearly saturated at the time of failure. However, no suction measurements were reported for this experiment and the pore-water conditions were inferred from measurements of the volumetric water content. Thus, to date, no direct field evidence is available for landsliding under partially saturated soil conditions. The dominant interpretation for landslide failure surfaces that occur above the water table is that at some point during the transient infiltration processes, part of the soil becomes saturated and positive pore water pressures develop [Sidle and Swanston, 1982; Reid et al., 1988; Johnson and Sitar, 1990]. So, whether shallow landslides commonly occur under partially saturated conditions is controversial in part because of the lack of instrumental data on the hydrologic conditions within a naturally occurring shallow landslide.
2. Monitoring Observations
 Here, we report on instrumental observations from a coastal bluff in the Seattle, WA area where a shallow landslide occurred in the apparent absence of positive pore water pressures under partially saturated soil conditions. Shallow landslides are common on the coastal bluffs in this area during the wet winter season [Galster and Laprade, 1991] and typically happen during extended wet periods lasting several days [Godt et al., 2006]. The steep (>30°) 50 to 100 m high coastal bluffs in the Seattle, WA area (Figure 1a) are the result of Pleistocene age glaciation, wave attack at the shoreline, and mass movement processes [Shipman, 2004]. In such a geologic, hydrologic, and climatic setting, shallow landslides are generally less than a few meters thick and typically occur in the loose, sandy, colluvial deposits derived from the glacial and non-glacial sediments that form the bluffs [Galster and Laprade, 1991]. The instrumented hillslope (Figure 1b) is a steep coastal bluff with a thin (<2.0 m) colluvial cover located along a section of coastline frequently subject to shallow landsliding about 15 km north of Seattle [Baum et al., 2000]. The initial instrument complement was installed in September of 2001 and consisted of water content instruments and open-tube piezometers. No positive pore water pressures were observed in the hourly measurements collected by the piezometers during 2 years of operation and they were abandoned in September of 2003. Two water content profilers and two nests of 6 tensiometers (Figure 1d) were installed and fully functional on 15 October 2003. During the entire 16-month operational period, no positive pore pressures or volumetric water contents in excess 40% were recorded in the hourly data. Measurements of bulk density performed in the field and the lab indicate that the saturated moisture content or porosity of the colluvial material is about 40%.
 On 14 January 2006 a 25 m long by 11 m wide shallow slide occurred in colluvium adjacent to the instrument array with a translational failure surface inclined at 45° located between 1.0 and 2.0 m below the former ground surface (Figures 1c and 1d). The failure surface was apparently coincident with the contact between the loose sandy colluvium and the better consolidated glacial outwash sand (Figure 1c). The slide exposed a thin (<2 m) silt bed near the headscarp and a shallow zone (<0.8 m below the ground surface) of small diameter (<2 mm) blackberry and grass roots. Only a few larger diameter Alder tree roots penetrated the failure surface. The landslide deposit did not mobilize as a debris flow.
 Soil moisture conditions leading to the slope failure were captured with an instrument complement consisting of tensiometers and water content profilers designed to record hourly soil suction and water content variation due to rainfall [Baum et al., 2005]. From late September 2005 to the middle of January 2006 nearly 500 mm of rain fell at the site (Figure 2a). Soil saturation and suction at depths greater than 1.0 m began to respond after the rainy period in late October (Figures 2b and 2c). Additional rainfall in late November and early December on already wet soils caused abrupt increases in soil saturation and corresponding decreases in soil suction, which were delayed and attenuated with depth consistent with the characteristics of dominantly vertical infiltration. Soil saturation at 1.5 m depth increased to 0.95 and soil suction decreased to about 4 kPa (tensile pore water pressure) in response to rainfall prior to the landslide at the site on 14 January 2006. Throughout the entire observation period hourly measurements indicated that soil at all depths never reached full saturation and soil suction was never less than 2 kPa (Figures 2b and 2c).
3. Slope-Stability Analysis
 These unique field and monitoring observations allow us to quantitatively analyze the effects of partially saturated soil suction stress on slope stability. Recently, a unified concept for effective stress under both saturated and unsaturated conditions has been proposed and validated [Lu and Likos, 2006]. It provides insight into the stress variation resulting from changes in soil suction and water content during infiltration [Lu and Likos, 2004]. These transient stress variations have been shown to be of the same magnitude as those associated with true soil cohesion. The expanded effective stress σ for variably saturated soil [Bishop, 1954; Lu and Likos, 2006] is:
where σ is the total stress commonly provided by the self weight of soil, σs is defined as the suction stress characteristic curve of the soil with a practical functional form of [Lu and Likos, 2006]:
where uw is the pore water pressure, ua is the pore air pressure, θ is the volumetric water content, θr is the residual volumetric water content, θs is the saturated volumetric water content, and Se is the degree of saturation. We use the transient data collected for both water content θ and soil suction (ua–uw) to calculate suction stress by equation (2). A generalized factor of safety (FOS), defined as the ratio of shear strength to shear stress, for a one-dimensional infinite slope under both saturated and unsaturated conditions as a function of vertical depth, z, below the ground surface is given by:
where ϕ′ is the angle of internal friction, c′, the cohesion, β, the slope angle, and γ, the water-content dependent soil unit weight [Lu and Godt, 2008]. When the FOS of a hillslope is reduced to less than unity by infiltration and changes in soil suction and water content, landsliding is predicted. Infinite-slope stability analysis is appropriate for translational landslides in which the failure depth is relatively small compared to the landslide length. For the studied landslide, FOS obtained using equation (3) and those from 2-D limit-equilibrium analyses [Janbu, 1973; Baum, 2000] incorporating the effect of soil suction on shear resistance vary by only 1–3 percent. Suction stress, σs and factor of safety (FOS) were calculated for measured values of slope angle, β, of 45°, angle of internal friction, ϕ′, of 36°, cohesion, c′, of 1.1 kPa, including both true soil (0.9 kPa) and estimated root cohesion (0.2 kPa) [Schmidt et al., 2001], a saturated moisture content, θs, of 40%, and a residual moisture content, θr, of 5%.
 The temporal pattern of suction stress (Figure 2d from equation (2)) corresponds with the variations in soil saturation (Figure 2b) and soil suction (Figure 2c). An increase in the value of suction stress indicates a decrease in the shearing resistance of the soil or factor of safety. During late September - early November 2005 suction stress varied between −10 and −20 kPa, more than a factor of 10 greater than the measured true cohesion (0.9 kPa) for saturated hillside materials at the field site. In response to the heavy rainfall during the last week of October 2005 (Figure 2a) suction stress (Figure 2d) increased to less than −5 kPa at depths greater than 1.0 m by early November. The end of December was marked by the onset of a month-long wet period (Figure 2a). During this period saturation levels increased to maximums, soil suctions decreased to minimums, and suction stress increased in response to rainfall and infiltration. For depths less than 1.5 m, the factor of safety computed from equation (3) is never less than unity (Figure 2e). At 1.5 m below the ground surface the FOS was less than unity during 4 periods of a few to several tens of hours in late November and early December 2005 coincident with the reduction in suction stress. Beginning on 23 December 2005 the FOS decreased to less than unity for 5 periods of between 24 and 77 hours (Figure 2e). These surges of potential instability resulted from rainfall in excess of 25 mm (Figure 2e). During the first three periods in late November and early December when the FOS was less than unity, no landslides were reported along the bluffs from Everett to Edmonds, WA (Figure 1a). However, slides were reported in the Seattle area during the second period on 25 December. During the two periods of potential instability in January of 2006 several shallow landslides occurred on the bluffs near the field site. Instrumentation at the field site was damaged during the final period on 14 January 2006; the FOS at 1.5 m was less than or equal to unity for about 50 hours prior to the failure of several instruments, which indicated the occurrence of the landslide.
5. Concluding Discussion
 Field observations of soil-water content and soil suction, combined with an unsaturated slope stability assessment, indicate that the landslide occurred under partially saturated conditions. The spatial and temporal resolution of the instrumental observations is insufficient to completely eliminate the possibility of thin, transient zones of positive pore pressures during slope failure. However, assumption of small positive pore pressures (<0.1 kPa) at the failure surface yields FOS of less than 0.85 from both 1-D and 2-D limit equilibrium analyses. For the slope to remain stable under these assumed conditions additional soil strength must either be assumed or attributed to reinforcement by vegetation roots. We estimate that the few roots observed to penetrate the failure surface impart less than 0.3 kPa to the cohesive strength of the soil [Schmidt et al., 2001]. The effect of the surficial (<10 cm depth) root mat is not quantified, but it is unlikely that it provided adequate reinforcement to support widespread positive pore pressures acting on the failure surface.
 Empirical [e.g., Wilson and Wieczorek., 1995; Godt et al., 2006] attempts to identify periods when soil-moisture conditions are such that landslides are likely have met with limited success. We show that the transient reduction of suction stress during infiltration can be used to identify periods when stress and moisture conditions are conducive to landsliding when even modest rainfall can initiate slides. Application of the novel suction stress concept in a 1-D analysis captures the first-order effects of infiltration on the stress distribution and slope failure in the unsaturated zone and predicts the depth and timing, within a few days, of the shallow landslide at the site. Because landslide occurrence requires propagation of a failure surface [e.g., Fleming and Johnson, 1989; Muller and Martel, 2000], point measurements and 1-D analyses are insufficient to precisely identify landslide timing. Greater temporal accuracy may be possible if the 3-D stress dynamics were considered and if more frequent and spatially dense instrumental observations were available. During periods of potential instability the magnitude of the change in suction, water content, and thus suction stress is greatest in the instruments near the ground surface (Figures 2b–2d). This suggests that near-surface stress variations may influence landslide timing and size by controlling the failure process along the slide margins. As with the generation of positive pore-water pressures, and unlike the strength contribution from vegetation, variations in suction stress are transient over timescales relevant to landslide initiation and controlled by infiltration.
 The reduction of suction stress during infiltration may be a significant mechanism for shallow landsliding on over steepened hillslopes or where soils or regolith are underlain by permeable substrates. In both environments, shallow slope failure may occur prior to the formation of positive pore pressures. Permeable substrates, such as the glacial outwash sands of the Puget Lowland, marine sandstones in central Japan [Matsushi et al., 2006], or the weathered gneiss of Rio de Janeiro, Brazil [Wolle and Hachich, 1989], tend to inhibit the formation of shallow saturated zones and positive pore pressures. Human and other disturbances may also create over-steepened slopes in which the shear resistance of the soil mantle or regolith is inadequate to maintain stability under saturated conditions [Rao, 1996].
 Quantitative assessment of the influence of climate, tectonics, and land use on landslide hazard requires predictions of shallow landslide distribution, frequency, magnitude, and timing [Keefer and Larsen, 2007]. Consideration of the stress field in the unsaturated zone should improve forecasts of the timing and extent of landsliding. The prevailing methodology for landslide hazard assessment typically ignores stress changes in the unsaturated zone. Neglecting the contribution of soil suction in partially saturated soils to the stability of slopes and leads to overly conservative and inaccurate forecasts of landsliding. Incorporation of the novel concept for effective stress under partially saturated conditions can provide a new framework for reliable and accurate assessment of landslide hazard. Because landslide initiation is a highly transient process, prediction of the moment of their occurrence in the field may be impossible in practice. However, we show that this framework, implemented with monitoring of soil moisture conditions and rainfall can be used to identify periods when landslides can be expected and could be the basis for improved operational forecasts and warning of landslides.
 We thank J. McKenna and E. Harp for assistance with the installation and maintenance of the field instrumentation, A. Wayllace for geotechnical testing of the landslide materials, and M. Reid, W. Savage, and two anonymous reviewers for constructive comments. Burlington Northern Santa Fe Railroad provided support and access for the field monitoring.