Keywords

1 Introduction

This study models landslide susceptibility on a watershed scale and estimates soil loss from landslide deposit sediment impacting Strahler 3rd Order and higher streams in the Drift Creek/Siletz watershed. Landslides tend to increase sediment bedload resulting in loss of pools, and/or fish habitat, habitat connectivity, and diversity. Suspended landslide sediment increases temperature, turbidity, and exceedance of Total Maximum Daily Load (TDML) targets established by the United States Environmental Protection Agency (EPA) and the DEQ. In addition to providing habitat and diversity, the streams in the Drift Creek/Siletz watershed are a source of drinking water for the Confederated Tribes of Siletz Indians and Lincoln City.

Sediment influxes, possibly associated with landslide activity, are suspected by the DEQ of decreasing water quality, affecting water treatment plant operations, and impacting ecology in the Drift Creek/Siletz watershed. During this study, we evaluated landslide susceptibility and estimated soil loss for landslide deposits mapped in a previous study (Korte and Shakoor 2017). In the previous study, the Oregon Department of Geology and Mineral Industries (DOGAMI) provided light detection and ranging (lidar) imagery and the mapping methodology (Burns and Madin 2009). In this study, lidar DEM derived slope angle, curvature (plan, profile, and standard), flow direction, landslide deposit aspect, USDA NRCS (2017) sourced soil erodibility factors and USGS MRLCC (2017) land cover were used to calculate the probability of landslide occurrence and estimate potential sediment delivery to streams. Precipitation was not used in sensitivity analysis because there are no recording stations in this watershed and interpolation from nearby stations results in uniform precipitation across the study area.

The Drift Creek/Siletz watershed does not have active USGS stream monitoring stations. The Schooner Creek station (14303950), located near the northwest boundary of the Lower Drift Creek watershed, monitors only flow. The Confederated Tribes of Siletz Indians have monitored Boulder Creek, Warnick Creek, and the North Fork Siletz River for temperature and Cedar, Euchre, and Sams Creeks for both temperature and turbidity. Temperature, turbidity, flow and stream geometry are needed to draw any meaningful water quality correlations with the streams in the Drift Creek/Siletz watershed. Therefore, this study will estimate soil loss to determine the most likely areas of stream impact by landslide-derived sediment.

1.1 Problem and Significance

The United States Forest Service (USFS) has designated the Drift Creek/Siletz watershed as a Tier 1 Key Watershed. A Key Watershed is defined by the National Forest and Bureau of Land Management District as a watershed containing: (a) habitat for potentially threatened species or stocks of anadromous salmonids or other potentially threatened fish, or (b) greater than six square miles with high-quality water and fish habitat. Coho Salmon (Oncorhynchus kisutch) and Steelhead Trout (Oncorhynchus mykiss) populations are severely depressed in Drift Creek watersheds. The watershed is listed by the Oregon Department of Fish and Wildlife (ODFW) as a “source area” for Steelhead. The MidCoast Watersheds Council has designated the watersheds as a high priority for protection and enhancement.

The DEQ is concerned about the TDML (the maximum amount of a pollutant that a waterbody can receive and still meet water quality standards and an allocation of that load among various sources of that pollutant) found in the watershed (DEQ 2013; Michie, Personal Communication 2014). Suspected sediment influxes of unknown origin are changing the ecology of the watershed by stressing fish habitat through increasing turbidity, increasing conductivity, increasing temperature, reducing stream flow, decreasing dissolved oxygen concentrations, and decreasing general water quality. The pH of Devil’s Lake has exceeded alkaline standards during winter/spring seasons since 2003 (Salmon-Drift Creek Watershed Council 2006). In addition, the Confederated Tribes of Siletz Indians obtain their drinking water from the watershed. Water treatment operations in the watershed are sometimes shut down due to turbidity resulting from excessive sedimentation, particularly below 3rd order (Strahler) streams (van de Wetering, Personal Communication, 2014). The DEQ needs to know if these sediment influxes are the result of natural or anthropological phenomena to properly plan and direct funding for remediation. The DEQ and the Oregon Department of Mineral and Geologic Industries (DOGAMI) suspect that landslides may be responsible for some of the deterioration of the Drift Creek/Siletz watershed (Michie, personal communication 2014).

1.2 The Drift Creek/Siletz Watershed

The Drift Creek/Siletz watershed is a combination of 2 Hydrologic Unit Code 6 watersheds, Upper and Lower Drift Creek, located within the Siletz watershed. The area of the Drift Creek/Siletz watershed is approximately 107 km2. Elevation ranges from 1 to 975 m and slope angles range from 0° to 85°. The area is heavily dissected with no preferred direction of ridgelines. Major streams trend from northeast to southwest and east to west. The 30-year annual average precipitation in the watershed is 310.6 cm; about half is rainfall and half snowfall. Fall and winter are the wettest seasons. Autumn receives 83.1 cm and winter receives 140 cm. Spring receives 72 cm of precipitation and summer is the driest season with 15.5 cm precipitation (National Weather Service 2016). Rainfall and snowfall are not differentiated by season in the in the precipitation records. However, the overall marine temperate climate and seasonal precipitation concentration leads to most landslides occurring during the winter months (Smith 1978) as bedrock erosion increases and soil strength decreases due to the flow and infiltration of water.

Douglas fir, western red cedar, western hemlock, big leaf maple, and red adler are abundant in the watershed. The understory and floor are covered with ferns and shrubs. The watershed is in its 3rd or 4th forest harvest (van de Wetering, personal communication 2014). Logging is currently active in the eastern region of the watershed (Upper Drift Creek). Foresters are required to reclaim logged areas with conifers per the Forest Practices Act (Oregon Department of Forestry 2014). Some of the logging bounds and/or crosses Strahler 3rd order or higher streams.

1.2.1 Geology

The watershed has variable lithology consisting primarily of Siletz River Volcanics and Tyee Formation, with minor areas of Yamhill and Nestucca Formations (Fig. 1). The watershed is heavily forested and geomorphic expressions of landslides are difficult to interpret from aerial orthophotographs. Temporal landslide data is limited because there is only one detailed digital elevation model (DEM) and little observational history.

Fig. 1
figure 1

Bedrock units for the Drift Creek/Siletz Watershed

Full size image

The dominant bedrock in the watershed is Siletz River Volcanics (74.6 km2) and the Tyee Formation (16.1 km2). The Yamhill Formation, Nestucca Formation, Intrusive Formations, Depot Bay Basalt, Alsea Formation, and alluvial deposits (listed in order of age) make up the remaining 107 km2 area. The Yamhill Formation, Nestucca Formation, and Alsea Formation are generally constrained to the eastern and western edges of the watershed. These units have eroded away from the interior of the watershed, leaving exposures of Siletz River Volcanics as the dominant bedrock by area.

The Siletz River Volcanics consist of aphanitic to porphyritic vesicular pillow flows, tuff-breccias, massive lava flows, and sills of tholeiitic and alkalic basalt (Fig. 2). The upper part of the sequence contains numerous interbeds of basaltic siltstone and sandstone, basaltic tuff, and locally derived basalt conglomerate. Zeolitization of the rocks and presence of calcite veins are pervasive within the unit (USGS MROSD, 2017). These rocks are of marine origin and have been interpreted as oceanic crust and seamounts, an off-shore terrain named “Siletzia”, that was accreted onto the Oregon coast 55–50 million years ago (Wells et al. 1998). Siletzia basalts (Siletz River Volcanics) form the basement of the Oregon Coast Range (OCR) and therefore compose the core of the OCR anticline. The age of the Siletz River Volcanics spans the Paleocene to Eocene Epochs.

Fig. 2
figure 2

Siletz River Volcanics showing faint pillow basalt features from Korte and Shakoor (2017)

Full size image

The Tyee Formation has been interpreted as a deep-water submarine fan complex that formed on top of the Siletzia terrane in a fore-arc basin 49–48 million years ago. The fore-arc basin was created by the collision of Siletzia with the North American continent. Later uplift of the OCR displaced the fore-arc basin to what is now the Williamette Valley which separates the Cascades from the OCR (Wells et al. 1998) and tilted the Tyee Formation slightly (<20°) to the east. Groove and flute casts indicate deposition by north-flowing turbidity currents. The proximal end of the fan complex is to the south and the distal end to the north. Therefore, the Tyee Formation is thickest to the south and thinnest in the north. The Drift Creek/Siletz watershed is at the very distal end of the fan complex, the end of the Tyee Formation. The Tyee Formation in this watershed is a sequence of thin rhythmically bedded, medium to fine grained micaceous, feldspathic, lithic, or arkosic marine sandstone and micaceous, carbonaceous siltstone containing minor interbeds of dacite tuff in the upper part (Fig. 3). Probable provenance of the unit is the southwest Idaho batholith (Heller et al. 2016). The Tyee Formation is Middle Eocene in age.

Fig. 3
figure 3

Tyee formation showing interbedded marine sandstone and micaceous carbonaceous siltstone from Korte and Shakoor (2017)

Full size image

In addition, there are 37 different soil types derived from extrusive, intrusive, and sedimentary bedrock parents mapped in the Drift Creek/Siletz watershed. The top five soil types with landslide deposits vary in depth to bedrock from 51 to 203 cm (USDA NRCS 2017). Runoff potential varies from moderately low to moderately high. Erodibility varies from moderate to high. The textures of these top five soils are silt loam—very gravelly loam, silt loam—gravelly loam, and gravelly silt loam. Grains > 0.075 mm classify as SM for Siletz River Volcanics and SP-SM for the Tyee Formation. Fines classify as ML for both Siletz River Volcanics and the Tyee Formation per USCS (ASTM D2487) classification. The average slope angle in the watershed is 25°.

2 Methods

2.1 Landslide Susceptibility Map

A landslide inventory map for the Drift Creek/Siletz watershed was prepared using the protocol for inventory mapping of landslide deposits from lidar imagery developed for DOGAMI by Burns and Madin (2009). The protocol was developed to provide consistent methodology for collecting and visualizing data, to minimize mapping errors, and to enhance the reliability of landslide hazard maps (Ardizzone et al. 2002; Haneberg 2006; Burns and Madin 2009). The inventory map was created by manually digitizing 570 landslide deposits, associated scarps and scarp flanks, and fans at 1:24,000, 1:10,000 and 1:4000 scales, over an area of 107 km2 (Korte and Shakoor 2017). Inventory mapping was limited to slope movements within 30 m of a third order or higher stream channel which is where sediment issues have been problematic (van de Wetering, personal communication 2014). The database model used to create the inventory map was sourced from DOGAMI and the Oregon Geospatial Enterprise Office. The inventory map was verified where possible on the ground through direct observation and serves as the basis for the landslide susceptibility map.

A new database for the landslide susceptibility map was created by adding the DEM derived variables of curvature (plan and profile), slope aspect, terrain elevation, flow direction, soil erodibility factors, and land cover designations to the DOGAMI database. Plan curvature is defined as curvature perpendicular to slope, profile curvature is the curvature in the direction of slope. Soil type and erodibility were sourced from a Custom Soil Resource Report for Lincoln County, Oregon generated by the United States Department of Agriculture Natural Resources Conservation Service and added to the database (USDA NRCS 2017). Land cover was sourced from the National Land Cover Database 2011 and added to the database (USGS MRLCC 2017).

Statistical methods have been shown to accurately predict the likelihood of landslide occurrence e.g. (Haneberg 2004; Guzzetti et al. 2005; Guzzetti 2006; van Westen et al. 2008). The data set in this study includes non-normally distributed data, continuous data, categorical data, and binary data (presence or absence of a landslide). Logistic regression analysis can accommodate non-normally distributed data, continuous data, categorical data, and binary data (Costanzo et al. 2014; Budimir et al. 2015; Zhang et al. 2016; Raja et al. 2017). A logistic regression model would use occurrence of a landslide as a binary dependent variable. For example, a value of (1) would indicate presence of a landslide and a value of (0) would indicate absence of a landslide. Values between (0) and (1) indicate the likelihood (probability) of landslide occurrence. Independent variables in the model are the factors contributing to landslides based on the data from the study area. Logistic regression calculates coefficients (weight factors) and significance (p) of each independent variable through an iteration process where each independent variable is evaluated on its own, in combination with other independent variables, and with all independent variables. The significance level defined for each individual independent variable in this study is 0.05. The null hypothesis is that each independent variable has no effect on the overall model. If p < 0.05, then the null hypothesis is rejected, and the variable is significant to the model. The process used in this study then creates an equation that maximizes the area under a Receiver Operating Characteristic (ROC) curve, which is a plot of sensitivity versus specificity (Eng 2005). The maximum area under the curve is 1. The model with area under the ROC curve closest to 1 was chosen as the most valid model. Quantitatively the regression equation can be written as:

$$ P = 1/\left( {1 + {\text{e}}^{ - z} } \right) $$
(1)

where P is the probability of landslide occurrence and z is the combination of coefficients and independent variables responsible for landslide occurrence. The parameter z can be written as:

$$ z = \beta_{0} + \beta_{1} V_{1} + \beta_{2} V_{2} + \cdots + \beta_{n} V_{n} $$
(2)

where \( \beta_{0} \) is a constant and \( \beta_{n} \) is the coefficient for each respective independent variable \( V_{n} \). The parameter z ranges from −∞ to ∞; therefore, the range of P is from 0 to 1. Landslide non-occurrence was added to the database by randomly selecting 570 points within 30 m of stream channels where landslides did not occur. These points were then tagged with the same attributes as landslide occurrence. The process tests for P = 0 and P = 1 while generating coefficients \( \beta_{n} \) and significance of the independent variables in the model. Variables with high significance (p < 0.05) were then used in the model which maximized the area under the ROC curve. The model was then rasterized to create a 10 m2 resolution landslide susceptibility map of the study area. 10 m2 pixel size was chosen because the mean area of the landslide deposits in the landslide inventory is 150 m2 and none of the landslide deposits mapped were less than 10 m2 in area. For mapping purposes, susceptibility was classified as Very High (P = 0.76–1), High (P = 0.51–0.75), Medium (P = 0.26–0.50), and Low (P = 0.01–0.25).

2.2 Estimating Average Annual Soil Loss

Soil loss in the Drift Creek/Siletz watershed was calculated by using the U.S. Department of Agriculture RUSLE 2 equation in ArcGIS (USDA ARS 2017). The equation calculates average annual soil loss in tons per acre per year. ArcGIS was used instead of the United States Department of Agriculture National Resource Conservation Service (USDA NRCS) RUSLE 2 v 2.6.8.4 program because of the variable soil erodibility factors and land cover designations in the watershed. This allowed for more accurate estimation of soil loss by accounting for spatial variations of these variables. The RUSLE equation is:

$$ A = R * K * Ls * C * P $$
(3)

where A = average annual soil loss, R = rainfall-runoff erosivity factor, K = soil erodibility factor, Ls = slope length and steepness factor, C = cover management factor, and P = support practice factor.

The R factor is measure of the erosive force and intensity of rain in a normal year. R was calculated using the U.S. EPA rainfall erosivity factor calculator for a 30-year period from January 1, 1987 to October 20, 2017 (R = 4260) and for January 1, 2016 to December 31, 2016 (R = 151) for Lincoln County, Oregon to approximate rainfall erosivity in the watershed because the calculator does not use watershed shapes (polygons) for input extent.

The K factor varies by soil type. There were 37 different K factors in this study. K factors are included in custom soil reports from the USDA NRCS but not included in the NCRS SSURGO databases used in GIS mapping. A table of the soil types and respective K factors was created and joined to the SSURGO database.

The Ls factor quantifies the combined effect of slope length and steepness. Ls factors for different slope angles and slope lengths can be found in summarized tables or calculated from the lidar DEM. For this study, the Ls factors for each 1 m2 pixel were calculated (Pelton et al. 2007) because of the many different slope lengths and slope angles in the study area. The equation for Ls uses a flow accumulation (“flowacc”) and slope (“slope”) raster multiplied together. The individual equations for L and s are:

$$ L = \left( { ` ` {\text{flowacc''}}*\left[ {resolution} \right]/22.1} \right) ^{0.4} $$
(4)
$$ s = (\sin ( ` ` {\text{slope''}}*0.01745)/0.09)^{1.4} *1.4 $$
(5)

The C factor reduces the soil loss estimate according to how effective vegetation is at preventing detachment and transport of soil particles. C factors were rasterized to 10 m2 using a table generated from RUSLE 2 guidelines for the state of Oregon and land cover data sourced from the Multi-Resolution Land Characteristics Consortium National Land Cover Database 2011 (NLCD 2011).

The P factor is the ratio of soil loss with a given surface condition to soil loss with up-and-down hill plowing. P is greater than 1 for compacted conditions and less than one for loose conditions. The control condition, P = 1, was used in this study.

Estimated average annual soil loss was then calculated and mapped at 10 m2 resolution over a 30-year period and for year 2016. Overall mean soil loss statistics were calculated for the entire watershed. The relationship between estimated average annual soil loss, mapped landslide deposits, and a 30 m stream buffer was investigated by intercepting the respective map layers and using zonal statistics to quantify how much soil sediment is delivered to the stream network by landslide deposits (Fig. 4).

Fig. 4
figure 4

Zone of landslide deposits used for average annual soil loss estimates

Full size image

3 Results

3.1 Landslide Susceptibility

Table 1 summarizes the significance of the independent variables from most significant to least significant used in the logistic regression model.

Table 1 Results of independent variable significance in the logistic regression model
Full size table

Running the model using only the independent variables with p < 0.05 (the most significant contributors to landslide occurrence) maximizes the area under the ROC curve and calculates the parameter z as:

$$ \begin{aligned} z & = - 15.383 + \left( {0.096* Slope Angle} \right) + \left( {0.003*Terrain Elevation} \right) \\ & \quad + \left( {36.536*Soil Erodibility Factor} \right) + \left( { - 0.063*Plan Curvature} \right) \\ \end{aligned} $$

The area under the ROC curve is 0.8761. For scale, a value of 1 would be a perfect prediction and a value of 0.5 would be random chance. Adding insignificant variables to the model only increased the area under the ROC curve by a maximum of 0.0033. The model was exported as an Excel spreadsheet which allows sensitivity analysis by varying the values of all the independent variables and observing the effect on predicted chance of landslide occurrence.

The equation \( P = 1/\left( {1 + {\text{e}}^{ - z} } \right) \) was used to create a landslide susceptibility map thematically displaying the chance of landslides occurrence (Fig. 5). Thirty-three percent of the watershed has a high to very high probability of landslide susceptibility. The high and very high probability areas of landslide occurrence are primarily above 390 m terrain elevation, with slope angles >25° (which is above the 25° average slope angle for the watershed), along stream channels where landslide deposits were mapped, logging areas and along roads leading to those logging areas.

Fig. 5
figure 5

Landslide susceptibility map of the Drift Creek/Siletz watershed showing probability of landslide occurrence

Full size image

3.2 Estimated Average Soil Loss

The estimated average annual soil loss over a 30 period was 27 tons/acre/year over the entire extent of the watershed. The estimated average annual soil loss for 2016 was 28 tons/acre/year. The result of using the RUSLE equation in this study to estimate average annual soil loss is within 5% of Milliman and Syvitski’s (1992) result of 29.5 tons/acre/year for a watershed of similar area and physiographic location. However, watershed scale soil loss can only be used to characterize the watershed. Sediment yield increases with the size of a watershed. Larger watersheds generate more sediment than smaller watersheds. Specific sediment yield (average annual soil loss, tons/acre/year) decreases with increasing watershed size because of the increase in watershed area.

The average annual soil loss model in this study allows examination of small area/large scale areas of interest. This makes the soil loss model particularly valuable because it can be used for a number of different scenarios. Average annual soil loss anywhere in the watershed, at any distance from the stream network, at any specific landslide deposit, any specific land cover condition, near any road, etc., can now be easily estimated in the Drift Creek/Siletz watershed.

Figure 6 shows the estimated average annual soil loss per landslide deposit. There were 4 landslides out of 570 that individually lost greater than 15 tons/acre/year. They are located within recent logging areas in Upper Drift Creek. The soil types in these areas have soil erodiblity factors that range from 0.37 to 0.4, which are among the highest in the watershed. The estimated average annual soil loss within 30 m of the stream network for these 4 landslides combined is 281.5 tons/acre/year. Other than previous logging, none of these four high soil loss landslides show any nearby human related activity.

Fig. 6
figure 6

Estimated average annual soil loss per landslide deposit

Full size image

The area of the landslide deposits within 30 m of the stream network in Upper Drift Creek is 128 acres and the total estimated annual soil loss for those 128 acres is 65 tons/acre/year. Lower Drift Creek has 85 acres of landslide deposits within 30 m of the stream network and the estimated annual soil loss is 29 tons/acre/year. The soil loss estimates show that landslide deposits contribute substantially more sediment to streams in Upper Drift Creek where there is more logging activity than in Lower Drift Creek.

4 Conclusions

This research was undertaken to provide a better understanding of landslide susceptibility and sediment loading to Strahler 3rd order or higher streams in the Drift Creek/Siletz watershed. The two dominant rock types, Siletz River Volcanics and Tyee Formation, have low strength and weather into poorly graded sandy and silty soils (Korte and Shakoor 2017). While this study cannot determine the effects of landslide sediment on water quality and ecological habitat, it does advance our knowledge of where landslides are likely to occur and estimates how much soil loss occurs in the Drift Creek/Siletz watershed.

The landslide susceptibility model exhibited a high degree of statistical validity. In addition, the model highlighted areas away from the stream network where landslides appear in the lidar DEM. Therefore, it will be useful in mapping a watershed wide landslide inventory. The most important omission in the model is water. There is no detailed precipitation variation data across the 107 km2 area of the watershed. Also, there have been no pore pressure studies in the study area. Precipitation and pore pressure could be used to further refine the regression model.

The Revised Universal Soil Loss Equation was used as a mapping technique to accommodate 37 different soil types and 13 different land cover factors. Upper Drift Creek had the highest average annual soil loss estimates in the Drift Creek/Siletz watershed. Current logging activity is prevalent to Upper Drift Creek. Logging activity, including roads and the time needed to re-grow logged areas appears to substantially affect soil loss in Upper Drift Creek by compacting soils, loading slopes, and removing established vegetation.