Development and testing of a snow interceptometer to quantify canopy water storage and interception processes in the rain/snow transition zone of the North Cascades, Washington, USA


Corresponding author: K. A. Martin, Department of Civil and Environmental Engineering, Box 352700, University of Washington, Seattle, WA 98195-2700, USA. (


[1] Tree canopy snow interception is a significant hydrological process, capable of removing up to 60% of snow from the ground snowpack. Our understanding of canopy interception has been limited by our ability to measure whole canopy water storage in an undisturbed forest setting. This study presents a relatively inexpensive technique for directly measuring snow canopy water storage using an interceptometer, adapted from Friesen et al. (2008). The interceptometer is composed of four linear motion position sensors distributed evenly around the tree trunk. We incorporate a trunk laser-mapping installation method for precise sensor placement to reduce signal error due to sensor misalignment. Through calibration techniques, the amount of canopy snow required to produce the measured displacements can be calculated. We demonstrate instrument performance on a western hemlock (Tsuga heterophylla) for a snow interception event in November 2011. We find a snow capture efficiency of 83 ± 15% of accumulated ground snowfall with a maximum storage capacity of 50 ± 8 mm snow water equivalent (SWE). The observed interception event is compared to simulated interception, represented by the variable infiltration capacity (VIC) hydrologic model. The model generally underreported interception magnitude by 33% using a leaf area index (LAI) of 5 and 16% using an LAI of 10. The interceptometer captured intrastorm accumulation and melt rates up to 3 and 0.75 mm SWE h−1, respectively, which the model failed to represent. While further implementation and validation is necessary, our preliminary results indicate that forest interception magnitude may be underestimated in maritime areas.

1. Introduction

[2] Tree canopy snow interception is a significant hydrologic process that affects the depth and duration of ground snowpack [Winkler et al., 2005]. In highly forested areas, canopy snow interception is the primary control on the amount of snow available for spring melt. During periods of snow accumulation, it has been found that as much as 60% of total snowfall can be intercepted in the canopy [Hedstrom and Pomeroy, 1998; Storck et al., 2002]. Once snow is intercepted, it is exposed to ablative processes such as wind and solar radiation and is subject to sublimation, evaporation, or drip melt. Compounded over an entire season, tree canopy snow interception can account for the removal of up to 40% of water storage from the ground snowpack, when compared to nearby open areas [Pomeroy and Schmidt, 1993; Storck et al., 2002; Lundberg et al., 2004; Jost et al., 2007; Lopez-Moreno and Latron, 2008]. This is particularly important in the Pacific Northwest where 80% of land is forested [Alig et al., 2000].

[3] In the Western United States, communities rely on the snowpack as a vital water resource, since up to 70% of annual streamflow is provided by snowmelt alone [Mote et al., 2008]. As climate change projections indicate a loss of snowpack and a shift in snowmelt timing to earlier in the season [Hamlet, 2011], we must look for ways to optimize the available snowpack depth and duration. Findings from studies on the impact of forest structure on snowpack suggest that silvicultural manipulation in second-growth forests may help to balance a negative trend in snowpack snow water equivalent (SWE) by reducing the magnitude of snow intercepted by the trees [e.g., Jost et al., 2007]. However, in order to fully understand and predict the effects of forest manipulation techniques, we must first be able to accurately quantify snow interception timing and magnitude in the natural environment. Although there have been numerous studies attempting to do so, no one method or model has come forward as the standard.

[4] Several modeling approaches [Hendrick et al., 1971; Davis et al., 1997; Koivusalo and Kokkonen, 2002; Essery et al., 2008, 2009; Stähli et al., 2009] and local-scale empirical studies [Schmidt and Gluns, 1991; Hedstrom and Pomeroy, 1998; Lundberg et al., 1998] (more empirical studies summarized in a comprehensive review by Varhola et al. [2010]) have attempted to represent snow interception timing and magnitude. However, the presence of trees affects the hydrology of a basin through processes that are complex and difficult to predict. Therefore, many models fail to accurately represent snow interception in maritime mountain environments due to the lack of quality field data [Van Heeswijk et al., 1996; Marks et al., 1998; Raleigh et al., 2013].

[5] As models and empirical methods have fallen short, many studies have addressed the need for a direct snow interception measurement technique. Despite the large number of existing studies, which have contributed valuable information in an attempt to measure snow canopy interception, no existing technique has been universally implemented. This is likely attributable to the difficulty involved in measuring snow interception for a living tree. Single branch experiments [Schmidt and Gluns, 1991; Bründl et al., 1999] require upscaling estimates from branch to canopy scale and do not represent the entire tree. Whole-cut-tree load experiments [Lundberg and Halldin, 1994; Nakai et al., 1994; Hedstrom and Pomeroy, 1998; Storck et al., 2002] and an artificial tree experiment [Schmidt, 1991] remove trees from the stand environment, are costly, and are not feasible in a remote environment. Furthermore, these techniques may not represent interception behavior in an undisturbed setting. Gamma-ray attenuation techniques [Calder and Wright, 1986] can be costly and require handling of radioactive material. Photograph fractal geometry analysis [Pomeroy and Schmidt, 1993] is limited to a small view area, not the entire canopy. For more information, Lundberg [1993] and Lundberg and Halldin [2001] provide comprehensive reviews of techniques to measure snow interception and evaporation.

[6] Additionally, these existing studies have been largely limited to cold continental climates. Field experiments in maritime climates have been few in numbers with the exception of a cut-tree weighing lysimeter experiment [Storck et al., 2002] and a cut-tree weighing balance [Nakai et al., 1994]. A robust technique to directly measure snow interception could be implemented in any climatic region or forest structure and would provide a much needed measurement of in situ snow canopy interception.

[7] The instrumentation developed for this study, hereinafter referred to as an interceptometer, directly measures trunk compression to determine changes in canopy water storage. This study modifies techniques used in the past for tree wind loading [James and Kane, 2008] and canopy rain interception [Friesen et al., 2008; Van Stan et al., 2013]. Unlike rain, which evaporates relatively quickly, the weight from snow interception can be magnitudes greater and can be stored for weeks depending on the amount of snow intercepted, temperature, wind, and presence of liquid precipitation, among other things [Lundberg et al., 1998]. Therefore, a much larger and more persistent trunk compression is produced from snow interception when compared to rain interception, which greatly enhances interceptometer signal-to-noise ratio. To our knowledge, this measurement technique has never been used to quantify snow interception.

[8] This paper presents a relatively inexpensive and noninvasive, in situ measurement technique to measure snow canopy interception events for living trees. This technique is applied to quantify tree canopy snow capture and ablation in maritime second-growth coniferous forests of the Pacific Northwest. The remainder of this paper seeks to analyze the effectiveness of the interceptometer in its first year of field deployment and provide insight into future use of this technology. The data is also compared to simulated canopy interception as output by the hydrologic variable infiltration capacity (VIC) model [Liang et al., 1994; Andreadis et al., 2009]. This comparison provides a brief analysis of accumulated interception magnitude and timing. Section 'Theory and Background' discusses the theory and background of this technology; section 'Methods' provides a thorough description of the methods used to implement this technology; section 'Results' summarizes the results of this study. Section 'Discussion' discusses future implementation of the interceptometer and suggestions for future studies, and section 'Conclusions' provides concluding remarks.

2. Theory and Background

[9] Friesen et al. [2008] demonstrated that any change in canopy mass will result in a measurable trunk compression. This compression can be physically predicted because a tree trunk's behavior closely follows that of a linear elastic material and can be estimated according to Hooke's law of elasticity. Hooke's law states that the amount a material compresses due to an applied load is proportional to the applied load and measurement length and is inversely proportional to the stiffness of the material, referred to as the modulus of elasticity (MoE), and geometrical properties (equation (1))

display math(1)

where εc is the measured trunk compression strain, P is the axial force or weight applied to the trunk, ΔL is the observed sensor displacement, L is the length of the instrumented trunk section (ITS), A is the trunk cross-sectional area, and E is the tree's mean MoE, over the ITS.

[10] The mean MoE for softwood green timber can be found readily in timber handbooks. For example, Green et al. [1999a] reports a mean MoE, for fresh-cut softwood timber, of 7.6 GPa. However, the MoE of living wood is sensitive to changes in trunk moisture content, tree trunk diameter, changes in temperature [Cannell and Morgan, 1987], and season [Onwona-Agyeman et al., 1995]. Therefore, the MoE for each living tree must be individually calibrated [Friesen et al., 2008].

[11] Weight-induced trunk displacements (ΔL) tend to be very small. For example, using a weighing lysimeter, Storck et al. [2002] found that for an 8 m tall, 2.5 m crown diameter Douglas fir, the maximum snow canopy interception approached 30 mm of SWE over the entire winter. Assuming a 7.6 GPa MoE (as cited above) and a 15 cm trunk radius, we can apply equation (1) to calculate a theoretical displacement of about 2.5 μm. Such small displacements may only be captured with sensors capable of extremely high precision. Linear motion potentiometers provide infinite analog resolution [Bourns, 2009], allowing for quantification of such small displacements. These displacement sensors are also robust enough to be deployed in a wet and cold environment for an entire winter.

3. Methods

3.1. Site Description

[12] The study site is located within the Cedar River Municipal Watershed, approximately 50 km east of Seattle, Washington (47°21′N, 121°33′W, Figure 1a). This area has been classified as coastal temperate, a characteristically wet and mild maritime climate [Franklin and Dyrness, 1973]. A western hemlock tree was instrumented with the interceptometer for the 2011/2012 winter. The subject tree had a diameter of 28 cm and was approximately 20 m in height. The subject tree was located at approximately 640 m elevation on a southwest facing slope of less than 10°. This elevation is within the rain/snow transition zone of the North Cascades, where midwinter temperatures regularly rise above 0°C. Most precipitation that falls at an air temperature below 0°C falls as snow and above 3°C falls as rain [Kienzle, 2008; James et al., 2012]. In this paper, we will assume that all precipitation that falls at an air temperature below 0°C and above 3°C falls as snow and rain, respectively, with a linear mix of rain and snow in between. Mean hourly maximum and minimum temperatures at the site are 20.3°C and 8.5°C in July and 1.9°C and −3.4°C in January. Average annual precipitation is 2420 mm (1971–2000 climate normals; derived from the Parameter-elevation Regressions on Independent Slopes Model (PRISM) [Daly et al., 2008]). Annual snow accumulation varies between 0 and 2000 mm. Tree species composition in this area is primarily western hemlock (Tsuga heterophylla (Raf. Sarg.) and Douglas-fir (Pseudotsuga menziesii var. menziesii (Mirbel) Franco), with some western redcedar (Thuja plicata Donn ex D. Don in Lambert). There are low densities of Pacific silver fir (Abies amabilis Dougl. ex J. Forbes) at the research site but the species is predominant at higher elevations in the area [Kane et al., 2011]. The instrumented tree is representative of the vast majority of trees in the area and is part of a second-growth forest stand of approximately 67 years of age (in 2012), reestablished following clearcut logging in the early 20th century. Second-growth western hemlocks characteristically have tall, slender trunks and a high crown center of mass. The forest canopy at the research site is very dense, and there is generally very little light transmittance to the forest floor (<10% above canopy light) [Sprugel et al., 2009; Lutz et al., 2012]. The study sites are situated in a remote location without an external power source or remotely transmitted data capabilities, which is representative of most mountainous environments. Observed ground SWE accumulation was measured with a snow pillow in an area free of forest cover at the Mount Gardner National Resource Conservation Service (NRCS) SNOTEL site, elevation 890 m (Figure 1a). The 30 year average annual snow accumulation at Mount Gardner is 394 mm SWE. In areas with a high frequency of melt and freeze, ice layers have been known to form in the snowpack, causing difficulty in snow pillow measurements [Sorteberg et al., 2001; Johnson and Schaefer, 2002; Lundberg et al., 2010]. To account for this, a check of the snow density at Mount Gardner using the snow pillow data and the snow depth sensor showed a specific gravity between 0.2 and 0.5 during the observation period, which are within the expected range for the Pacific Northwest.

Figure 1.

(a) Map of the study site, (b) subject tree cross-section and dimensions, and (c) photo of the installed interceptometer.

3.2. Snow Surveys

[13] Manual snow surveys measuring snow depth and density were conducted in areas of varying canopy cover, described in Sprugel et al. [2009]. Open areas were represented by established gaps approximately 60 m (equivalent to the average height of one tree) in diameter, and forested areas were represented by undisturbed, second-growth forest. Approximately 40 snow depth measurements and three snow density measurements were taken in each forest type. During the 2011/2012 winter, eight total snow surveys were conducted, one of which occurred on 15 December 2011, shortly after the observation period presented in this paper. Snow depth surveys provide a benchmark on how much snow could potentially be removed by the canopy in forested areas compared to open areas.

3.3. Interceptometer Installation

[14] Preceding interceptometer installation, the tree's trunk was scanned using the LaserBark automated tree measurement system [Van Stan et al., 2010] (Figure 1b). This technology scans the circumference of the tree 1 m above the ground, at the height of the lower bracket of the interceptometer, providing a trunk profile, in polar coordinates, with a resolution finer than 10 radii measurements per degree. The geometric center of the trunk is calculated by approximating each successive measurement as a small triangle; once the centroid and area for each individual triangle is found, the centroid and area can be estimated for the entire trunk cross section [Van Stan et al., 2011]. A laser then marks the exact position of each sensor on the tree, ensuring that the placement is level and evenly distributed about the circumference of each tree. This placement minimizes interceptometer noise and offset from the neutral axes [Friesen et al., 2008; Van Stan et al., 2013]. Accurate sensor placement is crucial, as sensor misalignment increases error and reduces the signal-to-noise ratio.

[15] The interceptometer is composed of four linear motion potentiometers installed on two orthogonal axes that pass through the geometric center of the trunk cross section (Figures 1b and 1c). The four sensor arrangement allows for measurement redundancy in the event of a failed sensor pair and estimation verification when using the bending ratio method [Van Stan et al., 2013] for data processing, further described in section 3.6.1. For the scope of this paper, two methods are demonstrated to isolate tree compression; the strain-diagram method, which uses all four of the sensors, and the ratio method, which uses one coaxial sensor pair. The linear motion potentiometers were extended with 1 m quartz rods in order to achieve a more appreciable displacement measurement. The sensors were glued to brackets, which were fixed to the tree using steel banding. Using live computer output, the vertical position of each potentiometer was adjusted until it was centered in the allowable electrical travel range. This provided the ability to extend or compress as needed in response to wind action or uneven canopy loading. Following installation, the entire ITS was wrapped in insulation to maintain as stable of a temperature as possible.

3.4. Interceptometer Data Collection

[16] Data in this study were collected using Campbell Scientific CR10X 12 channel, 13 bit data loggers with 2 MB of internal storage. Each of the four interceptometer sensors was sampled in succession with a factory reported delay of 2.6 ms per successive reading [CR10X Measurement and Control Module Operator's Manual, 1997]. Data retrieved from the interceptometer were output as changes in millivolts. The raw millivolt signal was converted to a displacement by dividing by the allowable electrical range (2500 mV) and multiplying by the allowable mechanical travel distance (3.81 mm). In an attempt to minimize wind noise and maximize logger storage, an interceptometer measurement was taken every second (1 Hz), and the 5 min average was recorded. This recording scheme was selected to optimize measurement temporal resolution, data logger memory capacity, and battery consumption in our remote field sites. Equipment was run off of 12 V marine deep cycle batteries. Low winter light and high, dense canopies prevented the use of solar panels.

3.5. Data Processing to Isolate Compression Signal

[17] Although tree properties can be estimated using assumptions of linear elastic behavior, in reality, wood is a slightly nonlinear and heterogeneous material. Additionally, the trunk of a tree is not a perfect cylinder, but instead it tapers and deviates from a straight vertical axis. Trees are naturally slender structures and predisposed to bend under axial and in-span loading. This is further complicated by the fact that the tree canopy does not have an even branch/vegetation distribution. Therefore, a snow interception event will not produce a pure compression signal, but rather will be a combination of bending and compression, causing two sensors to compress and two sensors to extend. The compression signal must then be calculated based on the raw bending signal; for this, we employ two independent methods. The bending ratio method (section 'Bending Ratio Method') computes bending ratios using only two of the interceptometer sensors and is useful in the event of a failed or suspect sensor. We compare this method to a secondary data processing technique using all four interceptometer sensors plotted on a strain diagram (section 'Strain Diagram Method').

3.5.1. Bending Ratio Method

[18] On 15 June 2012, a pure bending test was used to identify the nonlinear bending behavior of each tree and to determine the MoE (Figure 2). During the bending tests, a high temperature of 17.9°C was recorded. A neighboring tree was used as an anchor, and a 4:1 pulley system allowed up to 1000 N of tensile force to be established in rope at a height of approximately 5 m above the sensors. The resulting load can be read directly off the spring scale, and the interceptometer records the strain produced. By using a static bending test, the trees can be loaded to both a wide range and large number of bending stresses, from several bending directions. Following a successful bending test, the subsequent MoE may also be calculated (see section 'Modulus of Elasticity') [Friesen et al., 2008; James and Kane, 2008]. Rolling pulleys were utilized at every joint to minimize friction. Bending tests were conducted while the canopy was dry and wind was at a minimum.

Figure 2.

Schematic showing a calibration bending test. Figure adapted from Van Stan et al. [2013].

[19] During the bending tests, it was assumed that no tree canopy loading existed, and all bending was due to the horizontal loads. When bending is applied in line with two of the sensors, one will extend and the other compress; this is referred to as the bending axis (Figure 2). In this paper, sensors going into compression are reported with positive strain, and sensors extending are reported as having negative strain. No displacement should be recorded on the two perpendicular sensors, as these sensors lie on what is referred to as the neutral axis. A bending ratio, ϕ, can then be calculated as the ratio of the displacement between the two opposing sensors that lie on the bending axis (equation (2)),

display math(2)

where εb1 is the strain due to bending alone at sensor 1, ε1 is the total strain at sensor 1, εb2 is the bending strain at sensor 2, and ε2 is the total strain at sensor 2. During the bending test, large displacements were observed in sensors along the bending axis, and virtually no displacement was observed in sensors along the neutral axis (Figure 3a). Since there is assumed to be no canopy loading during the bending test, the bending strain will be equivalent to the total strain (equation (2)). In a linear elastic material, with a symmetric cross section perpendicular to the bending axis, and symmetric potentiometer installation, ϕ would be −1.0; however, as seen in Figure 3b, ϕ is computed as −0.96.

Figure 3.

(a) Strain and (b) calculated mean bending ratio (ϕ), from bending test conducted on 15 June 2012.

[20] Once the bending ratio is known, it may be applied to interception events to isolate the compression signal. During a snow interception event, the total strain signal in each sensor, ε1, can be separated into the bending strain, εb1, and compression strain, εc (equation (3)). Likewise, ε2 can be separated into the bending strain, εb2, and the compression strain, εc (equation (4)).

display math(3)
display math(4)

[21] Using the bending ratio, the compression strain, εc, can be calculated based on the total strain produced in each sensor during the interception event (equation (5))

display math(5)

3.5.2. Strain Diagram Method

[22] As a tree undergoes canopy loading, the entire cross section will experience strain. The interceptometer's four potentiometers, which are distributed orthogonally around the trunk, record the total strain experienced during a canopy interception event. A strain diagram [Hibbeler, 2011] can be produced by plotting distance from the neutral axis to each individual sensor versus the strain reported for each of the corresponding sensors. While the actual location of the bending neutral axis is not known during an interception event, an optimization scheme is used to find the rotation angle, θ, from the neutral axis to each sensor at each time step by maximizing the r2 value of the line fitting the data when plotted on the strain diagram (Figure 4).

Figure 4.

A conceptual diagram of the strain diagram method for extracting the compression signal. The distance of the sensor from the neutral axis (x) versus the total strain recorded by the corresponding sensor (ε) is plotted. The compression signal is then taken as the y intercept from the strain diagram.

[23] For the application of the strain diagram method to the November snow event, the optimized r2 values were all above 0.99. The strain component solely due to compression from the snow loading will be uniform across the trunk's entire cross section. However, due to the presence of bending induced during snow loading, the strain at any point in the trunk's cross section will theoretically be linearly proportional to the perpendicular distance to the bending neutral axis (Figure 4). When plotted on the strain diagram, the compression component is taken as the y intercept.

3.6. Modulus of Elasticity

[24] The MoE for a tree is dependent on many things including wood density, water content of the wood [Green et al., 1999b; Cannell and Morgan, 1987], and wood temperature [Green et al., 1999b; Schmidt and Pomeroy, 1990]. Because of this variability, the MoE must be calculated for each tree. While loading a tree through a range of bending stresses, the interceptometer will record the resulting bending strains. The data may then be plotted on a stress versus strain curve, and the slope of the resulting linear plot is the MoE (E) (equation (6))

display math(6)

where math formula is the arithmetic mean of the absolute strain recorded by the 0° and 180° sensors (sensors lying along the bending axis) and σ is the bending stress at the outer fiber. For a circular cross section, σ can be calculated according to equation (7),

display math(7)

where F is the applied tensile force, d is the distance from applied tensile force to the strain measurement, y is the distance from the neutral axis to the strain measurement, and D is the tree's diameter, measured at the instrumented cross section.

[25] On 15 June 2012, the trees were incrementally loaded and unloaded using the described bending test. Settling was observed on the spring scale at every loading increment, most likely due to rope stretch and relative movement between the neighboring tree used as an anchor and the subject tree. Where settling on the order of 100 N of tensile force occurred, the bending values were taken as two different points. The MoE was computed as 5.80 GPa (Figure 5), which falls within the range reported in various literature for tree species similar to, and including, western hemlock [Youngs, 1963; Cannell and Morgan, 1987; Langum et al., 2009]. While the actual MoE was likely to have varied during the course of the year, a midwinter calibration was not feasible. However, the computed SWE is less sensitive to the MoE than the geometric properties of the tree (see sensitivity analysis in section 'Interceptometer Evaluation and Sensitivity Analysis'), which were measured with great accuracy using the LaserBark system [Van Stan et al., 2010].

Figure 5.

Shown above are the (a) raw data from a bending test performed 15 June 2012 along a bending axis in line with the 0°and 180° sensors and (b) the resulting stress versus strain curve.

3.7. SWE Calculations

[26] Prior to processing, a second-order, low-pass filter ( was applied to the raw data to remove any additional noise caused during windy periods. Equation (5) was then applied to the data in order to isolate the pure compression strain, εc. Using the compression strain, the weight of the snow in the canopy, P, was calculated by rearranging the terms from Hooke's law (equation (1)). Using the weight of snow in the canopy, the equivalent depth of SWE was calculated using the density of water and geometric properties of the tree crown (equation (8)),

display math(8)

where γ is the specific weight of water (9800 N m−3) and Ac is the vertically projected crown area, the area of the shadow the crown would make if the sun were directly overhead. The vertically projected crown area was determined by manually staking out the perimeter of the crown using Graphic Resource Solutions (GRS) densitometers ( Densitometers provide a cross-hair view directly above the observer. The locations of individual branches were marked on the forest floor, outlining the vertically projected canopy profile. The mean distance from the center of the tree trunk to the furthest reach of the branches was 3.64 m; Ac was then approximated as a circle of area, 45 m2.

3.8. Modeling

[27] In order to compare observations of snow interception with simulated values, we implemented the VIC model [Liang et al., 1994; Cherkauer et al., 2003] for a single grid cell at the location of the experimental plots. VIC is a widely implemented distributed hydrologic model, which implements a formulation for canopy snow interception that is based on observations from a 3 year field investigation in western Oregon [Storck et al., 2002]. Although there are a variety of approaches to representing the influence of vegetation on snow in hydrologic models [e.g., Rutter et al., 2009], we focus on VIC as an example comparison between our data and simulations of canopy snow interception that are grounded in observations from our study region. The VIC model balances the energy and water budget over a gridded domain based on meteorological and land cover inputs. Ground snow accumulation and ablation are represented by a two-layer energy and mass balance approach, and interception, evaporation, and radiative inputs are modified by vegetation parameters [Andreadis et al., 2009]. Canopy snow interception is represented by a partitioning of snow between the ground snowpack and canopy storage up to a maximum canopy storage capacity. The model (version 4.1.2c) uses an interception efficiency that varies with the fraction of available canopy snow storage, such that interception at one time step, It, is given in equation (9) as

display math(9)

where I is the total intercepted snow stored in the canopy, B is the maximum canopy snow storage capacity, and St is the snow that falls in that time step. The maximum storage capacity, B, is given in equation (10) as

display math(10)

where Lr is the leaf area ratio, m is an empirical constant, and LAI is the single-sided leaf area index of the overstory vegetation [Andreadis et al., 2009]. The leaf area ratio represents the temperature dependence of canopy snow interception as a step function, based on investigations by Storck et al. [2002], Kobayashi [1987], and Pfister and Schneedbeli [1999]. The leaf area ratio can be represented as seen in equation (11),

display math(11)

where Ta is the air temperature. Melting, evaporation, and sublimation of canopy snow are determined by the canopy snow energy balance, and mass release, M, is calculated as a linear function of meltwater losses from the canopy [Andreadis et al., 2009], as in equation (12),

display math(12)

where D is the amount of snowmelt that exceeds the liquid water holding capacity of the snow stored in the canopy, and thus results in water dripping from the canopy. At our study site and during our melt period, atmospheric relative humidity hovered near 100% while the temperature remained close to 0°C. Therefore, the primary mechanism for canopy snow storage to change here is meltwater drip, as documented by Storck and Lettenmaier [2000], and not sublimation, as has been observed in more continental sites [e.g., Mahat et al., 2013].

[28] We implemented VIC using estimates of local vegetation characteristics and forced the model with hourly observations of air temperature and wind speed from a meteorological station, approximately 1 km from the instrumented tree (Figure 1a) and with disaggregated daily precipitation from the Mount Gardner SNOTEL station, located approximately 2.5 km from the experimental plots, in the Cedar River watershed (Figure 1a). We compared the simulated hourly ground SWE and aggregated intercepted SWE with observations to assess the skill of the model in reproducing the timing and magnitude of interception events recorded by the interceptometer.

4. Results

[29] The following data are presented for an early winter storm cycle spanning the dates of 10 November to 3 December 2011. Although the interceptometer was deployed for the entire 2011/2012 winter, reoccurring power outages, animal disruptions (elk and black bear), and glue failures affected continuous data collection. Snow surveys conducted during the 2011/2012 winter showed, on average, a 40 cm (60%) greater midwinter snow depth in open areas as opposed to forested areas, suggesting that interception has a strong impact on ground accumulation in forest versus open sites in this area. From a manual snow depth survey conducted on 15 December 2011, an average of 44.2 cm snow depth was reported in the open area compared to an average of 21.8 cm snow depth in the forested area. The average specific gravity of the snow on this day was found to be 0.37 in the open area and 0.33 in the forested area corresponding to a deficit of 91.6 mm SWE (56% of that in the open area) in the forested area following the snow event. During the observed storm event, incoming solar radiation remained low, reaching a maximum of 322 W m−2 (measured at Mount Gardner, Figure 1a). Relative humidity remained high during the observation period, with an average relative humidity of 94.6% and a standard deviation of 9.6%.

[30] Snow canopy interception timing and relative shape can be seen for a 3 week period spanning 11 November to 3 December 2011 from the raw data (Figure 6). Relative raw signal and compression magnitudes can be seen by comparing the compression data to the total strain signal. On average, the total strain signal was 17.3 times larger than the processed compression signal. Due to this fact, the processed compression signal can be sensitive to small changes in the overall strain signal. To reduce the compression signal to noise ratio, a low-pass data filter removed small noisy patches, likely due to wind signals, in the raw interception signal while maintaining the underlying shape of the data (Figure 6).

Figure 6.

Interceptometer raw filtered strain data along with processed compression data for (a) the 0°–180° bending axis and (b) the 90°–270° bending axis. The light gray lines represent the unfiltered data for each sensor.

[31] The strain diagram method and ratio method were applied to the raw strain signal in order to compute the compression (Figure 7a). These two methods had a mean difference of 0.56 mm SWE during the observation period and comparable standard deviations; 15.1 mm SWE for the ratio method and 14.0 mm SWE for the strain diagram method. Using the strain diagram method, the neutral axis was solved at each time step. The predominant direction of bending, averaged over all time steps, during the storm event was found to be 54.7°, relative to the 0° sensor (Figure 6). Over the period of observation, the bending direction ranged from 40.3° to 80.2°. During the observed storm cycle, 180 mm of precipitation fell in the form of snow, measured with a snow pillow at the Mount Gardner SNOTEL site (Figure 1a), the only available continuous SWE data in the area. Although Mount Gardner is 250 m higher in elevation than the instrumented tree, a snow depth sensor at an elevation only 30 m higher than the instrumented tree tracked well with a snow depth sensor at Mount Gardner. Mount Gardner maintained a snow depth, which was on average, 6.8 cm greater than the supplementary data station over the course of the observation period. Temperatures at the Mount Gardner station and the supplementary meteorological station (Figure 1a) were comparable. The average hourly temperature at Mount Gardner during the observation period was 0.7°C with a standard deviation of 2.4°C. The average hourly temperature at the supplementary meteorological station was −0.3°C with a standard deviation of 2.5°C. Temperatures routinely rose above 0°C, as is common at the research site, and wind speeds stayed relatively calm, never exceeding 2 m s−1 (Figure 7b). Although the recordings of zero wind during times with below-zero temperatures are suspicious, near-identical wind patterns were recorded at three surrounding anemometers. If the anemometers were freezing, this was happening uniformly across the study area.

Figure 7.

Snow canopy interception event showing (a) the total accumulated ground SWE and the intercepted SWE as computed by the strain diagram method and the ratio method and (b) the observed meteorological conditions during the same period, where the horizontal dotted lines represent the snow and rain thresholds.

[32] During the early part of the storm cycle, the trees intercepted 83% of total accumulated ground SWE, until the canopy reached a maximum carrying capacity of 50.1 mm (Figure 8, average of ratio method and strain diagram method). Because snow accumulated and was shed from the canopy in the early part of the storm, Figure 8 shows the efficiency for low- and high-ground accumulation periods, both in line with the 83% slope.

Figure 8.

Ground SWE measured at the Mount Gardner SNOTEL station versus intercepted SWE (average of ratio and strain diagram methods) with efficiency lines fit to a low- and high-ground accumulation period.

[33] The VIC model was used to output snow canopy interception using LAI values of 5 and 10 and compared to the measured snow interception using the interceptometer (Figure 9a). Second-growth western hemlock trees near the study site have been estimated to have LAI values ranging between 8 and 12. Therefore, the LAI estimate of 10 would be representative of the instrumented western hemlock. By comparing the LAI values of 10 in the model to the LAI value of 5, we gain insight to model behavior and interception dependence on the LAI value. For simplification, model results will be reported for LAI = 5(LAI = 10). The model reached a maximum interception capacity of 10.0(18.8) mm SWE. Compared to the measured interception, the model over reported intercepted SWE by 10(17)% for the initial accumulation period and underreported intercepted SWE by 33(16)% for the remainder of the snow accumulation event (Figure 9a). The model also failed to represent interception variability, such as canopy accumulation and ablation periods within the storm event, as demonstrated by the low standard deviations of 2.22(4.7) mm SWE when compared to the standard deviations reported by the interceptometer of 15.1 mm SWE as computed by the ratio method and 14.0 mm SWE as computed by the strain diagram method (Figure 9a).

Figure 9.

(a) A direct comparison of the VIC hydrologic model interception output to the measured interception calculated by both the strain diagram method and the ratio method and (b) the observed meteorological conditions during the same period, where the horizontal dotted lines represent the snow and rain thresholds.

[34] In order to more closely examine the behavior of intercepted canopy SWE, a smaller portion of the observation period is shown in Figure 10. During this period, one canopy snow accumulation event and two periods of ablation in the form of canopy melt and evaporation are highlighted. As captured by the snow interceptometer, snow accumulated in the canopy at 3.2 mm h−1 and was shed from the canopy at 0.75 and 0.51 mm h−1. The model failed to represent these periods of accumulation and ablation. While the canopy accumulation periods, as measured by the interceptometer, correspond to ground SWE accumulation at Mount Gardner (Figure 10a), the two ablation events correspond to only minor increases in temperature and wind speed (Figure 10b), signifying that long periods of supporting interception observations would be necessary to train a model to correctly simulate these events.

Figure 10.

(a) Modeled and observed interception as well as ground accumulated SWE measured at Mount Gardner for a subset of the time shown in Figure 9, one period of accumulation and two periods of ablation are highlighted and (b) the observed meteorological conditions during the same period, where the horizontal dotted lines represent the snow and rain thresholds.

5. Discussion

5.1. Interceptometer Evaluation and Sensitivity Analysis

[35] Through field calibration, the interceptometer, installed following techniques described in Van Stan et al. [2011], has been shown to lie on perpendicular axes, which minimizes errors resulting from signal processing and enhances the signal-to-noise ratio (Figure 3). Pure bending tests showed a bending ratio of −0.96 along the bending axis and virtually zero displacement along the neutral axis, demonstrating that a tree behaves almost ideally as a linear elastic material at low stress levels (Figure 3). The soundness of installation provides confidence in the LaserBark measurement system to identify the geometric center of the tree and estimate the radius and area of the trunk to a high accuracy. While these results are for a tree with a nearly symmetric cross section, future studies will evaluate performance on trees with less uniform cross sections. The strain diagram method (Figure 4) was used in comparison with the bending ratio method to compute the compression signal from the total strain signal and to quantify the bending direction of the tree. The fact that the two methods were within 1% agreement demonstrates the ability to extract the compression signal from the total strain signal using the strain diagram method for four functional sensors, which computes the bending angle, and the ratio method using only two sensors in the event of sensor failure. However, both methods were sensitive to noise in the total strain signal.

[36] While it is unclear what caused unrealistic jumps in the total strain signal in the latter part of the storm, a relatively smooth period void of such jumpy behavior, before 23 November 2011 (Figure 6a), demonstrates the potential for high-resolution and accurate interceptometer measurements. During instrument calibration, the interceptometer recorded a linear stress versus strain relationship, which allowed us to calculate the MoE of the tree as 5.58 GPa (Figure 5). With the exception of branch bending experiments [e.g., Cannell and Morgan, 1987; Onwona-Agyeman et al., 1995], the MoE has not been previously well quantified for living trees. Due to temperature variation during the winter, the MoE may likely have varied throughout the year. A study conducting bending tests throughout the duration of the winter could effectively quantify the variation of the MoE but was beyond the scope of this study. Alternatively, a sensitivity analysis was conducted to quantify the response of intercepted SWE estimate to realistic variations in the MoE and errors due to measurement of the projected crown area. Based on previous literature, decreases in temperature increase the MoE. For example, changing the temperature from 13°C (maximum hourly temperature on 15 June 2012, date of bending test) to 0°C, corresponds to an increase in MoE of 5% [Green et al., 1999b]. Additionally, it is feasible that measurement error of the vertically projected canopy area could have been up to 10% as either an overestimate or underestimate. By increasing the MoE by 5% and decreasing the projected crown area by 10%, we would calculate an average increase in interception efficiency of 15% (for a total efficiency of 98%) and calculated maximum intercepted SWE of 59 mm (a 16.7% increase). By maintaining the calculated MoE and increasing the projected crown area by 10%, we would calculate an average decrease in interception efficiency of 7% (for a total efficiency of 76%) and calculated maximum intercepted SWE of 46 mm (a 9.1% decrease).

5.2. Interceptometer Comparison to Previous Literature and Model Output

[37] Over the entire period of observation, the calculated canopy interception showed that our single western hemlock, within a second-growth stand, removed up to 50 mm SWE from the ground snow pack at an efficiency of 83% of snowfall measured in a clearing (Figure 8). The sensitivity study conducted provides some insight to what the errors in this estimate could be. The results of a manual snow survey on 15 December 2011, approximately 2 weeks after the observation period, shows a 92 mm SWE deficit in the forested area compared to the adjacent open area. While the snow depth deficit in the forested area could be due to a combination of interception and melt, the magnitude of the deficit shows that it is feasible that the instrumented tree could have intercepted up to 50 mm SWE during the observed period. Additionally, the actual removal of snow from the ground snowpack may be larger for an entire stand than the interception capacity of a single tree, due to the overlapping of tree canopy area in forests increasing interception capacity.

[38] We also ran the VIC hydrologic model for a comparison. We tested the effect of varying LAI on the model representation of the magnitude and duration of canopy snow storage because the maximum storage capacity depends on LAI (equation (10)), and canopy snow storage remained near its maximum capacity for much of the simulation. However, many other factors control the snow interception and storage in the model. In particular, we speculate that the model fails to represent losses from canopy snow storage due to undersimulation of melting of canopy snow. Storck et al. [2002] observed low sublimation rates in the maritime Pacific Northwest, where relative humidity is close to 100% for much of the winter and determined that meltwater drip and subsequent mass release from sliding are the main causes for loss in canopy snow storage. With air temperature fluctuating close to 0°C during the observational period, slight variations in the canopy energy balance will trigger whether intercepted snow melts or not. The lack of variability in canopy snow storage as represented by the model here may be due to errors in the canopy energy balance. Observations of canopy storage, such as those presented here, are needed to improve representation of canopy snow process. While more trees must be instrumented with the interceptometer for more snow events before we can recommend model changes, we can gain some insight from this preliminary study. As shown, the model both underestimated the magnitude of intercepted snow (Figure 9a) and also failed to capture intrastorm variability (Figure 10a). Although models are generally run on a large scale and represent considerable forest variability, the results of this study may indicate that the model representation of snow interception behavior could be improved through regional calibration in comparison with direct field measurements of snow canopy interception.

[39] To see how the results of this study compared to previous research, we refer to previously cited literature as a comparison of maximum interception capacity and efficiency. As seen in Table 1, the results of the present study show both a larger interception magnitude and efficiency than previous studies. Many of the previously cited studies are in cold, continental climates [Bunnell et al., 1985; Schmidt, 1991; Schmidt and Gluns, 1991; Pomeroy and Schmidt, 1993; Hedstrom and Pomeroy, 1998]. In continental climates trees often have thinner canopies when compared to trees in maritime climates, and the snow that falls is often less dense due to lower temperatures and may slough off the branches more easily [Ward and Trimble, 2004]. However, this study also reports higher interception magnitude and efficiency than studies in comparable maritime climates. This may be due to a difference in tree type and behavior. The trees in dense second growth stands compete for light and, as a result, have very high and dense canopies, which may be capable of intercepting more snow than an isolated tree. Furthermore, as noted at our field site, the canopies tend to bridge, which may allow for a higher interception capacity where the tree canopies overlap.

Table 1. Literature Review Comparing Maximum Intercepted Canopy SWE and Efficiency for Various Tree Types and Geographic Locations
ReferenceMeasurement TechniqueStudy LocationTree SpeciesMaximum Interceptiona
  1. a

    Units, mm SWE (efficiency).

  2. b

    Direct, indicates a direct measure of whole canopy tree interception.

  3. c

    Indirect, indicates an estimation technique for determining whole canopy interception.

Bunnell et al. [1985]Directb, tree-weighing apparatusBritish Columbia, CanadaDouglas fir30 (45%)
Calder [ 1990]Direct, tree-weighing apparatusCentral ScotlandSitka spruce30 (66%)
Johnson [1990]Indirectc, precipitation - (throughfall + stemflow)Central ScotlandSitka spruce22 (28%)
Schmidt [ 1991]Direct, tree-weighing apparatusColorado, USAArtificial coniferN/A (58%)
Schmidt and Gluns [ 1991]Indirect, cut branch experimentColorado, USA and British Columbia, CanadaEngleman's spruceN/A (50%)
   Lodgepole pineN/A (45%)
   Subalpine firN/A (45%)
Pomeroy and Schmidt [ 1993]Indirect, manual snow surveysSaskatchewan, CanadaBlack spruce3.5 (60%)
   Jack pine7 (45%)
Nakai et al. [1994]Direct, tree-weighing apparatusSapporo, JapanSix conifer species30 (50%)
Hedstrom and Pomeroy [ 1998]Indirect, paired undercanopy and open-area snow gaugesSaskatchewan, CanadaBlack spruceN/A (45%)
   Jack pineN/A (30%)
Storck et al. [ 2002]Direct, tree-weighing lysimeterOregon, USADouglas fir40 (60%)
This studyDirect, tree interceptometerWashington, USAWestern hemlock50 (83%)

5.3. Recommendations for Future Installations

[43] Although the interceptometer was disrupted several times during the 2011/2012 winter, this study provided a prototype-level installation. Therefore, future implementations of this technology could gain considerably from where our instruments failed. During the course of the winter, we noted several opportunities for improving upon the instrument installation presented here.

[44] The data collection system used in this study represents the lower end of the available technological spectrum. Increases in data storage capacity, data logger resolution, and battery usage efficiency will only improve instrument performance. During this application, the data were sampled at a high frequency (1 Hz), and 5 min averages were recorded in an attempt to maintain a high sample rate and optimize data logger storage space. We were able to utilize this recording scheme due to the persistence of snow in the canopy. An application of the interceptometer to record wind throw or rain interception would require a higher sampling rate and resolution and would not allow for averaging. Installing a small temperature sensor within the insulation, next to the interceptometer, would allow for monitoring of the instrument temperature and would provide insight to potential thermal expansion and contraction of the instrument.

[45] A bracket design, which is not strapped to the tree trunk, but drilled directly into the tree could significantly ease the installation process. An improved bracket design would allow for slight rotation about the x and y axes as well as translation in the z direction. The ability for slight adjustments during installation would ensure that the instrument would be aligned both vertically and horizontally, eliminating torque where the quartz rod meets the potentiometer. By reducing the number of connections in the interceptometer, it would reduce the number of places where glue has the potential to fail. During the installation, glue failure occurred on several instances when using two-part epoxy. Finding a reliable glue and, as mentioned, reducing the number of connections would lessen the probability of glue failure.

[46] The implementation of a track-and-hold circuit would allow for near-instantaneous measurement of all four sensors in the potentiometer. The data logger specifications for our installation report a 2.3 ms delay in successive sampling. This delay could produce errors due to displacements caused by wind while the interceptometer is being sampled. We employed a high-frequency measurement technique, while taking a longer-term average. However, a track-and-hold circuit would allow for a higher sampling frequency and the assurance of eliminating error as a result of sampling delay.

[47] Through further instrument testing and development, a more robust and simplified design could be achieved allowing for significant improvements in instrument performance. The interceptometer may then not only improve our understanding of canopy interception but also provide applications for research beyond snow interception. For example, the interceptometer may be used to gain information regarding trees that pose hazard to buildings and power lines, through observations of wind, and loading during ice storms. By identifying actual loading, models could be developed for failure criteria, and suspect trees could be removed avoiding potentially catastrophic destruction.

5.4. Broader Implications

[48] This study has demonstrated that the interceptometer is capable of measuring the canopy intercepted SWE for an individual tree. The development of the snow interceptometer represents a new technique to quantify snow canopy interception. Since the interceptometer is installed in situ on a single tree within a stand, the recorded interception behavior is representative of how trees behave in their natural environment. Due to the nondestructive installation methods, the interceptometer may be ideal for observations in environmentally sensitive areas such as old-growth forest stands. The relative ease of interceptometer installation could allow for examining the interception behavior of a larger sample size of trees. By sampling more trees, subsequent studies could effectively capture variation in interception behavior due to trees of different age and species, as well as trees of varying elevation, slope, aspect, stand type, and climatic regions. This type of variation has not previously been represented and would provide data essential for model validation and training. By improving model representation of interception, watershed managers interested in the effects of silvicultural manipulation to optimize snowpack, could examine the effects of manipulating forest cover on the watershed scale.

6. Conclusions

[49] The results of this study represent the successful deployment of the tree interceptometer to record snow canopy interception. Using the Van Stan et al. [2011] method for interceptometer installation, we have demonstrated even sensor spacing and achievement of the neutral axis. The interceptometer is a technology that can be installed relatively easily and inexpensively, in live trees within existing canopy structures in any watershed. Where models may fail to represent interception efficiency, maximum interception capacity, and intrastorm interception variability, the interceptometer serves as a practical method for model validation and regional calibration. While the quantification of tree canopy interception has previously been difficult to attain, the tree interceptometer may be a viable tool to increase the number of interception studies in any region, remote or accessible. By sampling more trees, subsequent studies could effectively capture variation in interception behavior due to trees of different age and species, as well as trees of varying elevation, slope, aspect, and stand type and climatic regions. With this information watershed managers may then plan for silvicultural manipulation in specific areas to optimize the amount of water stored in the ground snowpack for spring melt.


[50] This work was made possible through funding provided by NSF grant CBET-931780. The authors thank Mark Raleigh, Brian Henn, Shara Feld, Nic Wayand, Nicoleta Cristea, Laura Hinkelman, Courtney Moore, Alex Fischer, Jenna Forsyth, Melanie Richmond, Samuel Barr, and Jim Syvertsen for contributing to fieldwork and paper revision suggestions. Additional thanks to John Selker at Oregon State University and Jan Friesen of the Helmholtz Centre for Environmental Research (Leipzig, Germany) for numerous discussions and instrument help. The authors also like to thank Vince Chaijaroen for access to the University of Washington structural engineering laboratory and help with instrument testing and production as well as Lisa Berman for providing a tree for prototype testing. Additional technical and intellectual support was provided by Delphis Levia and Matthew Jarvis at the University of Deleware. We are also grateful for access to the Cedar River Municipal Watershed, which was provided by Seattle Public Utilities.