Canopy effects on snow accumulation: Observations from lidar, canonical-view photos, and continuous ground measurements from sensor networks

: A variety of canopy metrics were extracted from the snow-off airborne light detection and 1 ranging (lidar) measurements over three study areas in the Sierra Nevada, Providence and Wolverton 2 from the southern Sierra and Pinecrest in the central Sierra. More than 40 snow-depth sensors were 3 deployed at Providence and Wolverton since 2008 and about 10 sensors were deployed at Pinecrest 4 since 2014 for long-term snowpack measurements. At Wolverton, hemispherical-view images were 5 captured and the sky-view factors were derived from the images at each individual zenith angle. We 6 extracted the snow accumulation characteristics for each sensor measurements over multiple years. As 7 the sensors were deployed under various canopy-cover conditions, we studied the variation of snow 8 accumulation across landscape and found they are controlled by the canopy-cover conditions. We used 9 regularized regression model Elastic Net to model the normalized snow accumulation with canopy 10 metrics as independent variables, and found that about 50% of snow accumulation variability at each 11 site can be explained by the canopy metrics from lidar. 12


13
The snowpack in California's Sierra Nevada has long been served as the primary water resources for 14 agricultural and urban uses [1]. For seasonal forecasts of flood peaks following the onset of snow melt, 15 the estimation methods are turning from statistical estimates that use historical records to spatio-temporal of cumulative snowfall may be intercepted by forest in mid-winter and annual sublimation losses can be 23 30 − 40% of annual snowfall [8]. Being able to accurately quantify the canopy interception of snowfall 24 is the foundation to estimate the total snow melt with higher accuracy and precision during the Spring 25 season. 26 The canopy interception of snowfall can be quantified as the snow storage capacity of the canopy 27 and interception efficiency (interception/snowfall). The snow storage capacity is the maximum amount 28 of snowfall that can be intercepted by the canopy. It is determined by the leaf area, tree species, and initial 29 canopy snow load [9]. The interception efficiency is found to decrease with increasing snowfall, initial  The coniferous canopies interception on snowfall is difficult to measure and quantify. Previous 33 studies designed special weighing devices such that the weight of the intercepted snow accumulated 34 snow can be measured at the same time. The total snow interception is found to be correlated with the 35 accumulated snowfall [8,9]. Thus, several process models have incorporated this statistical finding and 36 account canopy-cover effect on snow accumulation [11][12][13][14]. 37 To calculate canopy interception, using canopy metrics that are highly correlated with the total 38 snow accumulations is a common solution. Retrieving canopy metrics has advanced in recent 39 years. The technology has been advancing from the traditional plant canopy analyzer [12,[15][16][17][18], 40 to hemispherical-view camera [19,20], and recently, to lidar [21,22]. The plant canopy analyzer was 41 commonly used for retrieving the LAI in the forest. By using the hemispherical-view camera, the 42 pixels of the taken images can be classified as either canopy-cover or clear, thus the percentage of 43 clear view for each zenith angle can be quantified as sky-view factor, which was also found to be a 44 statistically significant predictors for parameterizing snowfall interception in the process models [19,23].

45
The point-cloud data collected using lidar can be used for reconstructing the 3-dimensional canopy 46 structures if the point-cloud has enough density. Algorithms have been developed for deriving LAI from 47 the lidar point clouds and it will be interested to develop new canopy metrics from lidar for quantifying 48 the snowfall interception.

49
In addition to canopy-metric retrieval from lidar, the canopy effect can also be quantified by using 50 statistical models, with dense spatial measurements of snow depth or snow water equivalent (SWE) [ in canopy effects can also be determined.

57
One short-coming in using lidar is it lacks temporal completeness, especially during the precipitation 58 season, when it is difficult to take measurements. Lidar requires clear sky condition to take measurements 59 to prevent the laser pulse intensity from attenuating because of rain drops and snow flakes [25].

60
A dense cluster of snow-depth sensors can compensate the weakness of lidar in terms of temporal 61 consistency. Combining the vegetation structures derived from lidar measurements and continuous 62 snow-depth measurements, there is potential that the spatial variation of snow accumulation can be 63 accurately quantified. In our study, we used long-term spatially dense snow measurements in the Sierra 64 Nevada, together with the lidar-derived canopy metrics, to study the canopy effect on seasonal snow 65 accumulations.

66
The general objective of the work reported here is to explore the possibility of studying the spatial 67 variability of snow accumulation by using lidar-derived canopy metrics and clustered snow-depth sensor  The study was conducted over three areas in the Sierra Nevada: Pinecrest in the Central Sierra, 74 and Providence and Wolverton in the Southern Sierra (Figure 1(a)). For each study area, snow-depth 75 sensors (Judd Communications) are instrumented and they are placed into clusters (Figure 1(b, c)), with 76 topographic characteristics (elevation, aspect) varying between clusters and canopy-cover conditions varying within each cluster. Pinecrest is the lowest in elevation and also flat in terms of elevation gradient.

78
The lower site of Providence has similar elevation range as Pinecrest and the upper site is 200-m higher.

79
Wolverton is the highest of the three study areas, with elevation around 2200 m at the lower site and 2600  (Table 1).

Lidar data 83
The point-cloud lidar data were used for generating raster data sets. The raw point-cloud files were 84 divided into 250×250-m tiles using LAStools lidar processing software. We extracted the ground points  The canonical-view images were taken below each individual sensor node, facing straight-up to the 96 sky. The sky-view factors f at each individual zenith angle θ were derived from the raw image and the 97 sky-view factor of the entire image is also estimated using the equation below, The sky-view factors data are available for Wolverton only. We included the total sky-view 99 factor and the sky-view factors at each zenith angle as independent variables for modeling the snow The data availability over time for each site is shown in Table 1. To study the canopy effect on snow 104 accumulation, we extracted all events when most precipitation is in solid form. This kind of events can 105 be extracted from the time-series snow-depth data through the following procedure.
where y is the target value and X is the matrix of all covariates. The estimates of the regression 137 coefficientsβ is defined as, The Elastic Net was chosen than other regularized regression approaches for its ability of addressing 139 correlated covariates and when the number of covariates is high. In our case, the canopy metrics can be Pinecrest has a relatively short record, most of which is during the heavy drought of California, we did 152 not conduct the analysis for Pinecrest. Also, camera images are not available for Providence thus we 153 only radius dependency analysis at that site. For the data at Wolverton, we selected a few near-optimal 154 searching radii and zenith angles. We used these selected variables and conducted a step-wise linear 155 regression process for exploring the relative importance between variables. is similar to manual extraction that needs to be done by human. As is shown in Figure 2, the algorithm is 8 of 16 able to detect most major snow accumulation periods. And the summary of accumulation events detected 162 for each site is shown in Table 2

165
The variability that the Elastic Net model can explain over the three sites are shown as in Figure 4.  Considering Wolverton is the only study area that both SVF and lidar are available and the trends 179 observed in Figure 5. We constrained the valid mean precipitation in the range of 15−30 cm. We 180 conducted three sets of analysis, including using lidar-derived canopy metrics as the predictors, using 181 SVF as the predictors, and using both lidar and SVF as the predictors in the Elastic Net model. We  snowfall, which is similar to [10] has found, which stated that total interception of snowfall will saturate 214 when the total precipitation reaches certain thresholds for different tree species. In addition, we observed 215 some noise introduced by the low precipitation events in the regression analysis. Figure 5 suggests that 216 the spatial variability of precipitation is less explainable by the canopy-related variables when the total 217 precipitation is small. The snow-cover information is slightly more important than canopy metrics, including tree heights 227 and tree-height standard deviations. This was verified by correlation coefficients in the regression 228 analysis between snow accumulation and both tree height at increment searching radius and SVF at 229 increment zenith angles. The SVFs are more correlated with snow accumulation than tree height in 230 general. The step-wise regression analysis also suggested that sky-view factor at optimal zenith angle 231 is more important than tree height at optimal searching radius. Although the tree height is an important 232 metric characterizing trees in the forest but it does not necessarily represent the density and interception 233 capacity of the canopy. Even sky-view factor only represents the canopy-cover condition at the lowest 234 layer of canopy, it still explain partial variability in the interception capacity of the entire tree crown, 235 which is the reason that it can be more important than lidar-derived canopy metrics.

236
Comparing within lidar-derived canopy metrics at increment radius, Figure 8 suggests that the most 237 important canopy structures may not be the canopy layers right above the measured locations. The 238 canopy surrounding within a few meters could be even more important as the interception capacity can 239 be larger when the trees are clustered together than a single tree stand.