Results
Spatial Patterns of Snow and Forest
Structure
Forest structure metrics are highly correlated (Figure A3, Figure 4).
LAI’, fVEG, and canopy density tend to increase with elevation and
northness. Openness shows the opposite trend, decreasing with elevation
and northness (Figure A3). Following expected terrain relationships,
absolute SWE is greater at higher elevations on more northern-facing
slopes in all three accumulation flights (Figure 5). Lower elevations
have steady decreases in SWE in denser canopy, but these relationships
become less distinct at higher elevations (Figure A6). The relationships
between SWE and terrain are nonlinear and in agreement with previous
findings (e.g., Kirchner et al., 2014; Tennant et al., 2015; Varhola et
al., 2010) showing that areas with lower SWE tend to be more variable in
terms of fVEG and northness, particularly at higher elevations.
R-squared values for SWE-elevation regressions are 0.51, 0.68, and 0.43
and slopes are 0.05, 0.19, and 0.09 for the 2008, 2016, and 2022
flights, respectively. R-squared values for SWE-fVEG regressions are
0.12, 0.09, and 0.02 and slopes are -23, -58, and -12 for the 2008,
2016, and 2022 flights, respectively. We calculated that p
<0.001 for all models (Figure A9).
“Open Reference Site”
Approach
A space-for-structure approach
allows us to isolate the effects of vegetation structure by analyzing
differences within 30-m grids of up to 900 1-m pixels that we assume
experience homogeneous precipitation and the same terrain. We isolate
both coarse (fVEG) and fine (LAI’ and openness) forest structures using
lidar datasets. This approach is applied across three accumulation
flights (2008, 2016, and 2022) and two ablation periods (March to April
and April to May, 2016).
Accumulation
DSnowAc patterns from SCB are in general agreement with data used in
Varhola et al. (2010). DSnowAc decreases (becomes more negative) with an
increase in fVEG, indicating that open areas accumulate more snow
relative to average 30-m grid cell values (Figure 7; Figure A7). A
linear regression of DSnowAc and fVEG shows greater variability and
shallower slopes (-0.23, -0.37, and -0.15 for the 2008, 2016, and 2022
accumulation flights, respectively) than the data found in the Varhola
et al. (2010) meta-analysis (-0.37; Table 1).
The RF models have varying levels of skill, with R-squared values of
0.72, 0.62, and 0.16 for the 2008, 2016, and 2022 flights, respectively
(Table 2). Forest structure metrics explain a greater amount of variance
in the model than terrain metrics (Figure A10). Partial plots of snow
accumulation predict DSnowAc values between -4% and -20% per 100%
change in fVEG and show that on average the forest sites accumulate less
snow than open reference sites (Figure 9). A larger absolute DSnowAc
indicates that the discrepancy between open and total grid cell SWE is
increasing. In this study, we emphasize patterns consistent across
select flights. For the 2008 and 2016 flights, fVEG and LAI’ have the
greatest importance on the RF models with about three times as much
influence as terrain variables (Figure A10). There are distinct ranges
of influence where certain variables predict greater change in DSnowAc.
When fVEG < 0.3, we predict consistent decreases of
~5% snow accumulation. At fVEG values >
0.3, these patterns become more variable, and decrease steeply until
fVEG ~0.7, where the relationship becomes weaker but
continues to predict decreasing DSnowAc of around -15%. Qualitative
thresholds can be observed for both LAI’ and openness as well; DSnowAc
maintains a maximum around -10% until openness reaches -2.5. A sharp
decrease in DSnowAc is predicted until a minimum threshold at around
-17% DSnowAc and an openness index of 0 (indicating gap diameter
similar to canopy height). DSnowAc shows a steep decreasing trend with
LAI’, from DSnowAc around -4% at an LAI’ of 0 to a DSnowAc minimum
around -11% where LAI’ is 0.2 (Figure A11). Northness and eastness
showed inconsistent results with low importance and no clear trends
across the flights (Figure A15; Figure A16). However, the 2008 flight,
which best captures accumulation at lower elevations, indicates that
southern-facing slopes have a greater difference between open and
under-canopy accumulation than other aspects (Figure A14).
Response zones are chosen to quantify snowpack response to canopy
removal based on the thresholds identified above (Figure 9). Low
response zones are expected to experience the least amount of change
post-canopy loss, or post-treatment; high response zones are expected to
benefit snow the most. Areas with lower absolute values of DSnowAc
(values that approach zero) are typically locations with low density
canopy where we would expect greater total accumulation compared with
areas with dense canopy and more negative DSnowAc values. Using this
framing, we qualitatively identify high (fVEG>0.6 and
openness<-2.5), low (fVEG<0.3 and
openness>0), and moderate (all remaining fVEG and openness
values) snowpack response to canopy loss. Low DSnowAc response areas are
primarily at higher elevations, often in the steeper terrain, whereas
high response areas tend to be at lower elevations and closer to the
valley floor (Figure 6) due to pre-existing patterns in vegetation
structure (Figure 4). These response zones show clear differences in
average DSnowAc from the space-for-structure approach of
~0, -10, and -30% across low to moderate to high
response areas, respectively. Low response areas accumulate more
absolute SWE (60 cm in low-response areas vs. 40 cm in high response
areas), indicating they would not benefit as much from canopy removal
(Figure 10).
The response zones determined through our expanded analysis show clear
patterns with previous management and, particularly, forest fires
(Figure 11). Areas that have been previously burned are dominated by
Ponderosa Pine forest and shrub conversion and have low response
classifications. High response areas tend to be forests containing
Douglas, Grand, and White Fir. SCB is dominated by moderate response
areas, with the proportion of high response areas in planned treatment
zones (compared to low-response areas) increasing post-2014.
Ablation
Both DSnowAb metrics show weak increasing linear trends with fVEG and
have R2 values of 0.12 and 0.35 for the March-April
and April-May flights, respectively (Figure 8). These relationships
indicate that an increase in fVEG leads to greater total ablation
relative to the open reference sites.
Elevation explains most of the variance in the RF model for the early
season ablation (March-April), followed by fVEG and LAI’ (RF r-squared
of 0.52; Figure A12). DSnowAb increases as elevation, fVEG, and LAI’
increase. An increase in openness predicts a general decrease in
DSnowAb, indicating that sites with more openness show more similarity
between ablation in the open and ablation under the canopy. fVEG also
controls late season ablation (April-May). fVEG and openness explain
most of the variance for late season ablation. High DSnowAb response
regions emerge between openness values between -3 (gap diameter 1/20
average tree height) and 0 (gap diameter and tree height are equal) and
fVEG values between ~0.5 and 0.9 (Figure A13).