In the South Island of New Zealand, seasonal snow contributes between 10 and 25% of annual runoff. It is therefore important to understand the controls on snow distribution and the extent to which these change from year to year. In the Southern Alps of New Zealand there has been limited assessment of the controls on snow accumulation and its inter-annual variability. This study investigates the physiographic controls on snow accumulation in an alpine catchment in the Southern Alps and how the importance of these controls varied over a four year study period. In addition, inter-annual variation in the synoptic-scale weather patterns is linked to the inter-annual variation in the physiographic controls and subsequently basin snow water equivalent and catchment baseflow.
This study utilized a four-year (2007-2010) snow depth and density dataset collected at the timing of peak accumulation in the Jollie Catchment, a seasonal snow dominated catchment in the central Southern Alps, New Zealand. Physiographic controls on snow depth distribution in the catchment were assessed using multiple linear regression and regression tree methods incorporating fourteen different terrain-based predictor variables. These variables were elevation, slope, east-aspect, north-aspect, radiation, wind exposure in the eight cardinal directions, and an average sheltering index. The synoptic weather during the accumulation seasons was characterized using pre-defined weather types (the Kidson Weather Types). In order to determine inter-annual variation in snowmelt timing, magnitude and duration, a baseflow separation technique was used on discharge data collected at a gauging station located at the outlet of the catchment.
Regression tree methods proved most successful in modelling the spatial distribution of snow depth in all four years because of their ability to capture non-linear interactions between the predictor variables. The physiographic controls on snow depth distribution were shown to vary between the four study years. Elevation was the dominant terrain-based predictor variable in 2007 and 2008 (explaining 88% and 74% of the respective regression tree R2 values), whereas the east-exposure index was the dominant terrain-based predictor variable in 2009 (23%) and 2010 (27%). The other wind exposure indices also had a large effect on the snow depth distribution in 2009 and 2010 (combined, they explained 76 and 75% of the variance in the models, respectively). The lowest and highest total snow water equivalent values were modelled in 2007 and 2010, respectively. The comparative difference between total snow water equivalent values in each year matched annual peak snow depth values recorded at the Rose Ridge climate station nearby.
Differences in the frequency and duration of the Kidson weather types between years can be linked to the changing importance of the terrain-based predictor variables in the regression tree models. In particular, a higher frequency of troughing events (2009 and 2010) led to a reduced importance of the elevation variable and greater influence of the wind exposure variables. Comparison with a simple temperature and precipitation-driven degree-day model shows the importance of including physiographic data when modelling snow water equivalent, particularly when the synoptic weather patterns lead to the elevation variable having a small degree of influence on snow depth distribution. Snowmelt patterns, shown in the baseflow sequence, appear to also be associated with the changing dominance of the elevation variable. A greater importance of the elevation variable in controlling snow depth distribution leads to a sharp increase and a clear peak in baseflow during spring and summer.
The identification of the inter-annual variability in the physiographic controls on snow distribution highlights the importance of multi-year datasets in snow distribution studies in alpine environments. This study also demonstrates the benefits of linking changes in the synoptic-scale weather patterns to variations in the controls on snow distribution and snowmelt. Understanding these links provide an insight into how snow distribution and snowmelt will respond to changes in these synoptic-scale weather patterns without the need for extensive snow surveys.
