Abstract
The spatio-temporal distribution of seasonal snow in alpine terrain is strongly influenced by interactions between the atmospheric boundary layer and the snow surface. Wind transport of snow and variation in energy and mass fluxes are important controls on snow distribution across multiple scales which affect the timing and magnitude of snowmelt runoff. As pressure on water resources increases, there is an accelerated need for understanding and modelling snow cover processes to assess future changes to seasonal snow and predict the impact of such changes. However, lack of observational data, local hydrological knowledge and computational requirements often impose limitations on the modelling of snow cover and melt in New Zealand.
The aim of this research was to apply a hydrological model which incorporates a blowing snow routine to determine whether snow accumulation and melt of a seasonal snowpack can be determined using a spatially aggregated approach in an alpine valley. The Cold Regions Hydrological Modeling platform (CRHM) was used to create a physically based hydrological model for application in the Leopold Burn Tributary (0.4 km^{2}), located in the Pisa Range, New Zealand for the 2017 snow season. The model was implemented at a point scale and in a semi-distributed manner over landscape units known as hydrological response units (HRUs). A 5 m digital elevation model (DEM) and high resolution snow depth map observations, derived from a remotely piloted aircraft system (RPAS), were made available. These were paired with an available tussock density vegetation map which informed catchment delineation, model parameterisation, and provided observations for model assessment. Meteorological forcing data from the catchment weather station was paired with reconstructed precipitation data to drive the model at an hourly resolution and observations of daily streamflow were used for model evaluation.
Implementation of CRHM at a point scale revealed the model was highly sensitive to the phase partitioning of precipitation, and melt parameterisations. Default values used in Canadian settings, where the model was developed, needed adjustment for New Zealand conditions. Comparisons of observations and model estimates showed the model performed well at the point scale, with R^{2} values of 0.85 and 0.94, for snow depth and surface temperature, respectively. Root mean squared error (RMSE) between simulated and observed snow depth was 0.05 m while surface temperature achieved a value of 0.69°C. For the snow cover season, between June and September 2017, temperature at the weather station (1503 m.a.s.l.) fluctuated around a mean of -0.8°C, with melt events occurring in all months. Net radiation was the dominant source of energy for melt, providing 66% of melt energy, followed by sensible (29%) and latent heat (5%). While sensible heat was the most consistent contributor of energy to the snowpack, net radiation became dominant as the season progressed, which was primarily driven by increased shortwave radiation.
At the catchment scale, the model was able to capture the proportions of snow between sparsely and densely vegetated areas of the catchment but it underestimated snow depth and SWE overall. In particular, differences between HRU estimated and observed SWE were evident in the early season with greater snow accumulation observed on north facing slopes compared to south facing slopes while the model estimated the opposite. This was attributed to the rudimentary HRU setup of the fixed aerodynamic sequence, under which the model simulated blowing snow in the predominant (northerly) direction. In reality, the snow distribution pattern was spatially dynamic for the 2017 season and exhibited high variability. Results suggest a considerable amount of blowing snow was likely transported into the tributary from nearby areas due to the location and low topographic boundaries of the catchment. High levels of snow transport to incised areas of the catchment, like the gullies and valley floor also influenced the end-of-winter snow distribution and the timing of snowmelt runoff. Under-representation of these areas in the model resulted in overestimated peak flow, however, the model was able to capture the timing of the peak spring runoff.
The results show the spatial scale of the aggregated approach used here is too coarse to represent blowing snow and melt calculations in this type of environment. Such application has potential for larger areas of New Zealand or towards the Main Divide where coarser topographic information or greater contrast of physiographic variables exist and may be better suited to an aggregated approach. Prediction of catchment hydrology is possible without calibration if physically based models are used, yet physically based parameterisations and improved knowledge of catchment hydrology is required. The study has demonstrated the benefit of this modelling approach which provided a platform for diagnosing inadequacies in the understanding of local hydrological processes and indicated research priorities for improving snow accumulation and melt modelling in New Zealand.