SNOW ACCUMULATION MEASUREMENT

Study sites were established and field measurements undertaken during the austral winter of 2008. The Franz Josef Glacier study site (FJG) was established on 14 July, and measurements conducted over 24 days until 7 August. The Tasman Glacier study site (TG) was established on 17 July, and measurements conducted over 21 days, which were concurrent with the FJG site (Fig. 1). Both study sites were set up at approximately 2300 m a.s.l., on relatively flat snow surfaces in the center of the accumulation areas. Sites were chosen to be at least 100 m away from topography (ridges/depressions), which could potentially influence snow deposition patterns.

The FJG site was located in the Davis Snowfield (Fig. 2). There are no spatially comprehensive winter snow accumulation measurements for FJG. Previous net accumulation measurement shows that an area in close proximity to the winter site has recorded near average net accumulation (Purdie et al., 2010). The TG site was located close to a previous long-term mass balance measurement point (Chinn, 1994). Like at FJG, there have been no previous measurements of winter snow accumulation at high spatial resolution. However, a net accumulation survey in 2008 (Purdie et al., 2010) recorded a point nearby with lower than average net accumulation (Fig. 2).

FIGURE 2

The Franz Josef Glacier (A) and Tasman Glacier (B) accumulation areas, showing the location of the winter study sites in relation to previous net accumulation measurement.

At each study site four snowboards (0.4 m × 0.4 m) with central poles (2 m) were installed at 1 m intervals in a linear array aligned perpendicular to prevailing westerly flow (Fig. 3). Snowboards are a standard method for snow accumulation measurement (New Zealand Mountain Safety Council, 2008) and have been used to determine percentage under-catch for other precipitation gauges (Barry, 1992). However, a degree of under-catch is still likely, particularly during light snowfall accompanied by strong winds.

FIGURE 3

(A) Site layout at Franz Josef Glacier showing the four snowboards and automatic weather station. Photo taken looking to the southwest. (B) Snow depth measurement. The solid wooden ruler is easily inserted through the snow down to the wooden board. The central poles ensure relocation of boards after each snowfall.

Each of the four snowboards in the array was used to measure accumulation at different time scales. For the purposes of this paper, results are presented from two of these boards: the ‘total-board,’ at which net snow accumulation was measured on a daily basis, and which allowed measurement of snow loss as well as gain; and the ‘24 hour board’ at which daily snowfall was measured at 0900 hrs. Each time the 24 hour board was measured it was subsequently cleared of snow and repositioned on the snow surface. For each measurement, four separate depth measurements were made in each quarter of the boards and averaged (Fig. 3). Snow surface density was measured daily at the 24 hour board. Density was measured by weighing a 250 mL sample (taken with a Snowmetrics RIP 2 cutter) of fresh surface snow on a digital scale. Comparison of measurement between adjacent boards found snow depths to be within ±10 mm. Relative measurement errors are estimated as depth (±5%), mass (±0.5%), and volume (±2.5%), equating to a total relative measurement error of ±8%.

SNOW ACCUMULATION MODELING

Standard accumulation models, which predict snow accumulation using an elevation/temperature–based approach, have been found to provide effective estimates of snow accumulation on glaciers on seasonal and annual time scales (Oerlemans, 2001; Braithwaite et al., 2002; Anderson et al., 2006). Here we evaluate the usefulness of such models at a synoptic scale, by applying the model from Anderson et al. (2006). In this model, snow accumulation occurs on the glacier when precipitation occurs at a nearby low-elevation weather station and air temperature on the glacier is estimated to be below a snow/rain temperature threshold. A threshold of 1 °C was determined by best-fit modeling in New Zealand accumulation models (Moore and Owens, 1984; Barringer, 1989). Air temperature at the glacier site is calculated by applying a locally measured temperature lapse rate (−5 °C km⁻¹) to temperature measured at the low-elevation weather station (Anderson et al., 2006). A precipitation factor is applied to the low-elevation precipitation total, to account for orographic processes. For FJG, a precipitation factor (1.41) was empirically derived (Anderson et al., 2006) and for TG a precipitation factor of (1.46) was derived on the basis of interpolation of precipitation gradient maps (Henderson and Thompson, 1999).

Modeling was conducted using two configurations. The first model run (M1) utilizes the model parameters outlined above and is forced with hourly temperature and precipitation data from an automatic weather station (AWS) at Franz Josef (80 m a.s.l.) or Mount Cook (765 m a.s.l.) village for the FJG or TG site, respectively. These AWSs are owned and operated by the National Institute of Water and Atmospheric Science (NIWA). The second model run (M2) uses modified model parameters, including temperature lapse rates, based on AWS data measured at the low-elevation weather stations run by NIWA and AWSs installed at glacier study sites (see following section for AWS details). For M2, precipitation input data for each site are modified to better reflect regional variations in precipitation. This is done by combining precipitation from weather stations using distance weighting. A new measured precipitation factor is also applied to the distance-weighted precipitation. This derives a daily precipitation total for each site that is a reflection of precipitation recorded on both sides of the Southern Alps. For both model runs, hourly modeled precipitation is summed daily to enable comparison with daily (0900 hrs) measured snow accumulation.

SYNOPTIC CLASSIFICATION AND LOCAL CLIMATE

To assess the influence of synoptic type on snow accumulation, each day during the study was categorized as one of three synoptic regimes (troughing, zonal, or blocking; Fig. 4). Daily synoptic weather types were supplied by NIWA and classified according to the scheme of Kidson (2000). Kidson utilized 40-year NCEP/NCAR data to identify 12 different weather types for New Zealand, which were further categorized into zonal, troughing, or blocking regimes. Troughing regimes are characterized by frequent troughs and frontal systems crossing New Zealand, zonal regimes typically have high-pressure systems north of New Zealand with zonal flow to the south, and blocking regimes are characterized by high-pressure systems being more prominent in the south, stalling or slowing the passage of approaching weather systems (Kidson, 2000). Since their development, Kidson indices have been applied to studies of air pollution (Appelhans, 2009; Baldi et al., 2009), climate change (Lorrey et al., 2007; B. Mullan, personal communication), drought patterns (Salinger, 2009), and marine biodiversity (Dunn et al., 2009).

FIGURE 4

Examples of (A) zonal flow (15 July), (B) troughing (22 July), and (C) blocking (30 July) regimes at 00UTC during the study period. Synoptic analysis charts provided by Metservice, and Kidson diagrams from Kidson (2000).

Meteorological variables were measured and compared between the glacier study sites. Temporary AWSs were set up at each site, ∼4 m from the snowboard array, recording hourly temperature, humidity, and wind speed and direction (Fig. 3). Each site had a catch precipitation gauge (150 mL capacity) in case of liquid precipitation. Riming of AWS sensors occurred during storms, and both stations periodically required excavation and repositioning on the snow surface. Manual recordings (temperature, humidity, wind speed, and barometric pressure) were made daily at 0900 hrs at the sites with a Kestrel 4000 pocket weather tracker. Upper-level wind directions were determined by tracking cloud movement with a compass and used to confirm synoptic-scale flow. The degree of wind transport was categorized using the blowing snow scale from the New Zealand Mountain Safety Council (2008), which records snow transport as nil, light, moderate, or intense.

In addition data from an AWS situated on the mid–Tasman Glacier (1110 m a.s.l.), and data from the two low-elevation AWSs run by NIWA were used to consider local-scale climate variability. Data from the latter two were used to calculate the modified temperature and precipitation lapse rates for M2, and force both model runs.

SYNOPTIC VARIABILITY IN SNOW ACCUMULATION

Direct observations and the absence of any liquid catch in the rain gauges at the study sites indicated that all precipitation received during the study period at both sites was solid. Snowfall events occurred on 15 of 24 days at FJG, and 14 of 21 days on the TG (Table 1 and Fig. 5). During the period of concurrent measurement, FJG received more snow on 56% of the days when snow occurred, and under all synoptic regimes.

FIGURE 5

Daily measured snowfall recorded at 24-hour snowboard, cumulative daily snow, and snow depth at the total board (gains and losses) on Franz Josef and Tasman Glaciers. The study period is divided into synoptic regimes as per Kidson (2000). Snow was measured at 0900 hrs and assigned to the previous day as per standard meteorological convention.

TABLE 1

Snow accumulation measured at the 24-hr snowboard in relation to synoptic frequency, including the long-term synoptic average.

In order to determine the extent to which the 24-day study period was representative of mid-winter conditions in other years, daily Kidson synoptic types from 14 July–8 August for the previous 50 years (1958–2007) were compared to those occurring during this study. The long-term (1958–2007) average frequency of twice daily (00UTC, 12UTC) Kidson synoptic weather types shows that during this time of year, troughing is most frequent and zonal conditions occur the least (Table 1). During the study period, troughing was more frequent than mean conditions, but still within one standard deviation of the long-term mean. Twenty-three of the previous 50 years also experienced higher frequencies of troughing (e.g. 1962, 1965, 1985, and 1986). In terms of the wider synoptic picture, of the 50 years of record, 21 years experienced similar regime frequency to 2008; the remaining 29 deviated from the mean by more than 1 standard deviation (e.g. 1970, 1974, and 2003). The regime frequency in the study period is therefore inferred to be reasonably representative of synoptic conditions of this mid-winter period.

A comparison of daily meteorological variables during the different synoptic regimes shows that zonal conditions were generally warmer and less windy during the study (Fig. 6). Troughing was associated with westerly quadrant winds, cooler temperatures, high humidity, and low pressure at both sites, with wind speeds higher at FJG. Blocking resulted in a reversal of normal wind direction (easterly quadrant winds dominated) and cold, humid conditions were recorded on TG. Low humidity, cold temperatures and strong downslope winds were recorded at FJG during these periods.

FIGURE 6

Average daily meteorological variables measured at the FJG (dark gray) and TG (light gray) study sites under different synoptic regimes.

Of the 263 ± 21 mm water equivalent (w.e.) of daily snow accumulation recorded at FJG during the study, 77% occurred during troughing, 19% with blocking, and less than 10% during zonal conditions (Table 1). Likewise, at TG the majority of snowfall (76%) occurred during troughing regimes, with blocking conditions contributing the remainder. The later start date at TG meant that only two days of zonal flow were measured, and no snow accumulation was recorded on those days. Troughing conditions were in place over 50% of the time, and was the most persistent synoptic regime, sometimes lasting for five or more days.

By the end of the concurrent study period the western, FJG site had received 74 ± 6 mm w.e. (29%) more gross daily snowfall than the TG site, east of the Main Divide. However, on six occasions there was <2 mm w.e. difference in daily snow accumulation at both sites, which was below estimated measurement accuracy. On three occasions, when snow was recorded at both sites, the eastern glacier site recorded more snow accumulation (Fig. 5), in particular, on 28 July, when wind direction recorded at the site was from the northeast, and 12 mm w.e. more snow was recorded at TG in comparison to FJG. Excavation of a snow pit at the total boards indicated that at the end of the study, net accumulation was 231 ± 18 mm w.e. at FJG (indicating a loss of 22 ± 2 mm w.e.), and 217 ± 17 mm w.e. at TG. The net total at TG was 38 ± 3 mm w.e. more than the daily cumulative measurement (179 ± 14 mm w.e.). This difference between cumulative and net accumulation at TG could be influenced by a lack of density data recorded below 300 mm in the TG pit, due to weather and time constraints, or an indication that the total board at TG received more snow than the 24-hr board, which was not detected by depth measurements due to compaction. So overall, difference in net accumulation between the two sites was very small at only 14 ± 1 mm w.e. (6%).

REDISTRIBUTION OF SNOW BY WIND

During the study period, snow transport by wind was a regular occurrence, with 9 out of 24 days categorized as having ‘moderate’ or ‘intense’ snow transport. During intense transport it was difficult to determine how much snow was actually precipitating and how much was remobilized. Quantifying loss of snow is complex as changes to snow surface height can be a combination of wind deflation and compaction. In calm conditions, a reduction in snow surface height is expected due to snow compaction and settling (Dewalle and Rango, 2008), and during fine, calm weather, the snow surface height at the total board was measured to decrease by 10–20 mm per day. Density measurements of fresh (or surface) snow were taken on a daily basis, but a full density profile was only completed at the end of the study to determine overall net accumulation. Reductions in snow surface height, in terms of water equivalent, are therefore estimated by first subtracting potential surface reduction due to compaction, and increasing the estimated error (±2 mm w.e.) to account for the uncertainty in compaction rates.

Intense snow transport occurred during a SE storm on 30 July (Figs. 4 and 6). At FJG wind speeds up to 30 m s⁻¹ were recorded before the wind speed sensor failed. By 0900 hrs on 31 July there had been a reduction in snow surface height (recorded at the total board) at FJG of 195 mm (Fig. 5), equating to 16 ± 3 mm w.e. On the eastern (windward) side of the Southern Alps at TG, the reduction in snow surface height, during the same storm, was only 56 mm (8 ± 2 mm w.e.) (Fig. 5).

Although occasionally blowing from the SE, wind directions recorded at the TG site during the storm tended to come from N-NE-E directions, and at times, a W direction, with gust speeds also reaching 30 m s⁻¹. Lower down the glacier, the mid-Tasman AWS recorded winds from the S-SE with a maximum average hourly wind speed of 11 m s⁻¹. This indicates a degree of localized wind turbulence at the TG site. So although snow was observed blowing down and away from the accumulation area at FJG during this storm, it is likely that turbulence resulted in redeposition of transported snow at the TG site.

Under troughing or zonal regimes snow surface changes were small (≤3 mm w.e. per day), and not beyond that which can be accounted for by compaction alone. However during blocking regimes, total snow surface reduction at FJG was 30 ± 4 mm w.e., and despite an additional 16 ± 1 mm w.e. of snow falling, there was an overall reduction in snow depth of 14 ± 3 mm w.e. during this synoptic type. At TG, although 16 ± 3 mm w.e. of snow surface reduction was recorded, there were still gains of 32 ± 3 mm w.e., giving a positive total during the same synoptic conditions.

Sublimation can also result in snow loss, depending on temperature, humidity, and wind speed (Liston and Sturm, 1998; Wever et al., 2009). However, sublimation was not thought to be an important process during this study, as periods of snow surface height reduction corresponded with high humidity (>90%) and observed, extensive wind transport. Increases in snow density from wind packing (Sato et al., 2008) could account for some of the loss in snow surface height, but not of the magnitude of reductions recorded.

CALCULATING MODEL PARAMETERS

Local climate data measured at the sites enabled the assessment and modification of standard model (M1) parameters, including temperature lapse rates, snow/rain thresholds, and precipitation factors (Table 2).

TABLE 2

Model parameters utilized by the initial (M1) and modified (M2) model run based on data measured during the study period.

During the study, measured temperature lapse rates between low-elevation AWSs and glacier study sites were highly variable on an hourly scale, ranging from 0.1 to −10 °C km⁻¹, but had an average over the whole study period of −5.5 °C km⁻¹ and −5.7 °C km⁻¹ at FJG and TG, respectively. These values are similar to previously measured lapse rates in these catchments by Anderson et al. (2006) and Ruddell (1995) of −5.0 °C km⁻¹ and −5.4 °C km⁻¹, respectively. At FJG lowest average rates (−4.3 °C km⁻¹) occurred during zonal flow, with near-average lapse rates during troughing and blocking. At TG blocking produced the lowest average lapse rates (−4.7 °C km⁻¹), troughing near average rates, and zonal the highest (−6.8 °km⁻¹), although only two zonal days were measured at TG. Longer-term temperature monitoring on Tasman Glacier has found that throughout the course of a year, temperature lapse rates between Mount Cook AWS and mid-Tasman AWS vary seasonally, with lowest rates (<−4 °C km⁻¹) occurring in winter and highest (>−9 °C km⁻¹) in summer.

Although temperatures at TG remained below zero throughout, positive temperatures were recorded on occasion at FJG, including 4 °C during snowfall on 20 July. Snowfall during this event was light, lasting for ∼1–2 hours, and only resulting in 35 mm w.e. snow accumulation. Even so, this demonstrates that applying a 1 °C snow/rain temperature threshold will result in underestimation of snow accumulation. Kienzle (2008) conducted sensitivity analysis of snow/rain temperature thresholds in Alberta, Canada, finding a mean threshold of 2.6 °C. Likewise, measurement at mid-Tasman AWS in winter 2007 recorded snow accumulation events when air temperature was ∼2–3 °C. Therefore a higher snow/rain threshold of 2.5 °C is used in M2.

Measured average precipitation factors between the glacier study sites and lowland AWS over the course of the study period were 1.51 and 1.36 for FJG and TG, respectively. However, on a per storm basis, factors could be highly variable, often ranging from 0.7 to 3.9 and in some cases being >30 or <1 (i.e. the lowland weather station received more precipitation than the glacier). These average precipitation factors measured during the study were applied to the M2 model run (Table 2). Franz Josef AWS is actually closer to the TG site than Mount Cook AWS, so with the change to distance weighting precipitation from both low-elevation AWS in M2, more weight is given to precipitation recorded at the western AWS when calculating snowfall for TG. It is important to note that these modified model parameters are based on measurement made during this study. Measurement during other seasons (e.g. spring, autumn) may result in variations in snow/rain temperature thresholds and precipitation factors. Despite this, comparing results from M1 and M2 provides opportunity to identify which model parameters are most sensitive over short temporal scales, and whether tuning a simple model to site parameters actually provides noticeable improvement.

COMPARISON OF MEASURED AND MODELED SNOW

Comparison between measured snow accumulation at FJG with a standard accumulation model (Anderson et al., 2006) found that by the end of the simulation, both M1 and M2 had underestimated total snow accumulation by 27 mm w.e. and 15 mm w.e., respectively (Fig. 7), although the difference in M2 is below the estimated measurement error of ±21 mm w.e. Simple linear regression between modeled and measured snow accumulation gave a correlation coefficient of r²  =  0.59 for M1 and 0.70 for M2. On a daily basis results are variable. Both model runs underestimate accumulation during the first period of zonal flow. Orographic precipitation is not considered explicitly by the model, and although a precipitation factor is applied that accounts for some difference between low-elevation AWS and the glacier accumulation area, the factor is held constant under all synoptic regimes. On occasions <1 mm precipitation was recorded at the low-elevation Franz Josef AWS, yet >50 mm snowfall was recorded at the top of the glacier, suggesting the precipitation factor needs to be synoptically adjusted. M2 produced improved results for the first period of blocking (least windy), but there is still a large difference between modeled and measured precipitation with the blocking event at the end of July, although this was an exceptionally windy storm. During troughing both simulations overestimate (19th) and underestimate (22nd) snow accumulation. On the 23 July, large improvement is seen with M2, as a frontal system moved east across the Southern Alps, ceasing to rain at Franz Josef township, while rain continued at Mount Cook village. Overall, the largest differences between measured and modeled (either run) precipitation occurred during troughing, when differences exceed 50 mm w.e. for M1 and 30 mm w.e. for M2. Best model performance (least difference) in either run was during zonal flow.

FIGURE 7

Difference between modeled and measured snow accumulation (bars) and cumulative modeled and measured snow accumulation (lines) at the study sites with a simple accumulation model.

Simulation at the TG site also showed variable quality on an event-based scale (Fig. 7), but noticeable improvement was gained on some days by the additional precipitation data from Franz Josef AWS used in M2, with r²  =  0.31 and 0.79 for M1 and M2, respectively. The cumulative difference between modeled and measured precipitation with either run was very small (3 mm w.e.), with M1 slightly overestimating and M2 underestimating cumulative snow accumulation. Consequently the difference in both simulations falls well below the estimated accuracy of the manual measurement (±14 mm w.e.). On 28 July, large difference exists between modeled (either run) and measured precipitation at TG. This snowfall was accompanied by very light winds (<2.5 m s⁻¹) from NE-NW directions, meaning that wind redistribution cannot account for the difference. Temperature, both modeled and measured, was below zero, so neither the temperature lapse rate nor the snow/rain threshold can account for the difference. In other words, the local nature of precipitation means that it is difficult to simulate all events using lowland weather station data and a precipitation factor, even using values that are specifically tuned for the glacier in question.

SNOW ACCUMULATION AND SYNOPTIC REGIME

This study aimed to simultaneously quantify snow accumulation on glaciers east and west of the Southern Alps in order to gain better understanding of relationships between synoptic weather systems and snow accumulation processes. As expected, the western and predominantly windward site on the Franz Josef Glacier recorded more snow accumulation than the eastern, predominantly lee, site on Tasman Glacier. But by the end of the study, there was little difference in net accumulation between the two sites, attributed at least in part to wind deflation.

During the study period, some of the largest snowfalls occurred with northwesterly conditions during troughing regimes, and both sites received >75% of their total measured snowfall during troughing events. The largest difference in snow accumulation east and west of the Southern Alps occurred under blocking conditions, which resulted in snow loss from FJG during strong southeast winds. On a regional scale, blocking is not necessarily associated with strong winds (Kidson, 1994), but in this case, the stalling of low-pressure systems east of the South Island resulted in large pressure gradients, generating very strong downslope winds west of the Southern Alps. Zonal circulation occurred least during the study period, and also resulted in the least snow accumulation.

New Zealand's maritime climate means that snow accumulation can occur year-round, but seasonal shifts in freezing levels result in the majority of snow accumulation occurring during winter and spring. As this study was conducted during winter, a question remains about relationships between synoptic weather systems and spring snowfall patterns. Synoptic frequencies in the New Zealand winter are similar to those occurring in spring, with larger differences to frequencies in the autumn and summer seasons (Kidson, 2000). Although it would have been desirable to undertake this experiment over a whole year, results from this winter survey, in terms of proportions of snow accumulation received under different synoptic regimes, still provides some insight into the potential impact that a change to synoptic frequency during the accumulation season could have to snow accumulation on glaciers.

Variation in synoptic frequency is driven by changes in atmospheric circulation patterns (Sturman and Tapper, 2006). In New Zealand, ENSO/Interdecadal Pacific Oscillation (IPO) phases have been linked to anomalous southwesterly (El Niño, negative ENSO, positive IPO) and northeasterly (La Niña, positive ENSO, negative IPO) circulation anomalies (Salinger et al., 2001). La Niña (El Niño) conditions in the Pacific are associated with a reduced (increased) frequency of troughing regimes and higher (lower) frequency of blocking circulation in New Zealand (Kidson, 2000). On longer time scales, increased troughing frequency was identified as an important factor associated with cooling and increased precipitation in New Zealand during the Little Ice Age (Lorrey et al., 2008). The recent negative phase shift in the IPO (Folland, 2008) indicates that the Southern Alps are likely to become warmer and receive less precipitation in future years than they did between 1977 and 2001 (Salinger et al., 2004).

Associations between synoptic frequencies, atmospheric circulation, and snow accumulation demonstrated in this and other studies are important on longer time scales. During the last few decades, glacier advances in Patagonia have been linked to increased westerly circulation and reduced blocking events, associated with positive IPO (negative ENSO) phases, whereas an anti-phase relationship exists between New Zealand and tropical Andean glaciers (Fitzharris et al., 2007). Similar relationships are evident worldwide. For example, increased troughing synoptic regimes have been associated with higher rates of snow accumulation in the Peace River Basin, Canada (Romolo et al., 2006), and on the northern Great Plains, U.S.A. (Grundstein, 2003). Similarly, positive phases of the NAO have resulted in increased winter accumulation on western Norwegian glaciers (Nesje et al., 2000; Nordli et al., 2005), and an inverse mass balance response is seen between Wolverine Glacier, Alaska, and South Cascade Glacier, Washington, U.S.A., in association with phase changes of the Pacific Decadal Oscillation (PDO) (Josberger et al., 2007). In conclusion, one must consider the role of regional (synoptic-scale) climate variability and its affect on mass balance when considering the future response of temperate glaciers, as the direction of change may enhance or buffer the anticipated negative mass balance of glaciers.

MODELING SNOW ACCUMULATION

While there are a number of complex snow accumulation models available (Liston and Elder, 2006; Lehning et al., 2008) which would likely improve snow accumulation estimation on glaciers, neither the meteorological input data (i.e. multi-level wind speeds from alpine catchments) to drive these models, nor the detailed accumulation measurements to calibrate and validate these models exist, except for a few specialized studies (Liston and Sturm, 1998; Mott et al., 2008; Dadic et al., 2010). Many of the worlds regions suffer from paucity of data—for example, South America (Popovnin et al., 1999), China (Sakai et al., 2009), and the Himalaya (WGMS, 2008). Therefore our data provided an opportunity to evaluate the ability of a simple snow accumulation model to capture synoptic-scale variability.

Troughing synoptic regimes accounted for ≥75% of measured snow accumulation during the winter study period. However, the largest difference between modeled and measured precipitation also occurred during troughing regimes, which were most frequent (Fig. 7). Total snow accumulation during blocking regimes was underestimated, with largest difference coinciding with snow accumulation during high winds. Simulation of snow accumulation under zonal flow was most successful, although it is important to note that this synoptic regime occurred least frequently, and overall contributed least to total snow accumulation.

Model performance was better at the eastern (TG) site, where after 21 days there was only a small (2%) difference between measured and modeled snow accumulation. It is likely that wind redistribution of snow was a factor in performance at the western FJG site, with biggest discrepancies between modeled and measured snow accumulation often associated with strongest wind speeds. Wind redistribution not only potentially removed snow from the site (e.g. 19 July), but also enhanced deposition (e.g. 22 July).

The importance of redistribution and preferential deposition of snow by wind has been quantified on complex glaciated terrain in the European Alps (Liston and Sturm, 1998; Lehning et al., 2008; Mott et al., 2008; Dadic et al., 2010). Interaction between wind dynamics and topography tend to result in increased snow deposition on lee slopes and reduced deposition on exposed slopes (Liston and Sturm, 1998). Strong downslope (SE) winds under blocking conditions resulted in significant loss of snow from the FJG site, but during the same period the TG site recorded both wind deflation and deposition, with complex topography resulting in turbulence, generating winds contrary to the general synoptic flow.

Aside from wind transport, event-scale variability in precipitation factors between low-elevation weather stations and high-elevation sites means that accurate snow accumulation estimation cannot be achieved by using a single precipitation factor. Applying a range of precipitation factors based on synoptic regime may provide improvement. Likewise, increasing precipitation input data to include a larger number of surrounding weather stations (especially at high elevation) would better capture regional precipitation patterns.

During this study, cold winter temperatures at the low-elevation AWS meant that the model was not very sensitive to choice of lapse rate, or snow/rain temperature thresholds. In the New Zealand Southern Alps, snow accumulation can occur year-round, with some of the largest snowfalls in spring, when warm moist air masses cross the Tasman Sea and intercept the mountain range. On Tasman Glacier, temperature lapse rates between low-elevation AWS and glacier AWS have been found to vary seasonally, with largest rates during summer and smallest in winter. Kienzle (2008) recorded seasonal variations in snow/rain temperature thresholds in Alberta, Canada, with mean monthly maximums (∼5 °C) occurring in winter, and minimums (0 °C) in summer. How such thresholds vary in the Southern Alps is unclear, but during the study, snowfall occurred at the FJG site when the temperature was 4 °C. So although these two model parameters did not account for inaccuracies during this winter simulation, such may not be the case if the study were repeated in spring.

In the absence of increased meteorological observation allowing application of higher order models, this study demonstrated that a simple snow accumulation model did provide reasonable approximation of cumulative snow accumulation, but still failed to capture event-scale variability. Interestingly, there was only small cumulative improvement by tuning model parameters specifically to the sites (M2), thus highlighting the difficulty in utilizing input data measured at low-elevation weather stations to estimate snow accumulation at a single point on a glacier. Utilizing more precipitation input data, and gaining a better understanding of precipitation relationships between low-elevation meteorological sites and glacier catchments under varying synoptic flow is essential to improve snow accumulation simulation. The challenge is to find ways to refine models without large increases to input data and computational requirements.

Spatial variability in snow accumulation on glaciers is important to both accurate mass balance measurement and modeling (Fountain and Vecchia, 1999; Jansson and Pettersson, 2007). This study has demonstrated links between synoptic frequencies and snow accumulation, but in order to be applied to the glacier as a whole, an assessment of how snow accumulation is spatially distributed on these glaciers is essential and is part of ongoing research (Purdie et al., 2010).