A multi-scale approach to assessing the spatio-temporal variability of seasonal snow in the Clutha Catchment, New Zealand

Abstract

Seasonal snow is an important, but under-observed component of New Zealand's hydrological cycle. Measurement and characterisation of seasonal snow is complicated because it varies over a range of spatial and temporal scales. This makes spatially distributed in situ observations difficult to acquire, and limits efforts to scale point-based observations up to larger areas. Sparse observations of seasonal snow lead to reduced understanding of seasonal snow processes, and subsequent uncertainty in efforts to model seasonal snow. This thesis addresses these issues within the Clutha Catchment, New Zealand's largest, by leveraging remote sensing and geospatial approaches to map and characterise seasonal snow both regionally, and at very high spatial resolution over a small alpine basin.

A daily snow covered area (SCA) time series and regional scale snow cover climatology is derived from MODIS imagery for the period 20002016. Metrics including annual snow cover duration (SCD) anomaly and daily SCA and snowline elevation (SLE) were derived and assessed for temporal trends. On average, SCA peaks in late June (~30 % of the catchment area), with 10 % of the catchment area sustaining snow cover for > 120 d yr -1. A persistent mid-winter reduction in SCA is attributed to the prevalence of winter blocking anticyclones in the New Zealand region. No significant decrease in SCD occurred over the period 20002016, but substantial spatial and temporal variability was observed. Raster principal component analysis (rPCA) identified distinct modes of spatial variability within the time series. Spatio-temporal variability extends beyond that associated with topographic controls, which can result in out of phase snow cover conditions across the catchment. Specific spatial modes of SCD are associated with anomalous airflow from the NE, E and SE. Furthermore, it is demonstrated that the sensitivity of SCD to temperature and precipitation variability varies significantly across the catchment. In order to resolve sub-MODIS scale processes, the potential of remotely piloted aircraft system (RPAS) photogrammetry to map snow depth was evaluated within an alpine catchment of the Pisa Range. Differencing between snow-covered and snow-free digital surface models (DSMs) acquired during 2016 provided high resolution snow depth maps. The accuracy of snow depth maps was thoroughly assessed with in situ snow probe measurements, and by analysing residuals for snow-free areas between DSMs. This accuracy assessment demonstrated repeatability and revealed substantial departures of errors from a normal distribution. This reflects the influence of DSM co-registration and terrain characteristics on vertical uncertainty. Error propagation provided lower uncertainties for snow depth (±0.08 m, 90 % c.l.) than the characterization of uncertainties on snow-free areas (±0.14 m). Comparisons between RPAS and in situ snow depth measurements confirm this level of performance. Semivariogram analysis revealed that the RPAS outperformed systematic in situ measurements in resolving fine-scale spatial variability.

Following the successful evaluation of RPAS photogrammetry for mapping snow depth, further snow depth maps were acquired for 2017. A total of six snow depth maps that resolved fine scale spatial variability in snow distribution facilitated the assessment of topographic controls on snow depth and snow water equivalent (SWE) distribution. Topographic controls were assessed via regression tree analysis between snow depth and terrain indices, including the kernel density of tussock vegetation (KDtussock), elevation (ELEV), the topographic position index (TPI), a Shade index (Shade), and Sx (maximum upwind slope). Despite substantial differences in both total snow volume and spatial distribution, the range of spatial-autocorrelation for snow depth was comparable for both winters at 20 – 30 m. Regression tree modelling reproduced some of the observed spatial structure, and demonstrated temporal variability in the relative importance of controlling parameters. The impact of varying wind regimes on the spatial distribution of snow was highlighted. These findings illustrate the complexity of atmospheric controls on SCD within the Clutha Catchment and support the need to incorporate atmospheric processes that govern variability of the energy balance, as well as the re-distribution of snow by wind in order to improve the modelling of future changes in seasonal snow. Despite limitations accompanying RPAS photogrammetry, this study demonstrates a repeatable means of accurately mapping snow depth for an entire, yet relatively small, hydrological catchment (∼0.4 km2) at very high resolution. Snow depth maps provide geostatistically robust insights into seasonal snow processes, with unprecedented detail. This thesis demonstrates the utility of mapping snow at differing spatial scales for improved understanding of seasonal snow processes and highlights the need to robustly capture dynamic processes in spatial snow models.