1 Introduction
In the regions of Central Asia, owing to the arid and continental climate, snow cover melt provides an important contribution to runoff, determining the availability of freshwater for the local populations. Snow can, however, also act as a source of natural hazards, an issue which can be further exacerbated by climate change and human intervention (Dietz et al., 2014; Han et al., 2019). In Kazakhstan, snowmelt represents the primary source of water for agricultural and industrial use, dwarfing the contributions from liquid precipitation or groundwater; Kazakh rivers are characterized by large variability in annual runoff, with the largest annual discharges exceeding average annual runoff by 5-7 times and more than 75% of annual runoff occurring in a short period in early spring (Kozhakhmetov and Nikiforova, 2016). Floods caused by late winter and spring snow melt are one of the most severe natural hazards in the country, displacing 10-30,000 people each year and causing on average more than $30 million in damages (UNOCHA, 2016; Guha-Sapir et al., 2018) through loss of infrastructure, crops and livestock. Floods affect both lowland and mountain rivers, and there is evidence that the number of extreme events affecting mountain rivers has increased by 80% in recent years (Kozhakhmetov and Nikiforova, 2016). This issue is particularly relevant in the East Kazakhstan region, which was one of the regions most affected by flooding between 1967 and 1990 and the most affected between 1991 and 2015, with over 42 floods reported (Kozhakhmetov and Nikiforova, 2016).
In spite of the role of snowmelt for the hydrology of Kazakh rivers, comparatively little is known about large scale snow cover variations from year to year. While Mashtayeva et al. (2016) analysed the spatiotemporal distribution of snow depth and snow cover duration in all Kazakhstan, a detailed analysis of the patterns of snow cover depletion is necessary for the mitigation of floods, water management strategies and conservation of river ecosystems. The improvement of short- and long-term runoff models further requires an assessment of the relationship between snow cover variability, snowmelt and meteorological variables (Kuchment and Gelfan, 2007; Kang and Lee, 2014; Kult et al., 2014) and the prediction of timing and magnitude of peak runoff. Temperature exerts the greatest control on snow cover extent and duration (Bednorz, 2004; Hantel et al., 2000; Mote, 2006; Tang et al., 2017), although the spatial variation of the temperature-snow cover relationship with elevation is less well understood (Gurung et al., 2017). In addition, temperature anomalies and snow cover across Eurasia are related to large-scale atmospheric circulation patterns via feedback mechanisms (Cohen et al., 2012). Eurasian snow cover variability in different seasons has been linked with the phases of the North Atlantic Oscillation (NAO) and Arctic Oscillation (AO), Eurasian and Siberian pressure patterns and sea ice extent in the Barents-Kara Sea (Cohen & Barlow, 2005; Wegmann et al., 2015; Ye and Wu, 2017). Further still, spring snow retreat is also known to influence the East-Asian climate of the following summer by controlling the strength of the Indian summer monsoon (Zhang et al., 2017). However, the links between teleconnection patterns, temperature anomalies and snow cover retreat at the catchment scale and possible lagged effects of large-scale atmospheric circulation, which would provide further indications for long-range forecasts, remain to be fully explored.
The size of the East Kazakhstan region and the sparse network of meteorological stations (Mashtayeva et al., 2016) means investigations on snow cover cannot be realistically undertaken using field data, and leaves remote sensing as the most viable option. Information on snow cover extent from optical sensors such as MODIS (MODerate resolution Imaging Spectroradiometer) and AVHRR (Advanced Very-High-Resolution Radiometer) at moderate spatial resolution (500-1100 m per pixel) and at daily to weekly time scales has been employed for studying snow cover variability across Eurasia (Dietz et al., 2012; 2013; 2014). AVHRR data represent the longest temporal record from 1978 to the present, but requires additional processing for snow/cloud cover discrimination and the quality of the observations decreases significantly with the oldest generation of sensors (Dietz et al., 2014). Recently, the availability of Sentinel-2 data has opened up the possibility for investigating snow cover at a much higher spatial resolution (10-20 m) at weekly time scales (Hollstein et al., 2016), while Sentinel-3 continues the legacy of MODIS and AVHRR by providing data at moderate resolution (300 m) in large swaths. However, no long-term (i.e. decadal) records exist for the Sentinel satellites as the first was only launched in 2015. Alternative snow depth and snow water equivalent (SWE) datasets are available from passive microwave sensors (Pulliainen, 2006) or GRACE gravimetric data (Wang and Russell, 2016), though the spatial resolution of these products is much coarser, e.g. 25 km for the ‘Globsnow’ product (Luojus et al., 2013) and 1º for GRACE (Landerer & Swenson, 2012). In addition, passive microwave data are known to be unsuitable for snow cover detection in mountainous regions (Luojus et al., 2013), which form a large part of the study area. In this study, MODIS was chosen as the main data source, as it provides an easily accessible archive of snow cover data from 2000 to the present, produced through a robust methodology and requiring minimal additional processing (Hall & Riggs, 2007).
This study focuses on the analysis of snow cover variability in one of the main water catchments of Kazakhstan, using the MODIS MOD10A2 dataset. The aims of this study are: i) To investigate patterns of spring snow cover change in five sub-basins of the Upper Irtysh river catchment, including the magnitude and timing of early/late peak snow cover depletion rates and timing of snow cover disappearance at different elevations, and ii) to investigate the correlations between variability in snow cover, air temperature and atmospheric pressure patterns. The Irtysh basin is used as a case study to evaluate the general applicability of this approach to understanding the linkages between large scale snow cover change and runoff.