**4. Discussion**

Detection of snow from MODIS data has proven to exceed 90% accuracy on clear-sky days [44,48,53], but cloud coverage remains a major problem for all optical satellites. The temporal GF and CA (similar to those in [11,44]) applied to the MODIS data resulted in promising cloud reductions of up to 100%. This success is largely attributed to the SH, which enables a high number of clear-sky days over Mongolia (Figure 5). The obtained So, S m, and SCD did not indicate any significant trends since 2000, while mid-season winter (DJF) air temperature showed a decreasing trend over the last three decades (Figure 7) associated with the weakening of SH [8–10]. The SH starts to form in early autumn and causes early forming snow cover due to its contribution to changes in temperature and precipitation [6,8,9]. Therefore, the increased delay (statistically not significant) in So might be attributed to the seasonal strengthening of the SH. However, more detailed investigations on extreme weather and precipitation events in conjunction with the SH are necessary to draw strong conclusions.

In this study, we applied a simple alternative altitude-dependent correction to Landsat and Sentinel images to produce a time series of snow distribution in the Sugnugur catchment. Results showed the development of snow cover during the snow season in time and space and the classified seasonal intermittent and persistent SCAs (Figure 9). Classifying persistent SCA from intermittent SCA is important for more accurate hydrologic and water balance modelling. Since the temporal resolution of the produced snow maps from Landsat and Sentinel images is still coarse, SCD for intermittent SCA remains uncertain, and thus further investigations should also include the timing of intermittent snow cover where ground measurements are not available.

We also demonstrated the altitude-dependent SCD rate to be +6 days/100 m and is in good agreemen<sup>t</sup> with other rates across other Central Asian Mountains [7]. The overall accuracy of SCD was ~91% with the mean underestimation of ~12–13 days/year, which was probably caused by the reduced MODIS snow detection rate for thin occasional intermittent snow cover that occurs in autumn and spring months. The snow field measurements supported more comprehensive snow metrics, such as SWE and SD in the study area. In general, no changes in SD with increasing elevation were evident over the catchment. However, SD can vary at local scales because of forest interception, snow redistribution, as well as different sublimation rates [55], and it should be studied in future investigations.

As snow cover contributes to the surface energy balance [56], it appears that the daily mean surface temperature (Ts) of < −1 ◦C could be a good approximation of the overlaying existing persistent snow cover. Nevertheless, the applicability of using Ts for defining seasonal persistent SCD might be limited to certain spatial extents in cold regions because of the high climatic variability and characteristics of snow cover in the high altitudes and latitudes of the Northern Hemisphere. The Ts from iButtons also indicated that the changes in land-cover alter the timing of snow-melt processes, showing an earlier Sm in burned forest relative to unburned forest [57]. Therefore, the consequent effect of the earlier Sm in conjunction with energy exchange and long term snow monitoring, together with continuous river discharge measurement, should be discussed in further studies for observing the contribution of snow-melt water in the regional water resources.
