Accounting for the different feedbacks described in the above section and physical processes implied in the future rain/snow transition rise requires to rely on complex models of SMB, and major improvements in glacier SMB projections can be achieved by focusing on the accumulation processes and modelling. State-of-the-art glacier models are in their vast majority relatively simple (Marzeion et al., 2020). They calculate glacier SMB with empirical relations between monthly or annual precipitation and accumulation on one side, and temperature and ablation on the other side. Consequently, such empirical models are likely to miss important feedbacks related to the strong influence of the snow/ice albedo on glacier SMB (Sicart et al., 2008). More physical approaches that explicitly describe the energy and mass exchanges between the glacier and the atmosphere are required to provide realistic future glaciological projections.

The use of simple glacier SMB models is often justified by their modest forcing needs. Simple models are forced with temperature and precipitation only, whereas physically based models are much more demanding in terms of input variables (radiation, wind, humidity are also needed). Forcing a physicallybased model requires atmospheric variables that are difficult to measure and simulate at the glacier scale, especially for precipitation (solid and liquid) which shows a high spatial and temporal variability in complex topography areas (Mott et al., 2014). Atmospheric models are ideal tools to provide an estimate of these physical variables over long periods, including in future projections. However, their resolution is generally too coarse to simulate the spatial heterogeneity of the meteorological variables for glacier SMB modelling. Therefore, atmospheric models need to be bias-corrected and downscaled
to get an estimation of precipitation rates and phase at the local scale. Statistical downscaling is a common method to get an estimate of the meteorological variables at the local scale.

Local observations of atmospheric variables over multi decadal periods in mountain areas are sparse, except in some regions, like the Alps. Long term monitoring programs are very costly and demanding to maintain, but they provide some of the longest series in remote areas. For instance, in the Andes, meteorological measurements have been done in the neighborhood of Zongo Glacier since the 1990s. While these networks capture some spatial and temporal variability of the temperature, they are missing most of the precipitation spatio-temporal variability (Junquas et al., 2018). This is due to the intrinsic challenges of precipitation measurements (high spatio-temporal variability, representativeness of point measurements, and inaccuracy of precipitation measurements, especially for solid precipitation [Kochendorfer et al., 2017]), that is intensified in mountain regions due to orographic effects and the additional complexity of measuring solid precipitation. For example, in the Indus basin, Immerzeel et al. (2015) showed that precipitation at high altitude was 5 to 10 times higher than the spatialized precipitation products. As a consequence, alternative and innovative methods that combine deployment of remote-sensing instruments, such as ground-based Micro Rain Radars (MRR) with space-borne observations are key to expand the existing network of observations. Radar systems on board satellites are very efficient tools to measure the temporal and spatial distribution of precipitations (quantity and phase) within the atmosphere, but they need to be validated with point observations. With this background, we argue that we need, on the one hand, to maintain existing networks, such as GLACIOCLIM observatory (CNRS-INSU/IRD), and on the other hand, to combine specific pointscale observations, such as a MRR, and other remote sensing techniques that help understanding the variability of the rain/snow transition at regional scale.

Updated on 6 April 2022