Dr. Michael Schirmer

This study presents a data-driven reconstruction of daily runoff that covers the entirety of Switzerland over an extensive period from 1962 to 2023. To this end, we harness the capabilities of deep-learning-based models to learn complex runoff-generating processes directly from observations, thereby facilitating efficient large-scale simulation of runoff rates at ungauged locations. We test two sequential deep-learning architectures: a long short-term memory (LSTM) model, which is a recurrent neural network able to learn complex temporal features from sequences, and a convolution-based model, which learns temporal dependencies via 1D convolutions in the time domain. The models receive temperature, precipitation, and static catchment properties as input. By driving the resulting model with gridded temperature and precipitation data available since the 1960s, we provide a spatiotemporally continuous reconstruction of runoff. The efficacy of the developed model is thoroughly assessed through spatiotemporal cross-validation and compared against a distributed hydrological model used operationally in Switzerland.
The developed data-driven model demonstrates not only competitive performance, but also notable improvements over traditional hydrological modeling in replicating daily runoff patterns, capturing interannual variability, and discerning long-term trends. The resulting long-term reconstruction of runoff is subsequently used to delineate substantial shifts in Swiss water resources throughout the past decades. These are characterized by an increased occurrence of dry years, contributing to a negative decadal trend in runoff, particularly during the summer months. These insights are pivotal for the understanding and management of water resources, particularly in the context of climate change and environmental conservation. The reconstruction product is made available online.
Furthermore, the low data requirements and computational efficiency of our model pave the way for simulating diverse scenarios and conducting comprehensive climate attribution studies. This represents a substantial progression in the field, allowing for the analysis of thousands of scenarios in a time frame significantly shorter than those of traditional methods.

Restwasser bedeutet für die Wasserkraft eine Produktionsminderung, für die Gewässerökologie ein Existenzminimum. Bei der Verankerung der Restwasserbestimmungen im Schweizer Gewässerschutzgesetz setzte sich vor über 30 Jahren ein Ansatz durch, der zu niedrigeren Mindestrestwassermengen führt, als ökologische Untersuchungen empfahlen. Angesichts klimatisch bedingter Abflussveränderungen, dem Biodiversitätsrückgang und divergierender Ansprüche an die Ressource Wasser stellt sich die Frage nach der angemessenen Restwassermenge erneut.

Niederschlagsmessungen neigen dazu, den Schneefall zu unterschätzen, was als «Undercatch» bezeichnet wird. Dieser Effekt tritt auf, weil Messinstrumente das Windfeld verändern, und so nicht ausreichend Schneeflocken in das Messgerät gelangen. Aufgrund dieses Undercatch-Effekts kann die Messung des Schneefalls um etwa 40 Prozent zu niedrig ausfallen. Bei gegitterten Produkten wie etwa RhiresD kommt hinzu, dass es in höheren Lagen wenig Messgeräte gibt. In der Schweiz sind jedoch etwa 450 tägliche Schneehöhenbeobachtungen und -messungen mit Fokus auf hochgelegene Regionen verfügbar. Dieser Datensatz wurde verwendet, um den festen Niederschlagsanteil des weit verbreiteten Schweizer Niederschlagsgitters RhiresD zu korrigieren. Der feste Niederschlagsanteil von RhiresD wurde basierend auf stündlichen Lufttemperaturdaten bestimmt. Dazu wurden COSMO-Reanalysedaten genutzt, um tägliche Gitter von Niederschlags- und Lufttemperaturbeobachtungen in Stundenwerte umzurechnen. Dies begrenzte den Analysezeitraum auf sieben Jahre von September 2015 bis August 2022. Der Festniederschlag wurde zudem mit einem parametrischen Schneemodell ermittelt, das auf den Schneehöhenbeobachtungen basiert. Beide Datensätze wurden anschliessend kombiniert mit einer Datenassimilationsmethode namens Optimale Interpolation. Tagesgenaue Korrekturen für die vergangenen sieben Jahre, wie auch Statistiken für klimatologische Anwendungen konnten ermittelt und überprüft werden. Mit diesen Korrekturen kann insbesondere in hohen Lagen eine Verbesserung der Schneemengenschätzung erzielt werden.

The seasonal evolution of snow cover has significant impacts on the hydrological cycle and microclimate in mountainous regions. However, snow processes also play a crucial role in triggering alpine mass movements and flooding, posing risks to people and infrastructure. To mitigate these risks, many countries use operational forecast systems for snow distribution and melt. This paper presents the Swiss Operational Snow-hydrological (OSHD) model system, developed to provide daily analysis and forecasts on snow cover dynamics throughout Switzerland. The OSHD system is a sophisticated snow hydrological model designed specifically for the high-alpine terrain of the Swiss Alps. It leverages exceptional station data and high-resolution meteorological forcing data, as well as various reanalysis products to combine snow modeling with advanced data assimilation and meteorological downscaling methods. The system offers models of varying complexity, each tailored to specific modeling strategies and applications. For snowmelt runoff forecasting, monitoring snow water resources, and research-grade purposes, the OSHD system employs physics-based modeling chains. For snow climatological assessments, a conceptual model chain is available. We are pleased to present two comprehensive datasets from the conceptual and physics-based models that cover the entirety of Switzerland. The first dataset comprises a snow water equivalent climatology spanning 1998-2022, with a spatial resolution of 1 km. The second dataset includes snow distribution and snow melt data spanning 2016-2022 at a high spatial resolution of 250 m. To meet the needs of a multi-purpose snow hydrological model framework, the OSHD system employs various strategies for process representation and sub-grid parameterizations at the snow-canopy-atmosphere interface, particularly in complex terrain. Recent and ongoing model developments are aimed at accounting for complex forest snow processes, representing slope and ridge-scale precipitation and snow redistribution processes, as well as improving probabilistic snow forecasts and data assimilation procedures based on remote sensing products.

Climate projection studies of future changes in snow conditions and resulting rain-on-snow (ROS) flood events are subject to large uncertainties. Typically, emission scenario uncertainties and climate model uncertainties are included. This is the first study on this topic to also include quantification of natural climate variability, which is the dominant uncertainty for precipitation at local scales with large implications for runoff projections, for example. To quantify natural climate variability, a weather generator was applied to simulate inherently consistent climate variables for multiple realizations of current and future climates at 100 m spatial and hourly temporal resolution over a 12×12 km high-altitude study area in the Swiss Alps. The output of the weather generator was used as input for subsequent simulations with an energy balance snow model. The climate change signal for snow water resources stands out as early as mid-century from the noise originating from the three sources of uncertainty investigated, namely uncertainty in emission scenarios, uncertainty in climate models, and natural climate variability. For ROS events, a climate change signal toward more frequent and intense events was found for an RCP 8.5 scenario at high elevations at the end of the century, consistently with other studies. However, for ROS events with a substantial contribution of snowmelt to runoff (> 20 %), the climate change signal was largely masked by sources of uncertainty. Only those ROS events where snowmelt does not play an important role during the event will occur considerably more frequently in the future, while ROS events with substantial snowmelt contribution will mainly occur earlier in the year but not more frequently. There are two reasons for this: first, although it will rain more frequently in midwinter, the snowpack will typically still be too cold and dry and thus cannot contribute significantly to runoff; second, the very rapid decline in snowpack toward early summer, when conditions typically prevail for substantial contributions from snowmelt, will result in a large decrease in ROS events at that time of the year. Finally, natural climate variability is the primary source of uncertainty in projections of ROS metrics until the end of the century, contributing more than 70 % of the total uncertainty. These results imply that both the inclusion of natural climate variability and the use of a snow model, which includes a physically based process representation of water retention, are important for ROS projections at the local scale.

The snow cover spatial variability in mountainous terrain changes considerably over the course of a snow season. In this context, fractional snow-covered area (fSCA) is an essential model parameter characterizing how much ground surface in a grid cell is currently covered by snow. We present a seasonal fSCA algorithm using a recent scale-independent fSCA parameterization. For the seasonal implementation, we track snow depth (HS) and snow water equivalent (SWE) and account for several alternating accumulation–ablation phases. Besides tracking HS and SWE, the seasonal fSCA algorithm only requires subgrid terrain parameters from a fine-scale summer digital elevation model. We implemented the new algorithm in a multilayer energy balance snow cover model. To evaluate the spatiotemporal changes in modeled fSCA, we compiled three independent fSCA data sets derived from airborne-acquired fine-scale HS data and from satellite and terrestrial imagery. Overall, modeled daily 1 km fSCA values had normalized root mean square errors of 7 %, 12 % and 21 % for the three data sets, and some seasonal trends were identified. Comparing our algorithm performances to the performances of the CLM5.0 fSCA algorithm implemented in the multilayer snow cover model demonstrated that our full seasonal fSCA algorithm better represented seasonal trends. Overall, the results suggest that our seasonal fSCA algorithm can be applied in other geographic regions by any snow model application.

Das Klima wird sich in der Schweiz im aktuellen Jahrhundert ändern. Davon ist auch die Wasserkraft potentiell betroffen. Diese Studie präsentiert mögliche Szenarien für Einzugsgebiete mit Wasserkraftwerken im Portfolio des Elektrizitätswerks der Stadt Zürich (EWZ) eine Auf Basis der CH2018 Klimamodelle werden mit dem hydrologischen Modell PREVAH zukünftige Abflüsse simuliert. Dabei wird das künftig veränderte Klima (Temperatur, Niederschlag, Strahlung, Wind) sowie die schmelzenden Gletscher einbezogen. Folgende Gebiete wurden betrachtet: Bergell, Mittelbünden, Hinterrhein, OFima und OFible, Wägital und die Kraftwerke Letten, Högg und Wettingen in der Limmat.

The spatial distribution of snow water equivalent (SWE) and melt are important for estimating areal melt rates and snow-cover depletion (SCD) dynamics but are rarely measured in detail during the late ablation period. This study contributes results from high-resolution observations made using large numbers of sequential aerial photographs taken from an unmanned aerial vehicle on an alpine ridge in the Fortress Mountain Snow Laboratory in the Canadian Rocky Mountains from May to July in 2015. Using structure-from-motion and thresholding techniques, spatial maps of snow depth, snow cover and differences in snow depth (dHS) during ablation were generated in very high resolution as proxies for spatial SWE, spatial ablation rates and SCD. The results indicate that the initial distribution of snow depth was highly variable due to overwinter snow redistribution; thus, the subsequent distribution of dHS was also variable due to albedo, slope/aspect and other unaccountable differences. However, the initial distribution of snow depth was 5 times more variable than that of the subsequent dHS values, which varied by a factor of 2 between the north and south aspects. dHS patterns were somewhat spatially persistent over time but had an insubstantial impact on SCD curves, which were overwhelmingly governed by the initial distribution of snow depth. The reason for this is that only a weak spatial correlation developed between the initial snow depth and dHS. Previous research has shown that spatial correlations between SWE and ablation rates can strongly influence SCD curves. Reasons for the lack of a correlation in this study area were analysed and a generalisation to other regions was discussed. The following questions were posed: what is needed for a large spatial correlation between initial snow depth and dHS? When should snow depth and dHS be taken into account to correctly model SCD? The findings of this study suggest that hydrological and atmospheric models need to incorporate realistic distributions of SWE, melt energy and cold content; therefore, they must account for realistic correlations (i.e. not too large or too small) between SWE and melt in order to accurately model SCD.

Snow is an important component of the water balance of many mountain watersheds worldwide. In a warming climate, snowmelt modeling and consequent soil water input, is often challenged by complex conditions such as rain-on-snow situations. This is why detailed physics-based snow models are increasingly being used. These models however have much higher input data requirements, where in many cases accurate forcing fields are very difficult to provide.
This study investigates whether the latest advances in the development of snow model framework actually translate into improved discharge simulations. To this end we integrated a distributed multi-layer energy-balance snow model with two recently developed methods of updating snow model mass and energy fluxes using snow observations to improve snow accumulation and depletion predictions. Surface water input from these simulations was used as input for subsequent streamflow modeling of 25 catchments in the Swiss Alps over four hydrological years.
Our analysis clearly demonstrates the benefits of accurate snow simulations for hydrological modeling in Alpine catchments. Simulations that included the flux updates improved streamflow predictions, and offered best performance at high elevation, where snow most prominently affected watershed hydrology. These results were consistently achieved when analyzing model performance over entire hydrological years, over the snowmelt season only, and for individual events.

 

The reliance on distributed energy- and mass-balance snow models as runoff forecasting tools has been increasing. Compared to traditional, conceptual forecasting approaches, these physics-based tools are robust to conditions that deviate from historic norms and offer improved performance in potentially dangerous rain-on-snow events. The physics-based simulations, however, depend on a large suite of accurate forcing data. Current numerical weather prediction products are capable of supplying the full range of required data, but systematic biases are often present. Data assimilation presents a means of compensating for such errors as well as potential snow model errors, yet currently available data assimilation techniques have limited usefulness in these snow models. This study introduced an alternative technique that similarly uses observations to update and improve simulations. As such, it is the first method to incorporate point snow observations in a fully distributed, physics-based, multilayer snow model while conserving mass and maintaining physically consistent layer states that are in accord with observations. At the core of this technique is an ensemble of predetermined perturbations to the model forcings termed the bias-detecting ensemble. Ensemble members were evaluated using observed snow depths to ascertain potential biases at nearly 300 sites across Switzerland. The bias assessments were distributed to 38 independent sites and incorporated into the model. Tests were conducted over three winter seasons using two numerical weather prediction-based products with varying quality. Averaged across the 38 sites and 3 seasons, the Nash-Sutcliffe efficiency score for bias-detecting ensemble-corrected snow depth was 0.98 compared to 0.81 without the bias-detecting ensemble method.

The snow cover exhibits tremendous spatial and temporal variability, especially in complex topography. Fractional snow-covered area (fSCA) is a parameter used to describe how much ground is covered by snow. As such, fSCA is a relevant parameter in large-scale model applications, for instance to compute the surface radiation balance or for scaling snow melt runoff. Many studies therefore focused on formulating parameterizations for fSCA, requiring mean snow depth estimates and knowledge about the underlying topography. With the emergence of high-resolution satellite products, remotely-sensed fSCA estimates are becoming more readily available, opening new avenues to assimilate fSCA in models. Nevertheless, parameterizations are still required to fill the gap when satellite data are unavailable (e.g. between satellite revisits, when clouds obscure the ground and for forecasting). We therefore evaluated fSCA estimates from a recently proposed fSCA parameterization developed for peak of winter snow data with satellite fSCA from Sentinel-2 images generated by Theia over an entire winter season for Switzerland. Modelled fSCA maps were obtained from a 1 km gridded operational hydrological model that assimilated several hundred snow depth measurements and ran with data from the numerical weather forecast model COSMO. Furthermore, snow depth maxima were tracked over the season. While both fSCA estimates correlated significantly, intra-season differences were present. We therefore investigated fSCA maps derived from parameterized mean snow depth using flat field snow depth measurements whilst similarly tracking snow depth maxima. Overall, our results show that complementing satellite snow maps with a fSCA parameterization has great potential for large-scale models.

Quantifying the spatial distribution of snow is crucial to predict and assess its water resource potential and understand land–atmosphere interactions. High-resolution remote sensing of snow depth has been limited to terrestrial and airborne laser scanning and more recently with application of structure from motion (SfM) techniques to airborne (manned and unmanned) imagery. In this study, photography from a small unmanned aerial vehicle (UAV) was used to generate digital surface models (DSMs) and orthomosaics for snow cover at a cultivated agricultural Canadian prairie and a sparsely vegetated Rocky Mountain alpine ridgetop site using SfM. The accuracy and repeatability of this method to quantify snow depth, changes in depth and its spatial variability was assessed for different terrain types over time. Root mean square errors in snow depth estimation from differencing snow-covered and non-snow-covered DSMs were 8.8 cm for a short prairie grain stubble surface, 13.7 cm for a tall prairie grain stubble surface and 8.5 cm for an alpine mountain surface. This technique provided useful information on maximum snow accumulation and snow-covered area depletion at all sites, while temporal changes in snow depth could also be quantified at the alpine site due to the deeper snowpack and consequent higher signal-to-noise ratio. The application of SfM to UAV photographs returns meaningful information in areas with mean snow depth  >  30 cm, but the direct observation of snow depth depletion of shallow snowpacks with this method is not feasible. Accuracy varied with surface characteristics, sunlight and wind speed during the flight, with the most consistent performance found for wind speeds  < 10 m s⁻¹, clear skies, high sun angles and surfaces with negligible vegetation cover.

Avalanche mitigation programs measures snow surface temperature, Tss, for up to three objectives: 1) to infer near surface faceting (NSF) from Tss and the snow temperature 10 cm or 20 cm below the surface; 2) to measure change in the snow surface temperature over time (e.g. days) usually at study plots; 3) to determine the point-in-time surface temperature. We review the surface properties of snow and the energy exchange at the snow surface and identify the low albedo of contact thermometers as problematic for measuring snow surface temperature. Using field studies with contact thermometers, hand-held IR thermometers and an IR camera, we show that a contact thermometer on a shaded part of the snow surface can be up to 6 °C above the surface temperature. While hand-held IR thermometers are promising for measuring Tss, some units are more accurate than others and some units are slow to adjust to the ambient temperature. Since the true snow surface temperature varies widely within hours and the near surface temperature gradient usually reverses twice per day, a point-in-time measurement of the surface temperature – even with an accurate handheld IR thermometer - is less indicative of NSF than observations of the sky cover. We recommend observations or measurement methods for each of the three objectives of avalanche mitigation programs.

The spatio-temporal distribution of snow in a catchment during ablation reflects changes in the total amount of snow water equivalent and is thus a key parameter for the estimation of melt water run-off. This study explores possible rules behind the spatial variability of snow depth during the ablation season in a small Alpine catchment with complex topography. The snow depth observations are based on more than 160 000 terrestrial laser scanner data points with a spatial resolution of 1 m, which were obtained from 11 scanning campaigns of two consecutive ablation seasons.
The analysis suggests that for estimating cumulative snow melt dynamics from the catchment investigated, assessing the initial snow distribution prior to the melt season is more important than addressing spatial differences in the melt behaviour. Snow volume and snow-covered area could be predicted well using a conceptual melt model assuming spatially uniform melt rates. However, accurate results were only obtained if the model was initialized with a pre-melt snow distribution that reflected measured mean and standard deviation. Using stratified melt rates on the other hand did not improve the model results. At least for sites with similar meteorological and topographical conditions, the model approach presented here comprises an efficient way to estimate snow depletion dynamics, especially if persistent snow accumulation pattern between years facilitate the characterization of the initial snow distribution prior to the melt.

The heterogeneous mountain snow cover is challenging the eye and the analytical mind of the observer. The snow distribution affects water resources, natural hazards such as avalanches and ecology. While a lot of recent research has helped to better understand this snow distribution and the processes that cause the heterogeneity, it has not yet been possible to predict snow distribution satisfactorily on the basis of terrain parameters alone. We present a model of the mean snow depth in topographic control units as a function of two terrain parameters: the conventional elevation plus a fractal roughness parameter. For this we used a unique data set of high resolution measurements of snow depth from an airborne laser scanner. The model captures the heterogeneous snow distribution by merely analysing the terrain and the mean precipitation. This unusually simple relationship holds for clusters of the snow depths of small topographical units. By applying fractal analysis, we describe the roughness of the terrain and use this parameter for the prediction of snow deposition. Rougher terrain holds less snow than smoother terrain. This finding is important not only for avalanche warning or eco-hydrological applications, but also for reliably predicting how snow water storage may change in the light of the pronounced climate change already ongoing in mountain regions.

The spatial variability of the snow cover is driven by wind-induced snow transport processes. The main aim of this study is to investigate the link between the scaling behavior of snow depths and the wind-induced processes driving the spatial structure of snow depths. We used very high-resolution atmospheric and snow transport simulations to compare their spatial characteristics with snow depths obtained from airborne and terrestrial laser scans. The directions of the strongest autocorrelations of snow depths and wind velocities were predominantly perpendicular to the modeled local flow direction, independent of whether more homogeneous lee slope loading in one area was considered or the formation of drifts in another area. In the case of homogeneous lee slope loading, the same direction of anisotropy was only found in the modeled preferential deposition of precipitation as for both wind and snow distribution. This deposition process appears to dominate homogeneous lee slope loading. In contrast, saltation-driven processes seem to dominate the formation of drifts and snow dunes. Furthermore, the fractal analysis suggests similar scale breaks for the modeled wind velocity field and measured snow depths defining the upper scale of landscape smoothing through snow.

Terrestrial and airborne laser scanning (TLS and ALS) techniques have only recently developed to the point where they allow wide-area measurements of snow distribution in varying terrain. In this paper we present multiple TLS measurements showing the snow depth development for a series of precipitation events. We observe that the pattern of maximum accumulation is similar for the two years presented here (correlation up to r = 0.97). Storms arriving from the northwest show persistent snow depth distributions and contribute most to the final accumulation pattern. Snow depth patterns of maximum accumulation for the two years are more similar than the distribution created by any two pairs of individual storms. Based on the strong link between accumulation patterns and terrain, we investigated the ability of a model based on terrain and wind direction to predict accumulation patterns. This approach of Winstral et al. (2002), which describes wind exposure and shelter, was able to predict the general accumulation pattern over scales of slopes but failed to match observed variance. Furthermore, a high sensitivity to the local wind direction was demonstrated. We suggest that Winstral et al.'s model could form a useful tool for application from hydrology and avalanche risk assessment to glaciology.

We present an analysis of high resolution laser scanning data of snow depths of three different slopes in the Wannengrat catchment (introduced in part 1) using omnidirectional and directional variograms for three specific terrain features; cross-loaded slopes, lee slopes, and windward slopes. A break in scaling behavior was observed in all subareas, which can be seen as the roughness scale of bare earth terrain which is modified by the snow cover. In the wind-protected lee slope a different scaling behavior was observed, compared to the two wind-exposed areas. The wind-exposed areas have a smaller ordinal intercept γ, a smaller short range fractal dimension D, and a larger scale break distance L than the wind-protected lee slope. Snow depth structure inherits characteristics of dominant NW storms, which results, e. g., in a trend toward larger break distances in the course of the accumulation season. This can be interpreted as a result of surface smoothing at increasing scales. Similar scaling characteristics were obtained for two different years at the end of the accumulation season. Since snow depth structure is altered strongly by NW storms, this inter-annual consistency may strongly depend on their frequency in an accumulation period. With the analysis of directional variograms anisotropies of fractal parameters were detected, which were related to dominant wind directions.

Snow in rock faces plays a key role in the alpine environment for permafrost distribution, snow water storage or runoff in spring. However, a detailed assessment of snow depths in steep rock walls has never been attempted. To understand snow distribution in rock faces a high-resolution terrestrial laser scanner (TLS), including a digital camera, was used to obtain interpolated snow depth (HS) data with a grid resolution of one metre. The mean HS, the snow covered area and their evolution in the rock face were compared to a neighbouring smoother catchment and a flat field station at similar elevation. Further we analyzed the patterns of HS distribution in the rock face after different weather periods and investigated the main factors contributing to those distributions.
In a first step we could show that with TLS reliable information on surface data of a steep rocky surface can be obtained. In comparison to the flatter sites in the vicinity, mean HS in the rock face was lower during the entire winter, but trends of snow depth changes were similar. We observed repeating accumulation and ablation patterns in the rock face, while maximum snow depth loss always occurred at those places with maximum snow depth gain. Further analysis of the main factors contributing to the snow depth distribution in the rock face revealed terrain-wind-interaction processes to be dominant. Processes related to slope angle seem to play a role, but no simple relationship between slope angle and snow depth was found.
Further analyses should involve measurements in rock faces with other characteristics and higher temporal resolutions to be able to distinguish individual processes better. Additionally, the relation of spatial and temporal distribution of snow depth to terrain – wind interactions should be tested.

The spatio-temporal variability of the mountain snow cover determines the avalanche danger, snow water storage, permafrost distribution and the local distribution of fauna and flora. Using a new type of terrestrial laser scanner, which is particularly suited for measurements of snow covered surfaces, snow depth was monitored in a high alpine catchment during an ablation period. From these measurements snow water equivalents and ablation rates were calculated. This allowed us for the first time to obtain a high resolution (2.5 m cell size) picture of spatial variability of the snow cover and its temporal development. A very high variability of the snow cover with snow depths between 0–9 m at the end of the accumulation season was observed. This variability decreased during the ablation phase, while the dominant snow deposition features remained intact. The average daily ablation rate was between 15 mm/d snow water equivalent at the beginning of the ablation period and 30 mm/d at the end. The spatial variation of ablation rates increased during the ablation season and could not be explained in a simple manner by geographical or meteorological parameters, which suggests significant lateral energy fluxes contributing to observed melt. It is qualitatively shown that the effect of the lateral energy transport must increase as the fraction of snow free surfaces increases during the ablation period.