Alpine photovoltaic (PV) installations exploit the high winter solar irradiance above the fog and low-stratus layer to help close Switzerland's structural winter electricity deficit. Yet deploying PV at utility scale in complex Alpine terrain raises open questions: how solar radiation distributes where topographic shading, sky-view limitations, and snow reflectance interact at fine scales; what performance is realistically achievable and what drives snow-induced losses; and how wind-driven snow redistributes around elevated structures. This thesis addresses them through field monitoring, satellite-based radiation modeling, deep learning, and computational fluid dynamics.
Part I establishes performance benchmarks from four years of monitoring at Muttsee, Switzerland's first utility-scale Alpine PV plant. Excluding the exceptionally snowy 2023/24 winter, the plant reaches a mean annual yield of 1303 kWh/kWp with about 50% of generation in the winter half-year, confirming the Alpine advantage. Snow losses are disaggregated by mechanism: under normal conditions, shedding from panel surfaces dominates and total losses stay moderate at 2-3% of production. In the extreme 2023/24 winter, however, snow piling up beneath the panels buried the lower edges, causing 13.3% generation losses and structural damage, showing that ground interference becomes critical once clearance height is exceeded.
Part II presents HeMu, a deep-learning emulator for satellite-based surface solar radiation retrieval that matches state-of-the-art accuracy with a large computational speedup over legacy radiative-transfer methods. A key insight is that the model's temporal-context mechanism implicitly recovers background surface reflectance, mirroring the buffered-statistics approach of classical Heliosat algorithms and showing how physics-based modeling can guide neural-network design. Building on this retrieval, the CH-POA300 dataset provides hourly plane-of-array irradiance at 300 m resolution across Switzerland for 2004-2020, enabling Alpine site selection, rooftop PV profiling, and yield forecasting.
Part III investigates snow accumulation around PV structures through two complementary approaches. Physics-based CFD simulations with the snowBedFoam solver model wind-driven snow transport around the HELIOPLANT structure, revealing characteristic erosion and deposition patterns, while design-sensitivity analyses show how structure height, orientation to the prevailing wind, and inter-unit spacing control burial risk. These are complemented by an empirical interference-height model (R2 = 0.89) from drone-based snow surveys at Muttsee, relating local topography, snow climatology, and plant geometry to maximum accumulation beneath modules, a practical design tool.
Together, these contributions advance understanding of how snow shapes the radiation balance and PV yield in mountains, and cover the full Alpine PV planning chain: nationwide terrain-aware irradiance data to rank candi
