From observations to 3D forecasts: Data assimilation for high resolution lakes monitoring

Lake mesoscale processes have no secrets for artists, they have some for scientists and policy-makers. In the introduction to this dissertation, we saw that poets and painters depict the lake with rich variability. This vision conflicts with conventional approaches inferring the lake's bio-physical status from a few in-situ measurements with limited spatio-temporal coverage. Yet it reflects upon true physical processes, capable of disrupting the misleading lentic nature of lakes, the essential ecosystem services they provide, and even citizens' safety. Recent overarching policies now aim at securing those services, using novel approaches like numerical simulations and remote sensing observations. Harnessing the potentiality of in-situ, remote sensing data and model simulations, this thesis developed an end-to-end framework delivering reliable synoptic lake information, at high spatial and temporal resolution. In lakes, three-dimensional hydrodynamic models are the only information source capable of resolving transport and mixing, and forecasting their dynamics. However, they still rely on large observational datasets for their complex parameterizations and for constraining their uncertainties. This thesis addressed such challenges by (i) implementing an automated model calibration framework alleviating the need for expert knowledge, and by (ii) developing a data assimilative scheme capable of reducing and quantifying model uncertainties by incorporating satellite and in-situ data. Both yielded remarkable results: the former returned parametric values diminishing models Root Mean Square Error by up to 47 %, while the latter cut it further down by 54 %. Furthermore, the data assimilation enhanced the spatial coherence and magnitude of imperfectly resolved physical processes, and provided system uncertainties. Finally, this thesis delivered a practical outcome of its findings by developing an online pre-operational three-dimensional lake monitoring and forecasting system: meteolakes.ch. For two years, Meteolakes has been disseminating spatially explicit real-time lake information and data products to more than hundred thousand end-users. This pioneering platform has been featured in numerous media (newspapers, radio, television), public events, museum exhibitions, and benefited scientific, lake professionals, and public communities. A pinnacle of this research has been its early-warning and forecasting capabilities, which anticipated numerous mesoscale physical phenomena in Lake Geneva, such as upwellings, gyres, and strong currents. Two of those events, which impacted public and commercial activities, are illustrated in this dissertation. From observations and models to societal benefit, we created here a long value chain for water management. At the crossroads of scientific, computational and observational advances, this research paves the route for understanding lakes' delicate imbalance. By producing data society can use, it opens novel frontier