DATALAKES

Heterogeneous data platform for operational modelling and forecasting of Swiss lakes

Abstract

The objective of this project is to advance the forecasting capabilities of the data-driven hydrological and ecological lake modeling algorithms using methodologies inspired by data science and accelerated by high performance computing. We aim to develop a parallel framework interfacing high resolution 3D numerical solvers for the underlying lake dynamics with modern numerical Markov Chain Monte Carlo sampling methods for Bayesian inference, with particular interest in investigating particle filtering and multi-level variance reduction methodologies. The resulting framework aims at accurate data assimilation and uncertainty quantification in both model parameters and the associated forecasts. DATALAKES project is a collaboration with the Swiss Data Science Center (SDSC), EPF Lausanne and ETH Zurich, aiming at a sensor-to-frontend data platform providing and analyzing the dynamics of lake ecosystems at high spatial and temporal resolutions. Current version of the existing framework can be found at meteolakes.ch.

Starting Date / Status

November 2018

COMPLETED

PI / Partners

Scientific Computing group (EAWAG)

Aquatic Physics (EAWAG)

Aquatic Systems Laboratory (EPFL)

Read the article about this project on our blog:

Heterogeneous Data Platform for Operational Modeling and Forecasting of Swiss Lakes

Description

Problem:

  • Increasing pressure on lakes needs scientific support

  • 3D numerical simulations of lakes require input data – uncertainty quantification in parameters & forecast

  • New L’EXPLORE platform in Lake Geneva – increasing availability of high resolution data

Solution:

  • Sensor-to-frontend open data platform

  • Physics-driven hydrodynamic models

  • Data-driven modeling of input data processes

  • Parallel Bayesian inference – MCMC with ABC or PF

  • Multi-level speedup – hierarchical numerical models

  • Powered by Renku, the SDSC-developed platform for transparency and reproducibility in science

  • A neural network with uncertainty quantification properties in order to more accurately aggregate data from satellite imagery into the Bayesian inference

Impact:

  • Real time monitoring & future forecast of lakes

  • Platform for large-scale interdisciplinary collaborations

  • Research in hydrological / ecological lake modeling

  • Scientifically grounded water resources management

Publications

 

  • D. Bouffard, J Runnalls, T Baracchini, E Bouillet, H E Chmiel, T Doda, B Fernández Castro, F Georgatos, S Lavanchy, C Minaudo, F Ozdemir, D Odermatt, M-E Perga, P Perolo, S Piccolroaz, M Plüss, L Råman Vinnå, M Schmid, A Safin, J Šukys, V Tran-Khac, H N. Ulloa, C L. Ramón, A Wüest. Datalakes, a data platform for Swiss lakes. In prep for Earth System Data Science

  • Safin, J. Ŝukys, D. Bouffard, C. L. Ramon, F. Ozdemir, J. Runnalls, F. Georgatos, C. Minaudo. A Bayesian data assimilation framework for lake 3D hydrodynamic models with a physics-preserving particle filtering method. In prep for Geoscientific Model Development.