#ESR2

Machine learning methods integrating climate and water monitoring data to support modelling future water quality in lakes and reservoirs over decades

Host institution

Ruhr-Universitaet Bochum, Germany.

Project description

This project will integrate new machine learning tools to climate, water monitoring, and recent modelling results on both river and lake / reservoir water quality, vastly improving the confidence of modelling results and enabling to develop robust scenarios over decades.

Future water quality in lakes and reservoirs faces multiple challenges: 1) developing countries are expecting increasing nutrient loads while lakes in developed countries that were supposedly recovering are seeing re-eutrophication (e.g., Lake Erie), 2) large scale models currently do not take into account the effects of extreme weather, seasonal changes or simplify across pluri-annual time scales, 3) future reservoir construction has the potential to impact even more severely downstream river loads.

Machine Learning with emerging tools such as deep learning with convolutional neural networks are geared towards image recognition tasks and can extract features. Object recognition and automatic classification of water quality indicators result in vastly faster development of lake and reservoir water quality products, and pattern recognition can help to link eutrophication with weather events. The combination of techniques will greatly improve the robustness of modelling approaches, as the underlying database will be enhanced by several magnitudes. This in turn will reduce uncertainties when integrating and developing new scenarios.

Secondments

Aarhus University, Denmark. 2 months.
Helmholtz-Centre for Environmental Research, Germany. 3 months.
Wageningen University, The Netherlands. 3 months.

ESR

Ammanuel Bekele