Ecohydraulic modelling of eutrophication for reservoir management

Nahm-Chung had recognised that existing 2D and 3D physically-based models of eutrophication in reservoirs and lakes are limited by a number of physical, chemical and biological processes that are not yet resolved by the physically based models in current use.

This is mainly due to the importance of the fine temporal or spatial scales relative to the numerical grid and time steps, the neglect of some fundamental processes in the model equations, and difficulties in capturing the effect of these processes numerically.

As a consequence, the potential impact of external nutrient loads coming from the contributing catchments on the aquatic ecosystems of the reservoir cannot be determined adequately on the basis of physicochemical features alone. Together with the unresolved effects of hydrodynamic models, a wide range of environmental problems has lead to the emergence of data-driven modelling to make sense of multivariable data.

In recent years, there has been a growing tendency to use data-driven modelling based on machine learning to complement or even replace physically-based modelling, especially for forecasting. Consequently, Nahm-Chung coupled specific data driven models with physically based models to form a decision support framework for eutrophication.

He was fortunate to have a considerable amount of high quality data concerning the hydraulics and biochemical processes in the reservoir. These he analysed in detail, especially as the data was to form a vital foundation of his use, among others, of a recent data driven modelling technique called Clustering Partial Least Squares Regression (C-PLSR). He used C-PLSR to estimate successfully the dynamics of Chlorophyll-a based on measured data, but complemented by results for the physicochemical parameters provided by MIKE 31 or CE-QUAL-W2.