A hybrid empirical and parametric approach to forecasting dissolved oxygen in Lake Geneva to address long-term changes in lake biogeochemistry under re-oligotrophication and climate change

Abstract:

The last century saw a pronounced period of severely dropping lake water quality, characterized by overabundance of algae and dropping dissolved oxygen (DO), due to industrialization of agriculture and urban expansion. In some places, efficient management against eutrophication (e.g. banning of phosphorous in laundry detergent) led to strong reductions in lake phosphorous loading (total phosphorous, TP). The expectation has been that reducing the cause of the environmental degradation will let lakes return to a “pristine” or at least healthier state. In many cases, even with extensive remediation efforts, DO and chlorophyll (CHL) have not returned to pre-20^th century levels. One problem is there is another driver destabilizing systems: climate change. However, the effects of climate are not neatly separable from the effects of eutrophication and therein lies a second, deeper problem. The relationships between DO, CHL, and TP occur in the complex interactive nexus mediated by lake physics, chemistry, and ecology. For example, the effect of re-oligotrophication on DO can be dampened by an increase in the C:P ratio of the phytoplankton community to maintain biomass production, and at the same time, increasing air temperature promotes less edible harmful cyanobacteria in the primary producer pool that result in vastly different export fluxes of carbon to depth.

Resolving all these potential pathways and their interdependencies is a parametric headache. Here, we highlight how emerging data science tools offer an orthogonal approach, taking Lake Geneva as an iconic example. We use parameter free time-series analysis to probe the fundamental nature of the ecological processes and interactions. This then enables us to build a next-generation hybrid approach that combines physics-driven (equation-based) and empirically-driven (equation-free) components to diagnose and predict DO response to future climate and management scenarios. At its root, the approach uses data to leverage a more complete systems perspective without throwing away established, first-principles understanding, and so should be useful broadly as a tool for 21^st century management.