Rethinking Aquatic Ecosystem Models скачать в хорошем качестве

Rethinking Aquatic Ecosystem Models

Dr. Robert Ladwig - Aarhus University Limnology is a theory-driven science, built on conceptual frameworks and mechanistic models. These often take the form of process-based mathematical models that simulate system behavior using known physical, biogeochemical and ecological principles. These process-based models are used for mechanism understanding and to project past, present and future dynamics of aquatic ecosystems. The traditional focus is on applying vertical one-dimensional models to simulate ecosystem dynamics. The reasons for the one-dimensional approach are manifold but are mostly based on the assumption that vertical mixing is more pronounced than horizontal mixing, which is mirrored by the common monitoring of lakes using a buoy at their deepest location. Over the years, this has resulted in the development of multiple aquatic ecosystem models (AEMs), which are conceptually similar but differ regarding their specific algorithms and equations. But, projecting ecological dynamics with a good agreement to field data is often severely lacking as physical, biogeochemical and ecological processes can seldom be described with exact equations. It seems that the theory-driven approach is limiting our abilities to simulate their dynamics due to the stochastic nature of turbulence, the complexity of ecological systems, and the apparent fact that ecology does not follow the “unreasonable effectiveness of mathematics”. This talk aims to outline the next steps on improving our applications of AEMs and make them sounder and more robust with the eventual goal to enhance our ecological processes understanding. The basic axiom is that aquatic ecosystems are vastly complex dynamic systems with manifold causal feedback loops. Based on this, there is a need to clearly state the underlying model assumptions, openly describe their uncertainties and use them in a sound way to replicate possible ecological pathways. One way to do this is (1) the use of ensemble modeling. Running multiple AEMs together can highlight the uncertainties caused by potential equifinality, multiple models giving the same result, and model multiplicity, conceptually identical models giving alternative results. Further, modelers need to acknowledge sparsity, hence a lack of data and information in time as well as in space. Current lake monitoring research in Denmark is addressing this issue by (2) deploying multiple monitoring buoys and investing more into low-cost high-frequency sampling. These data are prime candidates for applying machine learning models. Finally, an emerging paradigm has risen over the last years that directly aims to handle complexity, uncertainty and sparsity in an integrated way: (3) Ecological Knowledge-Guided Machine Learning. This method directly couples ecological knowledge with deep learning models to improve projections and to upscale knowledge from data-rich to data-sparse field sites. Overall, by acknowledging complexity, uncertainty and sparsity, we can rethink our current model applications and develop innovative methods for the future.