Predicting how aquatic ecosystems will respond to increasing pressures — from changing climate to increasing demands for ecosystem services — is extremely challenging, due to complexities of aquatic ecosystems themselves, as well as their interactions with adjacent ecosystems and exogenous forcing. Process-based aquatic ecosystem models are powerful tools to simulate long-term ecosystem processes due to their low computational and input data need. Here, scientists and managers utilize process-based models to simulate how future climatic conditions would affect water quality or to forecast ecological processes in near-time. Especially by using standardized workflows, as well as ensemble modeling and intercomparison studies, we can address model uncertainty and provide stakeholders with critical information.
As process-based models have shortcomings regarding their performance, a recent stride in modeling is to combine mechanistic knowledge of ecosystem processes with data from long-term monitoring and high-frequency measurements in a data-driven deep learning model. The new paradigm of Knowledge-Guided Machine Learning (KGML) aims to improve ecosystem projections while still providing outputs that are ecologically and physically valid.
​
As the Computational Limnology team at Aarhus University, we address limnological questions using a plethora of methods, ranging from simple toy models, to data-driven approaches, and up to complex numerical models. We are also conducting field measurements and long-term monitoring to better capture ecosystems processes on multiple time scales. To complement our research, we work closely across sections, i.e., with Prof. Davidson to explore lake metabolism dynamics and carbon cycling.
Main research themes
Freshwater Metabolism
Aquatic metabolism is a key limnological concept to infer information about carbon, dissolved oxygen, nutrient and energy cycling in aquatic ecosystems. As metabolism distills water quality information and carbon cycling into a complex metric with a high order of information, it conceptually allows the inference of how freshwater ecosystems respond to environmental change. Our research aims to quantify how temporal and spatial feedbacks between biotic (phytoplankton succession, zooplankton grazing) and abiotic processes (stratification, ice formation, nutrient runoffs) will be affected by climate change and land use changes, and how this will impact aquatic ecosystem functions.
Take a look at: Rohwer, Ladwig et al. (2023), Ladwig et al. (2022).
Physical Limnology
Understanding mixing in lakes is critical for accurately simulating the distribution of heat, nutrients, and gases, which directly influence ecological processes such as primary production and biogeochemical cycling. Process-based models to simulate horizontal and vertical transport, water quality and ecosystem dynamics are powerful tools to explore feedbacks within aquatic ecosystems and to explore climate change scenarios. To ease model application and development, we focus on developing standardized modeling workflows, and conducting inter-model comparison studies to quantify performance and uncertainty. Further, we use state-of-the-art monitoring by collecting high-frequency data from instrumented buoys to improve our process descriptions of vertical mixing and ecological dynamics.
Take a look at: Moore, Mesman, Ladwig, Feldbauer et al. (2021), Ladwig et al. (2021)
Knowledge-Guided Machine Learning
Merging process-based modeling with deep learning shows promising results on how to improve lake modeling by training deep learning models on synthetic and observed data. Still such hybrid models have not yet widely been developed and tested for ecological research question. Developing novel hybrid models could revolutionize our applications of aquatic ecosystem models by providing improved projections that are still ecologically and physically valid.
Take a look at: Ladwig et al. (2023).
"The purpose of computing is insight, not numbers."
- Richard W. Hamming