Aquatic ecosystems are complex environments influenced by many variables, including weather, the biological activity of the organisms that live there, and anthropogenic nutrient pollution. The impact of these variables on aquatic ecosystems also depends on the characteristics of the water body, such as water temperature and depth. When these interconnected processes become out of balance, the consequences can be devastating.
To predict these outcomes, a group of researchers from the University of Connecticut developed a versatile computer modeling approach using machine learning to enhance existing efforts to monitor and predict lake water quality, recently published in the journal Environmental Modeling & Software.
Associate Professor Marina Astitha, from the School of Civil and Environmental Engineering and head of the Atmospheric and Air Quality Modelling group, explains that the research has been five years in the making and was conducted as part of former student Christina Feng Chang’s (Class of ’22) doctoral thesis, in collaboration with Professor Penny Vlahos, from the School of Marine Sciences and head of the Environmental Chemistry and Geochemistry research group.
Aquatic environments are susceptible to eutrophication, a process caused by excess nutrients, most notably fertilizer runoff from agricultural activities, which finds its way into aquatic ecosystems and causes algal blooms. The growth and eventual decomposition of these plant-like materials consumes most or all of the available oxygen, negatively impacting other organisms in the environment. Areas with low or no oxygen, known as “dead zones,” can cause fish kills, water quality problems, and other harmful environmental and economic impacts. Astisa explains that these eutrophication phenomena are expected to intensify with climate change, making such models even more important for monitoring and forecasting purposes.
The researchers focused their study on Lake Erie's central basin, which has suffered seasonal algal blooms and eutrophication events for decades. The lake's proximity to large agricultural areas where fertilizer is used, as well as large urban centers where air pollution is a concern, presents a unique set of challenges that the team sought to study.
Because millions of people depend on Lake Erie water, modeling has helped monitor water quality and will continue to do so, Astisa said.
“Currently, the forecast models are doing daily forecasts, which is very important especially for people living in these areas because these are densely populated areas. Water is not just for recreational purposes. People use water in their daily lives.”
But Astisa says no single model can account for all the variables that affect water quality. To address this, they began building a machine learning model that integrates data from different sources and trains machine learning algorithms based on observations in the lake.
Astisa said the first paper using this method focused on machine learning modeling of algal biomass and chlorophyll a, an indicator of eutrophication, while the second used the same methodology to look at nutrient pollution from rivers and streams. This latest paper looks at physical and biological processes limited to a physics-based model to understand the dynamic processes involved in eutrophication events.
Astisa says models will have to start from scratch for each process they study, but will need to assess the different physical, biological, weather-related and anthropogenic processes that influence eutrophication.
Chang explains that the eutrophication process begins in the spring when fertilizer applications on agricultural fields and subsequent rains wash nutrients into the lake. During the summer, Lake Erie's water separates into three layers: the epipelagic, a warm layer close to the surface; the mesopelagic, a middle layer that experiences the most rapid temperature changes; and the deep, deeper, cooler layer. The mesopelagic layer contains a thermocline, which causes rapid changes in temperature. During the stratified summer, there is little mixing between the epipelagic and mesopelagic layers, so the deepest water is deficient in oxygen all summer long.
The central basin of the lake is prone to the most severe hypoxia, so to study this phenomenon and understand its causes, Astisa explains, they designed a model to predict dissolved oxygen (DO), an indicator of hypoxia in the water, and apparent oxygen availability (AOU), an indicator of biological activity in the aquatic ecosystem. They trained the model using 15 years of data collected from 2002 to 2017.
The results were positive, Astitha says, and the model accurately predicted the observed DO and AOU conditions. The model also identified temperature stratification, or distinct temperature layers, in the water column as the most influential variable driving eutrophication in the study area.
“This was a good proof of concept, since we have very few data points on the lake,” Astisa says, “Ideally, any model would need to cover a larger area of the lake, but that's not happening. It's not feasible with the point observations we have. Nevertheless, the model performed very well.”
As the climate continues to change, models like this will become increasingly important for monitoring water quality and supporting decision-making. Astisa expects conditions like rising temperatures will increase stratification, but he says that extreme precipitation due to climate change could further increase the amount of nutrients entering the lake.
“Hypoxia occurs because in this natural system, nitrogen and phosphorus are naturally present, but when you fertilize hundreds of acres, some of that fertilizer leaches into the water. This is determined by lake mixing and stratification, and weather conditions affect these. Conceptually, we think climate change will make the situation worse, and we can now use our model within climate simulation conditions to look at hypothetical future scenarios.”
Astisa says future research will include applying this methodology to other freshwater and marine ecosystems, as well as conducting more thorough analysis using a range of climate change projection data to explore how climate change scenarios might affect water quality in those systems.
“From my perspective, we wanted to build a tool that would complement the models that were already doing this important prediction and monitoring. In the age of machine learning and artificial intelligence, we were trying to bring in that element and see how useful it could be, and that's what motivated me to start and continue this work.”
