New AI Model Predicts Water Quality Based on the Weather

Water quality sampling is an arduous task – exponentially so for remote areas, where difficulties in physical accessibility can lead to years- or decades-long gaps in data. This, in turn, leads to inefficiencies, delays and mistakes in remedying problems with water quality. Now, a team led by Penn State researchers has developed a new, AI-powered model to predict water quality in such remote regions using weather data.

The researchers figured that hydrometeorological data – essentially, water-related weather data – might prove a useful indicator of water quality, as it was recorded with much greater frequency than direct sampling data. “There is a lot of hydrometeorological data available, and we wanted to see if there was enough correlation, even indirectly, to make a prediction and help fill in the river water chemistry data gaps,” said Wei Zhi, first author of the paper and a postdoctoral researcher in Penn State’s Department of Civil and Environmental Engineering, in an interview with Penn State’s Tim Schley.

So the researchers fed some of this data into a long short-term memory (LSTM) network-based AI model, combining it with data from hundreds of watersheds that had been collected from 1980 to 2014. The goal: forecasting dissolved oxygen (or DO), which serves as a key indicator of the ability of those watersheds to host aquatic life. 

“People usually think about DO as being driven by stream biological and geochemical processes, like fish breathing in the water or aquatic plants making DO on sunny days,” said Li Li, a professor of civil and environmental engineering at Penn State. “But weather can also be a major driver. Hydrometeorological conditions, including temperature and sunlight, are influencing the life in the water, and this in turn influences the concentration levels of DO.”

The model performed well, the researchers found, highlighting how it had intuited key relationships, such as how dissolved oxygen is lower in warmer water and higher elevations. “It is a really strong tool,” Zhi said. “It surprised us to see how well the model learned DO dynamics across many different watershed conditions on a continental scale.” Still, it had some shortcomings: the model performed less well in areas with uneven DO levels or unstable water flows.

“If we can collect more samples that capture the high peaks and low troughs of DO levels, we will be able to reflect that in the training process and improve performance in the future,” Zhi said.

To learn more, read the coverage from Penn State’s Tim Schley here.