IFCA develops artificial intelligence model to estimate water chlorophyll concentration

Monitor water quality Contributing to the protection of this life-essential resource is a priority. recent, Latest United Nations report On global progress on the 2030 Agenda for Sustainable Development, Recognizing the lack of data related to water quality, Especially in less developed countries, hypothesis The health of more than 3 billion people is at riskTherefore, the application of detection measures becomes increasingly necessary., before severe degradation of the aquatic environment in these areas occurs.

To this end, a team of three researchers from IFCA’s Computational and Electronic Sciences Group, Judith Sainz-Pardo, Maria Castrillo and Alvaro Lopez, Developed a deep learning model (deep learning), is able to estimate the concentration of chlorophyll in water using techniques measured by powerful sensors that are cheaper and more commonly used than existing techniques.The results were published in the magazine “Water Research”, which ranks first in the field of aquatic resources Journal Citation Report

(JCR).

“When microalgae grow in water masses, chlorophyll appears. With this type of sensor, we can detect early the occurrence of large numbers of these organisms, as in recent cases of gastroenteritis,” he explains. castrillo. “It could also be applied to measure other types of pollutants, such as nutrients and even emerging pollutants, which are becoming increasingly important on the water governance agenda,” the researchers said.

Maria Castrillo and Judith Sainz-Pardo at Las Llamas Park in Santander.

at the moment, Measuring the presence of these particles in water masses using optical sensorsHowever, its use is very limited due to several disadvantages such as its cost or maintenance requirements, and it does not provide real-time information to guarantee adequate management of the risks involved in these pollution events.

The study used two tributaries of the River Thames as models, and obtained data on physicochemical variables of its waters, such as temperature, pH, and electrical conductivity, as well as on meteorological variables, such as solar irradiance (the total amount of solar energy falling on the surface during a given time), ambient temperature, and wind speed . This set of data can be used to estimate the chlorophyll concentration of these water masses.

Microscopic observation of chlorophyll.Pixel

Globally effective artificial intelligence model

The research team trained a deep learning), with three learning methods: individual, centralized and joint.What these three methods seek is Check which of these is most suitable when working at different study locationssuch are the two tributaries of the Thames.

researcher Judith Sainz-Pardo, Explaining that this type of architecture is “the most innovative part of this research because it has never been applied in this field before,” it’s called federated learning. “What we have achieved is Train global models without pooling data from two rivers“What we do is get the data from each river in a decentralized way, that is distributed, we train the neural network and then the server is responsible for adding that model to build a global model,” said IFCA researcher. “When faced with data that the model has not seen before, it allows us to generalize better than other classic cases,” he added. Sainz Pardo.

In addition to the accuracy of water quality studies, this research also achieved the following results: Reduce the costs involved in processing large amounts of data, Sometimes further research is hindered by issues of privacy, security, or technical limitations (such as lack of data storage or network connectivity). “There is therefore no need to share the raw data obtained in each river, nor to ‘travel’ from the equipment that measured them,” which reduces,

explain Sainz Pardo“not only cost of storage memory on a single server, but he Computing costsThat’s “the amount of resources required to train the model and make predictions,” the researchers said.This work represents a Knowledge transfer example For companies or other public research organizations that want to implement it, because with new training of the model, new water bodies can be studied.