Cleaning up a contaminated site isn’t just about taking action; it’s mainly about correctly interpreting what’s happening beneath the surface. That’s where the decision is made as to whether the intervention will be fast, effective, and cost-efficient. A new project is therefore relying on artificial intelligence to uncover hidden connections from available information and help select the most effective technology for removing contamination.
Remediation experts know from experience that the more information they have about a site before work begins, the greater the chance of choosing the right approach the first time. It was precisely this need that gave rise to the SanAI project. It relies on more precise data analysis and the use of artificial intelligence and neural networks to better understand how contamination is distributed underground, where it is spreading, and where intervention is needed first. The goal is to know as much as possible about the site before cleanup even begins.
SanAI: Practical Experience and Data Analysis
The SanAI project is based on collaboration among a group of scientists from various fields, ranging from advanced pollution removal methods to machine learning, data mining, and mathematical statistics. “Our goal is to combine practical experience with the possibilities offered today by artificial intelligence and advanced data processing,” says Jaroslav Nosek, the project lead for TUL. The main project lead is MEGA a.s., and another commercial partner is Dekonta, a.s.
How the picture of pollution has been pieced together so far
Until now, experts searching for pollution sources have relied mainly on experience, individual boreholes, collected water and rock samples, or field measurements. From this piecemeal information, they then pieced together an idea of what is happening underground. Such an approach is valuable and works in many cases, but it has its limits.
“The subsurface is a complex environment, and something significant—such as the source of contamination itself, an obstacle in the rock, or the direction in which the contamination is spreading—can remain hidden between two boreholes. In practice, this often means that knowledge of the site must be gradually refined, supplemented with additional boreholes, and the right solution sought step by step,” explains Jaroslav Nosek, noting that this is precisely what prolongs and increases the cost of the entire process.
Faster and Smarter Map Creation
As part of SanAI, researchers began automating map creation in the Surfer program. For geologists, this represents a significant time savings, as manually creating contamination maps is slow and involves many repetitive steps. Thanks to new scripts, they can process more data faster and more clearly. They have already verified that this approach works on real data from a site where contamination remediation is currently underway.
But this is just the beginning. Automated mapping is also crucial for the next phase of the project. If artificial intelligence is to help us better understand what’s happening underground, we must be able to clearly visualize its results. Without that, the data would remain just a series of numbers. But thanks to the maps, it takes on a comprehensible and practically useful form.
From Individual Measurements to the Big Picture
One of the most important parts of the project is the development of tools for automatically processing data from detailed site surveys. These tools will help identify erroneous values, continuously evaluate results, and show where we already have enough information and where, conversely, there are still gaps. It’s similar to putting together a puzzle when most of the pieces aren’t available yet. Even with just a few pieces, you still try to understand what the whole picture looks like and estimate where it would be best to collect additional samples. This is exactly how a cleverly designed network of boreholes can be created right from the start.
“It’s a challenge that requires patience and systematic experimentation, but at the same time opens the door to original scientific contributions,” emphasizes Šárka Horská, an expert in machine learning and data mining techniques.
In the future, artificial intelligence should be able to estimate the three-dimensional distribution of underground pollution. That is, not only where the problem is, but also how it spreads and where action needs to be taken as soon as possible. The result should be a kind of three-dimensional pollution cloud that shows where its concentration is highest and where, conversely, it is weakening. This can significantly improve decision-making and reduce the tedious search for the right approach through trial and error.
Goal: better decisions, fewer delays
Although the project is still in its early stages, if this approach proves successful, it could bring about a fundamental change in the field. Better knowledge of the site right from the start means more precise intervention, faster removal of pollution sources, and more efficient use of time and money. And in the world of environmental remediation, that is precisely the difference that can make all the difference.
