AI and Fluorescence Spectroscopy in Water Treatment

Drinking water treatment plants face the ongoing challenge of removing harmful contaminants from raw river water before it reaches the public. Two major threats are cyanotoxins, which are produced by harmful algal blooms, and organic matter that reacts with chlorine to form hazardous disinfection byproducts (DBPs). Traditional water testing methods are often slow and require days of laboratory analysis, leaving plants vulnerable to sudden changes in water quality. Today, treatment facilities are combining excitation-emission matrix (EEM) fluorescence spectroscopy with artificial intelligence (AI) to detect these specific threats in real time at raw river intakes. This article explains how the optical technology works, how AI decodes the resulting complex data, and how this integration protects public drinking water.

The Threat at the River Intake

Raw river water contains dissolved organic matter (DOM) from decaying plants, runoff, and aquatic life. While most DOM is harmless on its own, it acts as a precursor to disinfection byproducts. When treatment plants add disinfectants like chlorine or chloramine to kill pathogens, the chemicals react with this organic matter to create toxic compounds such as trihalomethanes (THMs) and haloacetic acids (HAAs).

Simultaneously, warmer temperatures and nutrient runoff frequently trigger harmful algal blooms in rivers. These blooms release cyanotoxins, such as microcystin and cylindrospermopsin, which can cause severe liver and neurological damage in humans. Detecting both DBP precursors and cyanotoxins at the exact moment water enters the plant is critical for adjusting chemical dosing and filtration processes.

How EEM Fluorescence Spectroscopy Works

Excitation-emission matrix (EEM) fluorescence spectroscopy is an optical technique used to rapidly analyze the organic composition of water. The process does not require chemical reagents and takes only a few minutes.

  • Excitation: The spectrometer shines ultraviolet and visible light into a raw water sample across a wide range of wavelengths.
  • Absorption: Specific organic molecules in the water absorb this light energy, causing their electrons to jump to a higher energy state.
  • Emission: As the electrons fall back to their normal state, the molecules emit light (fluoresce) at longer wavelengths.
  • The Matrix: The instrument measures the intensity of the emitted light across all wavelengths and plots the data Against the excitation wavelengths. This generates a three-dimensional map—resembling a colorful topographical map—that serves as a unique optical fingerprint of the water sample.

While cyanotoxins themselves do not always fluoresce strongly, the biological compounds associated with cyanobacteria—such as specific proteins, pigments like phycocyanin, and intracellular organic matter—produce strong, distinct signals on the EEM map. Similarly, the humic and fulvic acids that serve as the primary precursors for DBPs fluoresce in specific, predictable regions of the matrix.

The Role of Artificial Intelligence

A single EEM scan generates thousands of data points. In raw river water, the optical signals of harmless organic matter, DBP precursors, and cyanobacterial compounds often overlap. This makes it nearly impossible for human operators or basic software to quickly interpret the data. This is where artificial intelligence becomes essential.

Treatment plants utilize machine learning algorithms, particularly Convolutional Neural Networks (CNNs) and Support Vector Machines (SVMs), to analyze the 3D fluorescence maps instantly.

  1. Deconvolution of Overlapping Signals: AI models are trained to separate overlapping light signatures. Previously, a mathematical method called Parallel Factor Analysis (PARAFAC) was used for this, which required heavy manual calibration. Deep learning models now automate this step, instantly isolating the exact fluorescence peaks associated with humic-like substances (DBP precursors) and protein-like substances (algal activity).
  2. Pattern Recognition for Cyanotoxins: Because many cyanotoxins are non-fluorescent, AI uses proxy detection. Researchers train AI models using thousands of historical water samples where both the EEM fingerprint and the exact toxin levels (measured via traditional liquid chromatography) are known. The AI learns to recognize subtle shifts in the fluorescence of algal proteins and pigments that correlate directly with the release of toxins like microcystin.
  3. Predicting DBP Formation Potential: Rather than just measuring current organic matter, AI models combine EEM data with basic water parameters like pH, temperature, and bromide levels. The algorithm then predicts the exact concentration of THMs and HAAs that will form if the water is chlorinated under current plant conditions.

Real-Time Operational Adjustments

By placing EEM sensors paired with AI directly at the raw river intake, water treatment plants transition from reactive testing to proactive management. When the AI detects a sudden spike in cyanotoxin proxies or DBP precursors, it triggers immediate operational changes:

  • Activated Carbon Dosing: Operators can instantly increase the dose of powdered activated carbon (PAC) to absorb cyanotoxins before they pass through the plant.
  • Coagulation Optimization: The plant can adjust coagulant chemicals (like alum or ferric chloride) to bind with and remove the specific type of DBP precursors identified by the EEM scan prior to the disinfection stage.
  • Intake Management: If a river intake has multiple draw depths, the plant can switch to drawing water from a deeper level where algal activity and organic precursors are lower.

By merging the rapid optical scanning of EEM spectroscopy with the pattern-recognition power of artificial intelligence, drinking water plants can effectively neutralize invisible chemical threats before the water ever reaches the treatment basin.