Why dedicated data-mechanism AI models are required for proactively controlling algal-induced odorants at trace levels: A critical review from source to tap.

Algal odorants are threatening water supplies, but traditional odorant control is passive and in a single perspective, with shortcomings like delayed identification, lack of dedicated models, and blind treatment. Proactive source-to-tap control is thus proposed, where identification and treatment by artificial intelligence (AI) are the core. Yet, how to apply unique AI technologies to trace-level odorants and how to integrate identification and treatment are challenging. Rapid identification acts as a radar. Three promising methods with speed, accuracy, and coverage upgrades are compared, i.e., portable gas chromatography-mass spectrometry (PGCMS), electronic-noses, and biomarker fluorescence. For rapid identification, AI allows odorant interpretation from distinct raw signals (from specific to indirect) with dedicated algorithms. Targeted treatment acts as a navigation, with dedicated data-mechanism models to decide optimal windows (source water), dosages (water treatment plants, WTPs), and locations (drinking water distribution system, DWDS). Specifically, mechanism-constrained temporal models are for source water, extracting-mapping models for WTPs, and kinetic-hydraulic models for DWDS. AI-driven synergy closes the loop at technology levels: from identification to treatment, AI provides inputs and correction, upgrading treatment to adaptive treatment; conversely, AI directs where, when, and what to monitor, updating identification to proactive perception. Developing odorant control in interactive platforms enables its practical applications at scenario levels, though data scarcity, stage-isolated models, and limited online sensing remain practical challenges. Together, dedicated AI models are needed for odorant control, promisingly facilitating proactive odorant management and more scientific decision-making in the identification-treatment loop.