Current challenges in the field of air and water pollution monitoring require the capability to detect simultaneously a large variety of chemical compounds at very low concentration using low-cost, compact sensor nodes. While carbon nanotube-based (CNT) sensor arrays have long been proposed as a solution to this challenge, their sensing performances usually suffer from the large number of interferents in real-life conditions. Here we discuss an uncertainty-based calibration and prediction framework which allows to recover multi-parameter sensing even in a highly perturbed environment. We study a 10×2 CNT-sensor array for pH and active chlorine monitoring in drink water. While in deionized water pH and active chlorine are easily monitored, in tap water only the active chlorine level can be recovered by standard calibration. By contrast, using our Bayesian approach, both active chlorine and pH are recovered with mean absolute error comparable with reference sensors.