Author(s): Sait Mutlu Karahan; Wouter Vandenbruwaene; Jan Verwaeren

Linked Author(s):

Keywords: Electrical conductivity forecasting; Long Short-Term Memory; Uncertainty quantification

Abstract: Salinisation is a growing concern in river systems, requiring continuous monitoring to support effective water quality management. While data-driven models have become increasingly prevalent in forecasting water quality parameters, uncertainty estimation in such models remains limited in the literature. This study addresses this gap by integrating uncertainty quantification into Long Short-Term Memory (LSTM) models for predicting EC, which can be used as a proxy for salinisation. Sensor locations, representing upstream and downstream regions, were selected from the river’s sensor network. For each region, one-step-ahead and multi-step-ahead forecasting strategies were implemented. A comparative analysis of uncertainty across spatial (upstream vs. downstream) and temporal (one-step vs. multi-step) variability was conducted to assess the confidence in predictions.

DOI:

Year: 2026