EGU21-13092, updated on 04 Mar 2021
https://doi.org/10.5194/egusphere-egu21-13092
EGU General Assembly 2021
© Author(s) 2021. This work is distributed under
the Creative Commons Attribution 4.0 License.

Machine learning-based tools for water digitalisation

Asma Slaimi1,2,3, Susan Hegarty2,5, Fiona Regan2,6, Michael Scriney1,4, and Noel O’Connor1,3
Asma Slaimi et al.
  • 1Insight Centre for Data Analytics, Dublin City University, Glasnevin, Dublin, Ireland (asma.slaimi@insight-centre.org, noel.oconnor@insight-centre,michael.scriney@insight-centre.org)org)
  • 2DCU Water Institute, Dublin City University, Glasnevin, Dublin, Ireland (fiona.regan@dcu.ie, susan.hegarty@dcu.ie, asma.slaimi2@mail.dcu.ie)
  • 3School of Electronic Engineering, Dublin City University, Glasnevin, Dublin, Ireland (asma.slaimi2@mail.dcu.ie, noel.oconnor@dcu.ie)
  • 4School of computing, Dublin City University, Glasnevin, Dublin, Ireland (michael.scriney@dcu.ie)
  • 5School of History and Geography, Dublin City University, Drumcondra, Dublin, Ireland (susan.hegarty@dcu.ie)
  • 6School of Chemical Sciences, Dublin City University, Glasnevin, Dublin, Ireland (fiona.regan@dcu.ie)

Advanced technologies have proven to deliver significant outcomes in the water management sector. New technologies provide the capability to collect and correlate the information from remote devices, introducing smart tools that can leverage augmented intelligence for interpreting structured and unstructured, text-based or sensory data. However, most of the single feature or non-sequential prediction machine learning methods for understanding water quality achieve poor results due to the fact that water quality information exists in the form of multivariate time-series datasets.

At the catchment scale, there are many layers where relevant data needs to be measured and captured. For that, data warehouses play an essential role in decision support systems as they provide adequate information. 

In this paper, we started by extracting, transforming, cleaning and consolidating data from several data sources into a data warehouse. Then, the data in the warehouse was used to develop a computer tool to predict river water level using Artificial Neural Networks (ANNs), in particular, Long Short-Term Memory networks (LSTM). As the prediction performance is significantly affected by the model inputs, the feature selection step, which considers the multivariate correlation of water quality information in terms of similarity and proximity, is particularly important. The features obtained from the previous steps are the inputs to the prediction model based on LSTM, which naturally takes the time sequence of water quality information into account.

The proposed method is applied to two different catchments in the island of Ireland. Experimental results indicate that our model provides accurate predictions for water levels and is a useful supportive tool for water quality management. 

Ultimately, digitised representations of water environments will guarantee situational awareness of water flow and quality monitoring. The digitalisation of water is no longer optional but a necessity to solve many of the challenges faced by the water industry.


Keywords: Water digitalisation, water quality, data warehouse, machine learning, predictive model, LSTM.



How to cite: Slaimi, A., Hegarty, S., Regan, F., Scriney, M., and O’Connor, N.: Machine learning-based tools for water digitalisation, EGU General Assembly 2021, online, 19–30 Apr 2021, EGU21-13092, https://doi.org/10.5194/egusphere-egu21-13092, 2021.