PREDICTING URBAN WATER QUALITY WITH UBIQUITOUS DATA

Abstract

Our everyday lives are greatly impacted by 
the quality of urban water. Urban water 
quality predictions aid in the prevention of 
water contamination and the safeguarding of 
human health. Predicting urban water quality 
is difficult because there are many variables, 
including weather, water consumption 
patterns, and land uses, that contribute to the 
non-linear variation in water quality in 
metropolitan areas. Here, we take a datadriven approach to predicting a station's 
water quality for the next several hours by 
combining information from several citywide sources, including weather, pipe 
networks, road structures, and points of 
interest (POIs), with water quality and water 
hydraulic data reported by existing monitor 
stations. Using a battery of tests, we first 
isolate the components with the greatest
impact on the water quality in cities. To 
further integrate these diverse datasets into a 
single learning model, we next introduce a 
multi-task multi-view learning approach. We 
test our technique on real-world datasets, and 
the results show that our approach is 
successful and that our method has benefits 
over other baselines.