According to the International Water Association and the World Bank, a 30% gap exists between the volume of water delivered to end customers versus the amount of water that is billed (fig.1). This Non-Revenue Water (NRW) costs the global economy $14 billion every year, a massive discrepancy that, along with the rapidly increasing urban population growth, climate change and ageing infrastructure, are some of the many global challenges the water industry must address with urgency in order to keep up with the pace of growing demand and the reality of water scarcity. 
Fig 1 – Typical Water Balance Equation showing the gap between Revenue and Non-Revenue Water. Source: AySa – Buenos Aires Water Utility
Today, progressive water utilities are using digital technologies to extract the maximum amount of information within their operational data to provide actionable insights into the condition of their networks. With more employees working remotely, this will help utilities facilitate the automation of the water loss process.
DMAs and NRW Management
To achieve a better water balance and a deeper understanding of their networks, water utilities across the globe have been dividing their distribution networks into sectors called District Metered Areas (DMAs) since the late 80s. In short, a District Metered Area (DMA) is an area with strict hydraulic boundaries within a distribution system with measured inflow into this discrete area (fig. 2). This configuration allows engineers and operators to closely monitor flow and apply Minimum Night Flow techniques, where measured night minimum inflow is compared to the customer’s night consumption to calculate water loss, thus helping identify anomalies in the networks.
According to the DMA Guidance Notes a district should have between 500 and 3.000 service connections and consist of up to 10 miles of pipeline. When the water utility lacks the resources or information technology to have in place an hydraulic condition modelling, a separation into districts with strict boundaries is a simple and efficient method to get an overview of the actual condition, making it easier to classify water loss by water balance. 
Fig 2 – Typical DMA configuration Source: WHO (2001) Leakage Management and Control. 
Machine Learning Aided Tools
The use and application of computer aided systems such as Automated Meter Reading (AMR) and Supervisory Control and Data Acquisition (SCADA) have been, for decades, the norm in water distribution systems to monitor the day to day operational activities, allowing massive amounts of data points to be gathered and analyzed, and enabling the water utilities to act in the shortest amount of time when an anomaly is detected. However, this siloed system required constant human supervision and technical expertise that is often also lacking.
For the first time technology has given us the ability to build near real-time water balances to target leakage through smart meter data and granular DMA pressure and flow condition monitoring, creating a significant opportunity for water utility companies to search for innovative approaches for early leak detection using Machine Learning Techniques (fig. 3) that are, not only more accurate, but reduce the burden of data analysis exponentially.
Fig 3- Process chart of software based leak detection. (Navas, 2016) 
One of the advanced statistical approaches that could be used for leaks detection and classifications is the use of multi-class SVM advanced pattern recognizer which includes Water Distribution Systems historical data to create recognizable signature of different type and scale of leaks and breaks throughout the water distribution pipeline networks system, without implementation of the actual DMA. 
Artificial Neural Networks (ANN) operating on quasi-static pressure and flow readings have been used for leak detection as well. Using this process, a series of patterns characterising the system status in several conditions of normal or abnormal operation are determined and the ANN trained successively on the data sets representing the system status (i.e. pressures and flow rates) in different operating conditions, either with or without leakages. Satisfactory leakage identification and location performance was obtained using their system; in particular, the ANN was always able to correctly identify the leaking branch. 
Stor Water Machine Learning Tools
Stor Water ML tools are being developed to help the industry improve the way in which we tackle the massive NRW water gap. Our tools will allow you to increase your leak detection rate and to visualize the likelihood of leakage for every water main across your DMAs and infrastructure, while giving you an holistic view of the entire system from a single management point
Get in touch with us and let’s solve water together!
 Bill Kingdom, Roland Liemberger, Philippe Marin. (2006). The Challenge of Reducing Non-Revenue Water (NRW) in Developing Countries. WATER SUPPLY AND SANITATION SECTOR BOARD, 1, 52
 AWWA. (2002). Minimize System Losses by Implementing Water Loss Controls. Feb 2021, de AWWA
 Farley, Malcolm, World Health Organization. Water, Sanitation and Health Team & Water Supply and Sanitation Collaborative Council. (2001). Leakage management and control : a best practice training manual / Malcolm Farley. World Health Organization
 Navas (2016). Leak detection in virtual DMA combining machine learning network monitoring and model based analysis
 D. De Silva, J. Mashford, & S. Burn, Computer aided leak location and sizing in pipe networks. Urban Water Security Research Alliance, 2009.
 Using neural networks to monitor piping systems, Process Safety Progress, Vol. 22, No. 2, pp.119-127, 2003.