Real-Time Anomaly Detection in Rural Water Systems with meter.me

A powerful solution for the hardest-to-reach places

By: Andrew Emerick for meter.me

Water is a vital resource. Groundwater, in particular, is a critical and sensitive component of the national water supply, providing approximately 27% of drinking water across the U.S. This percentage is even more significant for rural homesteads, irrigation for farms, and for particular states, such as California. However, the availability of clean and accessible groundwater is sensitive to how it is extracted and used. Over-extraction can lead to irreversible damage to groundwater aquifers and permanent loss of clean water. This can threaten the financial stability of farms that rely on groundwater for irrigation, the liveability of houses that need clean drinking water, and the safety of homes that require reliable water access to protect against wildfires.  It is essential to ensure existing groundwater is used efficiently through monitoring, leveraging data-driven solutions for automation, and active alerting in the event of leaks.   

The water systems that serve rural homes and farms are dependent upon maintaining consistent, controlled water access. These are multi-component systems consisting of a combination of some number of wells, pumps, connections, spigots, and storage tanks, and the piping to connect each of these. 

There are many potential failure points across these systems, and even with careful monitoring and regular maintenance, failures can and do happen. Leaks due to component failure or abnormal operation (such as a spigot accidentally being left open) are often extremely challenging to identify. This is in spite of the fact that these kinds of leaks can result in 10s or 100s of thousands of gallons of lost water. That represents a substantial increase in the groundwater extraction for a given water system, risking permanent damage to the source well and aquifer. 

Sustainable water use means proactive leak prevention and leak detection. The meter.me solution is designed specifically for rural applications, leveraging battery-powered, LoRaWAN communication technologies to avoid the pitfalls of traditional approaches that require wired power, WiFi, and/or cell phone reception. This is all at a price point that is 1/100 of legacy SCADA systems with a fraction of the installation overhead. Below, we discuss these points in more detail as we outline our solution for proactive leak detection and abnormal water use that can help detect and alert water loss in real time. We conclude with a few illustrative case studies demonstrating its application

The Problem

Estimates are that about 16% of water is lost across water systems in the U.S. Individual water systems can expect anywhere from 5-15% of their total water supply lost due to leaks throughout the year and up to 40 -60% in some cases. This can be substantial for rural homes and small farms, as leaks often go unnoticed for weeks to months, resulting in water loss events equal to or higher than the entire normal water usage for the year, setting the wasted water in hundreds of percent. When leaks can drain anywhere from a few to several tens of gallons per minute, the potential water loss can easily approach many 100s of thousands of gallons. 

It is not a matter of if but when, how often, how much water will be lost, and how long it takes to identify and fix the problem.

The problem is being able to actively and rapidly detect leaks in water systems that are composed of several components connected by underground pipes spread over many acres and often have little to no connectivity to power or the internet. 

How do we monitor this diverse network to enable active leak detection in minutes – yielding water savings in the 10s to 100s of thousands of gallons?

Leaks happen beyond the WiFi range

Modern leak detection technologies are not suited for large-scale, rural applications. The best consumer leak detection systems on the market today (for example) work extremely well if and only if there is access to wired power, WiFi, cell reception, and/or areas where water may collect and pool. This works well for applications within homes or other structures with access to power and the internet or applications where it is feasible to install sensors in every possible location where a leak could occur. 

But rural water systems can extend over acres with many – if not most – components not only outside of WiFi range but also outside of cell phone reception and without access to a power source. Retrofitting a system to add this kind of connectivity is almost always prohibitively expensive for all but the largest commercial farms. Even if this were to be done, this still leaves unmonitored the many hundreds or thousands of feet of underground pipes connecting it all together. 

Monetary Impact

On the one hand, rural, well-fed water systems generally do not pay for this water as a household connected to a municipal water supply would. So, in a sense, these kinds of leaks do not have a direct, immediate monetary impact. Leaks in fields do not have the same kind of costly structural damage implications that a flooded basement does. However, the costs are real, lasting, long-term, and expensive. Leaks leading to large, prolonged water losses can:

• Wells: Wasting the water supply in wells can lead to limited water supply in the short term and growth of the cone of depression around the well in the long term. This can have long-term implications for water access and water quality and risks necessitating the costly undertaking of drilling existing wells deeper or adding new wells to your property. 
• Electricity and Pumps: Water leaks mean running water pumps more often and longer than necessary to sustain normal water use, which has direct and immediate costs in terms of electricity consumption but also can—long term—hasten the need to conduct maintenance on or install replacements for pumps. 
• Fire: Ensuring access to a reliable water supply is crucial for protecting your home or farm during a wildfire. Water can be used to extinguish small fires and maintain defensible space. Sufficient water availability can significantly increase your property’s resilience against the spread of wildfire. Leaks threaten that availability.

Without monitoring and active anomaly detection, how do you know you have access to the water you need? Not just for daily use and irrigation, but in the worst-case scenario, how do you know your home has the water it needs to protect itself in the event of a fire?

Expanding on Water and Fire

In rural homes and farms, monitoring water supply ensures sufficient water is available to combat fires quickly and efficiently, enhancing the property’s fire resilience. In that way, being firewise means being waterwise. Water-based fire hardening is only as reliable as the supply of water. Yet this is often neglected in implementing fire-hardening solutions. Without active monitoring, ensuring there is sufficient water in tanks or pressure in the water system is operating on faith alone.    

Being firewise means being waterwise.

Our advanced water system monitoring solutions help maintain optimal water levels and detect leaks early, ensuring that water resources are always ready for emergency use. By being proactive in water management, you also strengthen your defenses against fire, making your home or farm safer and more secure.

Leak detection and mitigation – therefore – is an integral part of being firewise. [anecdote]

Existing Leak Detection Methods:

Water loss estimates and water leak detection technology is not new. However, all of them are resource-consuming (in either time, money, or both), rendering them impractical for active leak detection (within hours) or out of reach for all but large farms with substantial financial resources. This includes:

• Modern Digital, Active Leak Detection Systems: As mentioned, these systems, often powered by machine learning and AI,  work well – provided there is access to power and internet. Rural water systems are generally predominantly without access to power and internet, and retrofitting a system to support these solutions is prohibitively expensive. Some of these solutions are discussed in more detail below.  
• Physical Leak Detection: Various technologies exist for detecting water leaks directly, by measuring water accumulation. Many of these require either power or WiFi connectivity (or both), but there are exceptions that do leverage LoRaWAN communication. Regardless, these sensors require careful placement in areas where water may leak or accumulate; this is intractable at the scale of a rural water system save maybe for the most critical, accessible areas. 
• Water system audits: These can be expensive, time-consuming, and happen too infrequently (quarterly / yearly) to be of use in trying to address leaks as they happen. In addition, a proper audit requires flow meters to be placed frequently throughout a water system. Even if these are all iOT connected to enable remote monitoring and continual remote auditing, installing the necessary number of flow meters (e.g. at every entry and exit point in the water system) is invasive and can be cost prohibitive.
• Acoustic leak detection: whereby escaping water in underground pipes is identified by the sound or vibrations it makes in the surrounding environment. This is very accurate, but is extremely time consuming. This is more feasible for pinpointing a leak once one is known, rather than monitoring for one in the first place.
• Water pressure changes: with monitoring of water pressure in a water system, leaks can manifest as significant and  unexpected changes in that pressure. Similar to flow meters, this requires frequent water pressure sensors throughout the water system. 
• The hard way: you run out of much / all of the water in your water system or you notice substantial amounts of water pooling in areas where it should not be present. Both of these are exactly the situations we aim to avoid with real time, proactive anomaly detection.

All of these solutions have their place, but none provide the necessary reliability, coverage, and cost-effectiveness required for rural applications. Two of the biggest hurdles here are in power and communication. Rural applications require monitoring solutions that can function on battery or solar power alone, in remote areas without WiFi or cell reception. For this reason, LoRaWAN (long range wide area network)  is a powerful alternative for communication because it operates on a low-power, long-range radio frequency that can transmit data over several kilometers even in remote locations. This makes it ideal for rural and agricultural settings, enabling reliable communication for monitoring systems without traditional infrastructure.

The meter.me® Solution

meter.me enables water system monitoring by leveraging LoRaWAN technology to continuously monitor every possible component of a rural water system: tanks, wells, pumps, flow meters, ponds, generators, and more. This enables high-frequency, real-time data acquisition from components regardless of their connectivity to power, WiFi, or cell reception. 

Leak Detection is Anomaly Detection

Leaks, whether slower, gradual leaks caused by deteriorating pipes or loose connections, unintended use from a left-open faucet, overflowing tanks due to faulty pump shutoff controls, or burst pipes, are all anomalies in normal water system use. With real-time monitoring of the flow of water throughout a water system, meter.me captures a detailed historical record of what constitutes normal operation for any given water system.  This allows us to leverage machine learning techniques to build anomaly detection solutions to identify leaks in real time as they happen.

At its core, anomaly detection is about finding outliers in a dataset. The data meter.me captures time series data of water levels, flow rates, pump status, etc. – can be mined just the same way other kinds of data would be, searching for unusual patterns. Machine learning algorithms can then be trained on this data to learn historical patterns customized to the individual water system while simultaneously incorporating more general knowledge from analyzing data from many water systems. 

A very common application of anomaly detection is in credit card transactions, looking for patterns and individual transactions that may be fraudulent use. Detecting leak-driven anomalies in water systems has some very key parallels: they represent a very small portion of the entire history when identified action must be taken quickly or risk larger damages, and while every user is different, there are still general patterns that can be leveraged. Furthermore, care must be taken to avoid alerting on unusual (but in a way expected) use. In the same way, you don’t want your credit card declined and locked when you check into your hotel on a big once-a-year vacation, you also don’t want alerts just because you decided to take a longer shower than normal. There is a balance that needs to be struck to ensure accurate, trustworthy alerts that drive appropriate action to identify and stop leaks quickly. 

Enabled by real-time monitoring, meter.me employs real-time anomaly detection that quickly identifies anomalous use, and alerts end users within minutes of unusual activity. This permits action to be taken within minutes/hours of a potential leak being identified instead of the weeks to months that it would normally take.   

The solution is flexible and powerful. While the accuracy and time-to-action improve significantly, the more components of a water system are monitored by meter.me, this kind of anomaly detection is possible on an individual component basis. Knowing only the water level of a single tank is sufficient for anomaly-based leak detection. This means meter.me anomaly detection does not require the installation and monitoring of more costly and complex components such as flow meters or pressure sensors, and it can work without retrofitting the entire water system. Critical components can be monitored with meter.me today while the remainder of the water system is integrated as time and cost allow. 

Case Studies

We recently tested our new amplitude and frequency water use anomaly detection on an existing customer data set and noticed anomalous use patterns in a dozen ranches and farms. We contacted these customers, and all but one confirmed that our algorithms identified real water loss. 

In one case, our customer’s tank fill frequency had gone unnoticed. After reaching out, they suspected faulty meter.me equipment, but after updating their equipment, the behavior continued. It was only weeks later that they finally discovered a leak in their system that was just below the pump’s ability to keep the tank full, which further masqueraded the leak. By this point, it had been ongoing for two months, and well over 90,000 gallons of water had been lost. 

Anomaly: Normal use is clear here.  Prior to Feb 29 daily  water use was around  ~250 gallons per day, but after Feb 29 this spikes to over 1600 gallons. Automation maintained the water level in the tank, and without high frequency data sampling and meter.me, the leak would have gone unaddressed for a long time.

As large as 90k gallons is, this was typical for hard-to-identify leaks, especially during the rainy season. By their own admission, this loss could have been even more substantial (>> 100k gallons) had meter.me anomaly detection not been used at any point, continuing for many more months before being discovered. 

This is not an unusual occurrence. All of our customers can recall at multiple times they’ve had a leak in their water system that had gone unnoticed for days, weeks, or even months. Instances, for example, where they discovered an underground leak only after finding a small patch of their field looking unusually healthy during a hot, dry summer. 

In another example, a tank had a  leak that went undetected and unidentified for over two weeks, from late October into mid-November 2023. This went unnoticed because – similar to the prior example — tank fill automation maintained the tank water level fairly high, so the leak never resulted in an empty tank or triggered more basic tank minimal-level alerts. To the end-user, this would have gone unnoticed for even longer had they not happened to examine their tank data in the meter.me a couple of weeks after the leak began. In this case, the leak doubled the system’s daily water consumption from ~300 to ~600 gallons per day, resulting in a loss of 4000 – 5000 gallons over the period. meter.me anomaly detection would have detected this change in water use within days, saving the customer thousands of gallons of wasted water. 

Conclusion

Leak detection in urban environments and SCADA systems is a solved problem. Detecting water loss in rural water systems means disrupting legacy infrastructure that has been chronically ill for over a century. But, one farm using meter.me to sustainably consume water can easily trump water savings in an entire city block, which makes it all worthwhile. 

Low power, low cost, LoRaWAN, and sensors in multiple nodes are all important ingredients, but it is the high-frequency data coupled with being able to apply machine learning in the cloud that finds the anomaly and ultimately saves water…  a lot of water.