Detecting Outdoor Irrigators

Outdoor water use, especially irrigation, during the summer months can put a huge strain on public water systems. Many areas of the country are experiencing long term drought and are facing potential water shortages. And in all areas, effective management of peak demand can mitigate the need for expensive capital improvements to water supply, treatment and distribution systems. The chart below shows system-wide demand peaks (in green and yellow) over a ten year period, and it also shows a clear ‘summer spike’ usage pattern for one account with high residential irrigation water use (red line).

Many water systems, suppliers and regulatory agencies have responded to the stresses of high residential and commercial irrigation demand by implementing outdoor water usage restrictions. These measures typically restrict the days and hours during which outdoor water use can take place. Often these times are staggered based on property address such that ‘addresses with a number ending in an even number can water on Wednesday between the hours of 9pm and 10am’ etc. Another measure which is gaining popularity is tiered pricing for irrigation water use, often accompanied by a requirement that irrigation systems be placed on dedicated meters.

Implementing such measures poses challenges; requiring dedicated irrigation meters is easy enough for new construction, but what about existing structures? Do you retroactively enforce such a requirement? And if so how do you know which properties will be impacted, short of a system-wide, house-by-house audit process? Who has time for that? Similarly, if you have watering restrictions in place, how do you audit/enforce them?

One valuable technique is to use statistical analysis of historical meter data to identify which customer accounts are likely using outdoor irrigation systems during the summer months. This information can serve as a foundation for meter retrofit programs and for watering restriction compliance audits. Two common statistical techniques can be used in tandem to identify your likely irrigation clients.


Detecting Seasonality in time-series data can be achieved numerous ways as described in this great Wikipedia article. We use a variation of the ‘ratio to moving average’ method whereby we bucket usage on each account by quarter and then produce a ‘signal’ or ‘feature’ from the data which is an indicator of how much of that account's annual average usage is consumed during each quarter. This is a good measure of the ‘seasonality’ of that account’s usage pattern.


Once we’ve generated these seasonality features for each account, we need to evaluate all accounts and see which ones exhibit a ‘summer spike’ pattern which is indicative of high outdoor water usage and irrigation. In a large water system with many thousands of customers this is clearly not a job you want to do manually. So, we feed the seasonality features for all accounts into the K-Means clustering algorithm which partitions the accounts into groups (or clusters) which exhibit similar characteristics. In the example above, k-means has created two clusters which have the 'peak-and-valley' usage characteristic of heavy summer water use. Using our seasonality features produced with the ratio-of-moving-average approach outlined above, k-means creates clusters based on quarterly/seasonal usage levels, effectively defining characteristics such as ‘fairly even water use all year’, ‘summer spike usage’, etc.