Application and assessment of model-based leak localization methods in an irrigation network

Abstract

The availability of water resources in agriculture is under significant pressure due to climate change and global warming. In some regions, such as the south of Spain, rainfall has become increasingly scarce and concentrated over shorter periods, reducing the available water allocations and forcing the maximization of the irrigation system's efficiency. This situation highlights the importance of leaks within irrigation networks, which exacerbate the issue by decreasing the amount of water available for crops. These leaks not only represent a loss of water but also lead to increased operational costs due to the energy consumption required to maintain the network under pressure and in operation. Detecting and locating leaks is particularly challenging in buried irrigation networks, where root systems and depth complicate interventions, or in highly permeable or rocky soils, where the water loss caused by the leak can quickly move underground.

In this context, there is a pressing need to adapt and improve current leak detection techniques, which have traditionally been designed for water distribution networks (WDN), to the unique characteristics of irrigation systems. Unlike standard mesh WDN, irrigation networks exhibit a hierarchical, tree-like structure, with branches feeding drip lines, and with pressure regulating valves to control downstream pressure.

This work addresses the application of model-based methods designed for WDN for leak localization in irrigation networks. The evaluated methodologies rely on sensitivity matrices for leak management, considering pressure or flow variables as tools to determine measurement nodes. These include techniques such as binarization, vector angle, correlation comparison, Euclidean distance, and least squares; linear programming problem-solving; and the residual projection strategy, which monitors virtual leak flow values based on a dual model. These methods are analyzed against various magnitudes and interference sources, including additive and multiplicative noise, leak size, sensor placement, and more. Hydraulic simulations are performed on standardized WDN widely tested in the literature, such as Hanoi, Net3, and Quebra, as well as on a real irrigation network spanning 125 hectares of citrus orchards located in southwestern Spain. This work establishes an initial analysis and assessment of the state-of-the-art leak detection methods in a real irrigation network, highlighting the need for specific solutions for this type of infrastructure.