Today there is no systematic way to identify and locate failures of irrigation systems mainly because of the labor costs associated with locating the failures. The general aim of this study was to develop an airborne thermal imaging system for semi - automatic monitoring and mapping of irrigation system failures, specifically, of leaks and clogs.

Initially, leaks and clogs were simulated by setting controlled trials in table grapes vineyards and olive groves. Airborne thermal images were acquired over the plots. The canopy temperature of the trees under the different treatments was compared to measured values of ​​stem water potential (SWP). Initial results showed that it is possible to identify abnormalities of approximately 11 bars in olive groves and 3 bars in table grapes.

Consequently, detection of real faults was attempted: a preliminary algorithm was developed to identify suspicious areas (suspected faults) based on the distribution of canopy temperature. To examine the accuracy and reliability of the algorithm five sites were selected with olive groves (Gshur and Revivim), vines (Lachish), palm dates (Kalia and Almog) and almonds (Lavi). Assessing the accuracy and reliability of the algorithm, it was combined with the estimated potential savings in labor. The results indicated the following: 1. according to the map produced by the algorithm, 14-20% of the area has to be scanned, which corresponds to a 60% saving of the time needed to scan the whole area 2. The automatically detected suspicious areas contain 80% of the visible faults. 3. Most of the area that is detected for scan will not contain visible faults. However, it was found that in 70% of the locations suspected for leaks the trees indeed received excess water relative to their surroundings, and in 90% of the locations suspected for clogs the trees suffered from lack of water relative to their surroundings. That is, the map produced by the developed algorithm allows to save about 60% of the scan time, detects about 80% of the visible leaks, and detects leaks and clogs that are not visible with a reliability of 70% and 90% respectively.

Despite the benefits of the semi - automatic algorithm it requires input of four empirical parameters that can change its performance: minimum distance between pixels to create the histogram, size of pixels clusters associated with leaks and clogs, parameters associated with image enhancement (erosion) and allowed tolerance in estimating the  location of the leak or clog. The algorithm has not yet been tested on palm dates and we intend to test it in the coming months.