Abstract
In the Mediterranean region, the quantity of water utilized for agricultural purposes ranges from 50 to 70%. Among the most water-demanding agricultural sectors are arboriculture and perennial crops. Orchards are particularly reliant on irrigation, a dependency that has been further intensified by climate change and the resulting reduction in water resources. This study aims to classify farms at the watershed scale according to their irrigation water consumption, and starting from this classification we aim to propose a method for estimating water consumption for irrigation at large scale and for heterogeneous land covers. The classification employed a variety of statistical methods to ensure robust results, including machine learning and regression approaches. Each method was applied independently, and the most common class allocation was retained. The study was conducted in the Ouvèze-Ventoux basin in south-eastern France, using data from various sources at both field and watershed scales. The data obtained from 21 farms provided accurate information on irrigation water usage, which was validated by data from the watershed's water manager. The benchmark analysis identified farms with high irrigation rates with 90% accuracy. Within these heavily irrigated orchards, a second benchmark identified heavily irrigated plots with 68% precision. Maps estimating water consumption were created at the watershed and municipal scales. The estimated total irrigation water use closely matched the actual consumption, with only a 14% deviation. This methodology offers an accessible estimation of water consumption at the watershed scale, without the need to rely on crop models. Moreover, the methodology accurately identifies areas with high irrigation demand based on actual irrigation practices.