European Journal of Sustainable Agroecosystems

Background: Agricultural irrigation takes nearly 70% of the total extracted freshwater from the earth. High-value crops such as grapevines, almonds, tomatoes, and potatoes suffer significant effects from water application since the value of their yield depends a lot on plant water consumption. Traditional irrigation scheduling techniques rely on established dates or ETo readings from limited weather stations which negate the variability of ETc rates over the fields. The new irrigation scheduling practices can be achieved through AI technologies, especially ML, DL, and DRL technologies, and satellite remote sensing, IoT sensor networks, and hydrological models, which can be applied together.
Purpose: The aim of this paper is to review published research from 2015 to 2026 about the use of artificial intelligence to schedule irrigations and evaluate its impact on the accuracy of evapotranspiration estimation and the effectiveness of water use (WW) concerning high-value horticultural and specialty crops. 
Literature sources: Literature was gathered from journals indexed by Web of Science and Scopus, records sourced from PubMed/PMC journals, reports from various international organizations like FAO, USDA, and good scientific databases like MDPI and Elsevier, collected through search processes on scientific websites like Springer, ScienceDirect, and Google Scholar.
Key results: Models that use AI technology, like randomized forest, long short-term memory networks, and hybrid ensemble architectures, more consistently provide better predictions than empirical ETo equations. When used together with machine learning, satellite-derived actual evapotranspiration (ETa) products, such as OpenET, allow irrigation recommendations to be generated in-field and almost in real time. Water conservation has been reported at 15-40% for specialty crops, often without affecting yield, and in some instances showing water productivity improvement.
Gaps in studies: Some gaps identify problems with model interpretation, validation in different agro-climatic regimes, difficulty in incorporation of soil-plant-atmosphere continuum indicators, high costs for implementation by smallholders, and missing long-term multi-season field studies on high-value crops.
Summary: AI-enhanced irrigation scheduling has potential to harmonize water-saving measures with yield and quality needs in high-value crop production, but for its wider implementation there is a need for transparent, inexpensive, and regionally proven schemes with the support of appropriate policies and extension activities.