AI-Driven Precision Irrigation in High-Density Apple Orchards: Comparative Analysis of Water Application, Soil Moisture Dynamics, and Fruit Yield
Riyaz Ashraf *
College of Agricultural Engineering and Technology, Sher-e-Kashmir University of Agricultural Sciences and Technology of Kashmir, Shalimar, Srinagar, Jammu and Kashmir, India.
Rohitashw Kumar
College of Agricultural Engineering and Technology, Sher-e-Kashmir University of Agricultural Sciences and Technology of Kashmir, Shalimar, Srinagar, Jammu and Kashmir, India.
V. P. Singh
Texas A&M University in College Station, Texas, USA.
*Author to whom correspondence should be addressed.
Abstract
Inefficient Irrigation practices in agriculture systems has encouraged the development of sensor-based irrigation systems to increase resource use efficiency and profitability. However, comparisons between automated and manual drip-irrigation systems must consider treatment effects as well as the volume of water applied. This study was conducted at Sher-e-Kashmir University of Agricultural Sciences and Technology of Kashmir, India. The study included 182 soil-moisture readings recorded over a 100-day period (15 June to 14 August 2021), together with irrigation data, namely the mode of drip-irrigation application (manual or automatic) and the volume of water applied in litres per irrigation event. The records contained 18 automatic irrigation events totaling 79,090 Litres and 36 manual events totaling 126,720 Litres, representing a 37.59% reduction in recorded irrigation water under automatic drip irrigation compared with manual drip irrigation. The system used three identical, indigenously developed soil-moisture sensors installed at a depth of 15 cm in the experimental plot having clay loam soil type. The mean of the three moisture-sensor readings was 54.96%, with a standard deviation of 18.45 percentage points. Soil moisture and relative humidity were positively correlated (ρ = 0.360, p < 0.001). The Random Forest Machine-Learning Model achieved a mean absolute error of 6.34 percentage points, compared with 10.74 percentage points for the persistence method. Harvested yields were 62.5 versus 36.8 kg for Golden Apple, 46.0 versus 22.7 kg for Decoste Robinjin, and 54.0 versus 28.75 kg for Mitch Gala (automated versus manual; five trees per irrigation mode and variety). These values represent descriptive increases in harvested mass of 69.84%, 102.64% and 87.83%, respectively, relative to manual irrigation. Normalized yield-water response scenarios over the 100-day period illustrate potential improvements in water management. The study suggests that predictive irrigation is feasible and provides data relevant to evaluating irrigation volumes, water-use efficiency and crop yield for resource use efficiency and profitability.
Keywords: Precision irrigation, soil-moisture sensors, machine learning, random forest, water-use efficiency, fruit yield