Data-driven predictive model forirrigation management in greenhouses under CO₂ enrichment and high solarradiation

Machine learning models have emerged as a viable method to predict plant water requirements with the ability to form non-linear correlations between plant response variations and microclimate conditions. An artificial neural network is used in this work to predict irrigation water requirements in a greenhouse located in a hyper-arid region with high solar radiation and encompassing an HVAC cooling system and CO₂ enrichment. The prediction model is developed from direct gas exchange measurements of transpiration and takes as input parameters microclimate data including greenhouse solar radiation, temperature, humidity and CO₂ concentration along with hyperspectral imaging-based vegetation indices. Results demonstrate the high performance of the data-driven artificial neural network (ANN) model with microclimate and vegetation index (VI) features. The ANN model resulted in a comparatively higher performance than the FAO56 Penman Monteith empirical model and linear regression with an R² of 91.2%, RMSE and MAE of 0.0648 mm/h 0.0528 mm/h respectively. The proposed data-driven predictive model will enable the determination of irrigation water supply in greenhouses under CO₂ enrichment and with varying microclimatic conditions, and support the implementation of precision irrigation.