Hyperspectral spectroscopy and machine learning for non-destructive monitoring of leaf water and chlorophyll content under greenhouse lighting conditions
Mohammed AlDwairi, Fahim Abdel Gafoor, Mariam Alcibahy, Abdel Rahman S. Alsaleh, Adil Al-Mahdouri, Sona Alyounis, Delal E. AlMomani, Tiejun Zhang, Maryam R. Al-Shehhi
Abstract
Monitoring plant conditions in greenhouses often requires extensive labor work and necessitates smart non-destructive solutions. This study explores a new approach of combining hyperspectral spectroscopy with machine learning for real-time monitoring of leaf water content (LWC) and relative chlorophyll content (RCC). It is successfully applied to cherry tomatoes across three greenhouse types: Polycarbonate (PCGH), Radiative Cooling (RCGH), and Shaded (SGH). By comparing Partial Least Squares Regression (PLSR) and spectral indices, we demonstrate that PLSR is more effective for LWC estimation (R² = 0.67 in PCGH; 0.65 in SGH), while spectral indices provide a simpler alternative for RCC monitoring (mND705 R² = 0.70 in PCGH; CI R² = 0.51 in SGH). The results highlight the significant impact of greenhouse lighting conditions on predictive accuracy, with SGH yielding the most reliable results due to stable light diffusion. The identified new spectral regions (542 nm, 564 nm, and 1134 nm) are strongly correlated with LWC variations, presenting novel insights into plant water status monitoring. Additionally, our temporal analysis reveals distinct LWC and RCC variations across environments, emphasizing the role of sunlight exposure. The developed model can be integrated into real-time greenhouse monitoring systems, IoT-based decision support tools, handheld hyperspectral scanners, UAV-mounted sensors, and precision irrigation strategies for smart plant health assessment and resource management towards sustainable agriculture.