Thermo-gas dynamics affect the leaf canopy shape and moisture content of aquaponic lettuce in a modified partially diffused microclimatic chamber

Abstract Precise and timely monitoring of leaf water content and growth is necessary for precision agriculture. In this study, the impact of temperature and gas concentration variations was synthesized with leaf canopy shape and moisture content of aquaponic lettuce in a customized partially diffused microclimatic chamber equipped with an RGB camera. Using the acquired thermo-gas sensor data during the light-dependent photosynthetic reaction and dark-period aerobic respiration, it was analyzed with the extracted spectro-morphological leaf canopy signatures. Applying spectro-morphological leaf canopy data as input to the recurrent neural network it was possible to sensitively predict full moisture content (FMT) and equivalent water thickness (EWT) with 92.34% and 85.96% accuracy, respectively. A decrease in red green blue vegetation index (RGBVI) and an increase in excess of green (ExG) optical vegetation stress index corresponds to an increase in FMT. According to our data, it was possible to observe that the water content balance during the third week after sowing, follows the same trend observed for the oxygen production increase during the light period. Additionally, it was also possible to observe that uneven leaf broadening (canopy shape index > 1), low EWT, high photosynthetic quotient (PQ), and photothermal unit (PTU), and minimal temperature and humidity stretch resulted in higher oxygen production rates. A growing degree day of 23.46 ± 0.30 °C day is ideal for aquaponic lettuce. Plants with six (6) weeks exhibited the highest thermal sensitivity (Q1) of 0.205. Overall, the developed model and observations allow concluding that the combined temperature-O2 CO2 dynamics monitoring must be considered to achieve maximum production rates. Moreover, in the framework of the new approaches developed within precision agriculture schemes, the collected optical data allied to the recurrent neural network are suitable tools for this continuous monitoring in an automated way.