Filmes de partículas de óxido de cálcio e silício em campo e casa de vegetação, na avaliação de déficit hídrico e aprendizagem de máquinas na predição da produtividade de abóbora “mini jack”

Abstract

Climate change has intensified the frequency and severity of drought periods, increasing risks to agricultural production and the need for strategies capable of enhancing crop tolerance to water deficit. In this context, calcium oxide (CaO)-based particle films and silicon (Si) have the potential to modulate physiological responses and alleviate the effects of stress, while machine learning enables the integration of biological and agronomic information for predicting crop productivity. This study evaluated the effects of CaO and Si on mitigating water deficit in ‘Mini Jack’ pumpkin (Cucurbita moschata) plants, considering physiological, biochemical, growth, and productivity responses, as well as the potential of machine learning algorithms for productivity prediction. The study was structured into three articles. In the first article, different CaO concentrations (0, 2.5, 5, and 7.5%) were evaluated under two irrigation regimes (66 and 100% of crop water requirements). CaO application modulated photochemical functioning and productivity performance, with the 5% concentration promoting electron transport, preserving photochemical efficiency, and improving yield. In contrast, the 7.5% concentration resulted in less favorable responses, highlighting the importance of optimizing particle film concentration. In the second article, the individual and combined applications of Si and CaO were evaluated under different water regimes. Water deficit impaired plant growth and altered photosynthetic pigments, the chlorophyll a fluorescence transient, and osmotic and antioxidant responses. CaO alone contributed to maintaining leaf area under water deficit, whereas the Si + CaO combination increased chlorophyll indices, preserved photochemical functioning, and reduced proline accumulation As well as CAT and APX activities, indicating lower oxidative stress and a predominantly preventive strategy for protecting the photosynthetic apparatus. In the third article, Linear Regression, Random Forest, and SMOreg algorithms were evaluated for productivity prediction using progressively integrated datasets comprising pigments, OJIP parameters, biometric traits, and management variables. Pigments alone showed low predictive capacity, whereas the incorporation of OJIP parameters improved model performance, with Random Forest showing the best results. The integration of photochemical, pigment-related, and biometric information provided the highest predictive performance, with SMOreg outperforming the other algorithms, demonstrating that biological complementarity among predictors was critical for accurately representing productivity. Overall, the results demonstrate that CaO and its combination with Si are promising strategies for modulating the responses of ‘Mini Jack’ pumpkin plants to water deficit, while the integration of plant ecophysiology and machine learning enhances the ability to relate crop management, photosynthetic functioning, growth, and productivity.