Uncertainty in greenhouse tomato growth models

Fresh tomatoes are an important component of a healthy diet, and their cultivation is often facilitated by protected structures, especially in harsh climates. While greenhouse tomato models have potential applications in controlling these environments, they have not been extensively evaluated. This study aims to characterize the sources of uncertainty in such tomato models, particularly in greenhouse tomatoes grown in low-technology environments. These environments, where optimal environmental conditions may not be taken for granted, are often overlooked in existing studies. We used three tomato models: two designed for protected environments (the Reduced State Tomgro model and the model developed by Vanthoor et al.) and one was developed for field growth, the Simple model. We conducted the first sequential global sensitivity analysis for different weather conditions in tomato models. This approach allowed for uncovering the most important parameters through growth and ensuring the parameters chosen as most important would likely be important in multiple seasons and not only the season in which growth was observed. For instance, for the models and locations studied, intra- and inter-annual weather variability affected parameter importance more than which location was evaluated. These results informed our choice of parameters to be calibrated, and the preliminary evaluation of model performance suggested that both models developed for protected environments may be adequate to represent indeterminate growth in both locations assessed. For the Simple model, some adaptations are required, such as in the radiation use efficiency and a continuous growth of new leaves to compensate for senescence. The uncertainty analysis, also derived from the assessment in different weather conditions, showed that errors in the estimates of the most important parameters, which are often related to environmental factors, such as the fruit abortion temperature threshold for the Reduced Tomgro model or the threshold temperature for crop growth inhibition for the Vanthoor model, impact the outcome substantially. These aspects need to be considered when using these models for low-technology greenhouses, since these environments do not ensure optimal temperatures. Therefore, parameter importance determination and calibration should be performed with data that comprise these more extreme conditions, which cannot be mitigated by technologies like ventilation or evaporative cooling.