Abstract:
Abstract: Greenhouse cultivation has a strong impact of crops on the complex process, because various time scales existed in the controlled environment. The greenhouse model can be divided into two types, including the crop growth and microclimate model. The crop growth model was usually used to simulate daily change of crop, where the variables of crop is updated at each time step within a day. The microclimate model has a shorter calculating step, because the climate in a greenhouse changed quickly, due mainly to rapid fluctuation of weather outside. In general, the climate physics is considered as a fast process, while the crop physics is considered as a slow one. The difference of time scales has brought a great challenge at the level of crop state, such as the rapid fluctuations of greenhouse climate or the elusive ambient inputs in the monitoring system of a greenhouse. In this study, taking tomato as a research object, a crop-climate interactive model at small timescale was established to balance the time scales of crop and climate in greenhouse. Firstly, the growth model was divided into three sub-modules, including the SUPPLY, PARTITION, and GROWTH. The replacement, structural transformation were implemented in the model, when three modules were transformed from a long timescale (day level) to a small timescale (second level). Two types of uncertain parameters were divided in the model under a global sensitivity analysis (Extended Fourier Amplitude Sensitivity Test, EFAST), such as sensitive and insensitive parameters. Insensitive parameters were fixed in the model, whereas, the sensitive parameters needed to be identified, according to real production data in specific greenhouses. Secondly, the general interactive model was obtained to combine small time-scale crop growth model and greenhouse microclimate model. Microclimate in the interactive model was different from other microclimate model, because it fully considered the reaction between the microclimate and crop, where the microclimate model was be considered as an input for the crop model. The proposed interactive model was also calibrated and validated in the field test. The real data was collected from A8 Venlo type greenhouse at Chongming Island, Shanghai of China. The 4-year (2015-2018) observed data of tomato yield was used in the model. It was found that the root mean square error (RMSE) between the simulated and real yield value was 7.3-18.85, and the average relative error was between 5.8% and 18%, both less than TOMGRO and Integrated model. The data demonstrated that the interactive model presented a better performance on the yield prediction of tomato. The microclimate simulation result also proved that the interactive model behaved a higher accuracy at different crop growth stages than that without considering the influence of crop growth. The average relative error was less than 10% for the prediction of microclimate environment at three stages of crop growth (growing, blooming and setting fruiting), indicating high efficiency to simulate the real dynamics of greenhouse microclimate. Nevertheless, there were relatively larger deviations in the small part of simulation from actual data, such as simulated yield in 2018 and temperature trajectory when LAI=2. Bayesian optimization was also used to identify the uncertain parameters in both crop growth and microclimate model. Model structure and parameters were totally determined after sensitivity analysis and parameter identification. Consequently, the interactive model can provide a theoretical basis for cultivation and environmental control in a greenhouse.