Full text
| |
| Author(s): |
Monique Pires Gravina de Oliveira
Total Authors: 1
|
| Document type: | Doctoral Thesis |
| Press: | Campinas, SP. |
| Institution: | Universidade Estadual de Campinas (UNICAMP). Faculdade de Engenharia Agrícola |
| Defense date: | 2022-03-04 |
| Examining board members: |
Luiz Henrique Antunes Rodrigues;
Daniel Wallach;
Fábio Ricardo Marin;
Alim Pedro de Castro Gonçalves;
Paulo Rodrigues Peloia
|
| Advisor: | Luiz Henrique Antunes Rodrigues |
| Abstract | |
Dynamic crop growth models coupled with the vast amount of available data have been seen as part of the answer to the problem of more resource-efficient agricultural production. Although such models require calibration steps without which their predictive performance may be insufficient to aid decision making, real-time monitoring could be able to overcome this need. Dynamic models and satellite images have been combined using data assimilation techniques to reduce the prediction errors of state variables related to crop canopy, soil properties, or yield. In protected environments, however, where the use of models and sensors allows the monitoring and automation of control systems so that it is possible to optimize environmental conditions for greater production profitability, the assimilation of monitoring data is not explored. This project aimed, then, to determine the performance of data assimilation techniques using environmental and crop sensing in a greenhouse, as well as to determine the acquisition frequency required and the technological level necessary for the approach to be replicated under production conditions. To do so, the meteorological factors of a greenhouse with tomato cultivation were monitored, as well as crop growth, through direct weighing of plants and the use of images captured with low-cost cameras. Using state estimation techniques such as the Unscented Kalman Filter and the Ensemble Kalman Filter, data assimilation in the Reduced State TOMGRO model was performed. Since the use of these techniques in protected cultivation had not been carried out, it was necessary to carefully characterize the elements that affect the performance of the models, as well as the filters. It was observed that: 1. depending on the growth model used, the assimilation of one state variable may not impact the others, as suggested by sensitivity analyses, 2. the quality of observations is crucial for good performance of the assimilation techniques, 3. the assimilation performed better when there was a need to adjust the estimates to growth disturbances, 4. when filters lead to better productivity estimates, continuous observations are not required. Although, in general, it has not been possible to obtain better performances than the calibrated model, this potential exists, as long as better observation models and better-quality observations are available (AU) | |
| FAPESP's process: | 18/12050-6 - Data assimilation for integration of data obtained by wireless sensor networks and a dynamic crop growth model |
| Grantee: | Monique Pires Gravina de Oliveira |
| Support Opportunities: | Scholarships in Brazil - Doctorate |