Precision farming based technologies for olive grove management optimization

Abstract

Precision oliviculture aims to enhance the quality and productivity of olive orchards while reducing environmental impact through optimized resource utilization. The implementation of these strategies requires the development of methodologies to characterize the state of olive trees with high spatial and temporal resolution. This thesis aims to develop accessible methodologies for assessing the water and nutritional needs olive crops, as well as the olive fruit quality. The first milestone of this thesis has been the development of a thermal camera based on a low-cost infrared sensor to assess the water status of olive trees. The canopy temperature and the crop water stress index (CWSI) were compared with two standardized water status indicators (predawn leaf water potential and stomatal conductance), obtaining promising results (R² ≥ 0.80). A significant aspect of this work is the proposed method for obtaining the thresholds needed to calculate the CWSI. This approach simplifies the automation of the process, as the reference limits are extracted from the temperature histogram, avoiding the need to measure artificial surfaces or meteorological variables. The second milestone involved developing a methodology to assess the nutritional status of olive trees based on analysing and modelling spectral images captured by an unmanned aerial vehicle. A crucial step of this methodology is the proposed image processing technique. It utilizes a digital surface model to filter out background information, improving the quality of spectral data by reducing the impact of background noise. The 5 reflectance data extracted from the images were used to train various modelling tools (partial least squares regression, artificial neural network (ANN), support vector regression, and gaussian process regression) using reference values of NPK leaf content. The ANN models achieved the best results (LNC: R² = 0.63; LPC: R² = 0.89; LKC: R² = 0.93). The third milestone focused on developing a low-cost multispectral device capable of characterising key quality parameters of intact olives. A prototype based on a commercial sensor was initially built and evaluated in a controlled laboratory experiment. The 18 reflectance values acquired by the sensor were used as input for ANN models, with three key olive quality indicators serving as reference data: moisture (M), titratable acidity (TA), and oil content per fresh weight (OCFW). The results obtained from the ANN models were promising (H: R² = 0.78; TA: R² = 0.86; OCFW: R² = 0.62). Encouraged by the laboratory results, a portable device based on the same sensor was developed. Its potential was evaluated in a field experiment, taking spectral measurements on-site. The results of this work were encouraging as the estimates of the oil content per dry matter (OCDM) (R² = 0.86), OCFW (R² = 0.86), and M (R² = 0.89) were better than those obtained under laboratory conditions, although the estimation of TA (R² = 0.21) was worse. Alternatively, the potential of the device to characterize quality indicators of red grapes in field conditions was evaluated, obtaining good results (Solid soluble content: R² = 0.70; TA: R² = 0.67). This suggests the potential of the device beyond olive trees. The results obtained in the research conducted in this Thesis indicate the potential of the developed solutions to support decision-making in the context of precision oliviculture. The low cost and ease of use of the proposed solutions make them accessible for all kind of olive growers.