The olive grove of the future: how to select the best materials for productivity… 93 olive oil, reinforcing the role of precision agriculture in modern olive cultivation. The olive cultivation research team at INIAV’s Elvas Innovation Centre – Herdade do Reguengo, alongside conservation, the study of native olive tree materials and genetic improvement, has been involved in studies on the development and application of digital monitoring tools for olive groves. a)Artificial Intelligence and Data Science Solutions in Digital Agriculture The optimisation of rational fertilisation in crops, particularly under more intensive conditions, is of great importance for the efficient use of nutrients, for reducing losses through leaching and/or volatilisation, and for food quality and safety. As part of an approach to reduce the cost of nutritional analysis, three fertilisation regimes were implemented in experimental plots of hedgerow olive groves at Herdade do Reguengo: a) the recommended regime for hedgerow olive groves; b) a higher rate than recommended; and c)a control treatment (with levels below those recommended). Three types of reflectance sensors were used (Figure 5). The variations induced by fertilisation have, sooner or later, repercussions on the nutrient content of the leaves, such as the ratios between the main macronutrients – Nitrogen (N), Phosphorus (P) and Potassium (K) – which result in changes to their colour. To assess leaf reflectance in the laboratory, we have been using three types of sensors, namely: a sensor considered to be low-cost (AMS S7265x 18channel multispectral sensor), developed by the University of Huelva (Spain); a high-resolution spectrometry sensor from Ocean Insight (FLAME-TXR1); and a multispectral sensor (MicaSense RedEdge-MX, AgEagle Aerial Systems Inc., Wichita, KS, USA) mounted on an UAV – Unmanned Aerial Vehicle (HEIFU®, Beyond Vision, Lisbon, Portugal). The images were captured by the multispectral sensor under field conditions, using a UAV flying over the experimental plot. On the same dates that the sensors were used, leaf samples were collected for chemical nutrient analysis in order to ‘train’ and ‘validate’ the machine learning models by correlating the collected data. The machine learning models revealed a strong correlation between the sensor data and the nutrient levels in the leaves. The multispectral sensor performed best for the nutrients P and K (R²= 0.75 and R²= 0.73, respectively). The FLAME spectrometer was more accurate for N (R²= 0.64). As for the low-cost sensor, the results were not promising, and a different approach is required. In conclusion, these results have demonstrated the potential of optical sensorbased technologies for the non-destructive, real-time monitoring of chemical nutrients. Figure 5 – Low-cost sensor (left), ‘FLAME’ spectrometer (centre), multispectral sensor mounted on an unmanned aerial vehicle (right)
RkJQdWJsaXNoZXIy MTgxOTE4Nw==