CITRIAFORO Operational Group: Use of new technologies for the correct estimation of citrus capacity

• Mobile app for estimating fruit count at the tree level and protocol for standardizing the acquisition of gauging data using a mobile device.
• Pre-industrial prototype of an image acquisition device for capacity prediction mounted on a tractor or land vehicle.
• Protocol for the standardization of drone-based capacity data acquisition and a management platform for images acquired by land and air vehicles.

• Supra-plot capacity prediction model based on the fusion of information from various sources and the implementation of a digital platform for estimating capacity at different levels.

This result has been achieved through the implementation of a series of algorithms to estimate field performance and the design of a platform for monitoring and managing the data and results of the implemented system.

The developed algorithms consider various parameters to determine, as accurately as possible, the expected yield or production in kilograms from the field, 90 days before the fruit is harvested. This process is particularly complex because, at that time, the fruit is still in an immature, green state, making it even more difficult to identify and count the pieces on the tree.

In addition to counting the number of fruits on the tree, specifically within a selected quadrant, it is necessary to determine the average size or caliber of the fruits. From this size, the unit weight of each fruit can be estimated and then extrapolated, using growth evolution curves, to the expected weight just before harvest.

Finally, the resulting weight, based on the number of pieces, their size, and weight, is projected onto the total number of trees in the evaluated field. This gives us a result in kilograms or tons of total fruit production that the field will yield.

Outcome 3 of Specific Objective 1 has been achieved through the definition and implementation of a protocol specifically designed for collecting data and information to feed the performance forecasting system. This ensures that, both now and in the future, the system can self-improve, increasing its accuracy in delivering performance forecast results.

The protocol considers various parameters such as:

• Cultivation and variety
• Location
• Time of year
• Growth evolution curves
• Historical production data
• Tables of equivalence between calibers and weights
• Fruit count
• Correction indices
• Plot area
• Number of trees on the plot

All of this data is collected and stored in a database specifically designed to exploit this information on a large scale and use it to improve the system and increase accuracy, reducing the margin of error in the results.

The protocol is implemented both on mobile devices, through the installable app, and on the server, where the algorithms and data models that perform the counting and calculations are hosted, which requires high processing capacity.

The protocol enables the secure and rapid transmission of information between users' mobile devices and servers. Furthermore, scalability was considered during its development, allowing for the integration of other technologies in the future.

A pre-industrial field imaging device has been developed, designed to be mounted on a tractor. It operates autonomously while the driver performs routine agricultural tasks.

Various image capture and analysis systems were evaluated to identify the optimal solution. Using 3D design tools, the structural components required for device assembly were manufactured from plastic.

The system consists of two cameras positioned to inspect both sides of the crop row, a real-time processing unit capable of counting fruit, and a touchscreen with a simple user interface. Power is supplied directly from the vehicle's battery.

Tests confirmed that the most effective algorithms are based on YOLO (You Only Look Once), especially its latest version (v11). To develop real-time decision models, more than 50,000 images of citrus fruits were captured and labeled, covering multiple varieties and lighting conditions, which contributed to the model's high robustness.

The device was field-tested in commercial orchards, yielding preliminary results with an accuracy of 80–85% based on the total visible fruit, which corresponds to 60–70% of the actual total count. These results demonstrate the system's potential as a reliable support tool for yield forecasting under real-world conditions.

Objective 1 activities, under Outcome R3, developed a methodology that uses multispectral drone imagery and vegetation index analysis to identify optimal sampling points. This allows farmers to measure fruit yield (number and weight), assess production before harvest, and improve forecasting calculations and associated logistics.

Key Results:
• Optimal sampling points improved through soil variability studies and analysis of historical vegetative development.
• High-resolution georeferenced maps for decision-making based on homogeneous areas within the plot.
• Critical periods for drone flights were identified during the campaign, reducing costs by focusing on key growth stages.

Added Value/Benefits:
• Profitability: Fewer drone flights and the ability for farmers to perform them independently.
• Forecast Accuracy: Increased accuracy in performance predictions through strategic measurement of performance.
• Precision agriculture: Optimizing water and fertilizer use through mapping of homogeneous areas and guided navigation.
• Increased productivity: Correlation between NDVI data, soil variability parameters, and historical yields to improve crop performance.

Practical Applications:
• Real-time monitoring of crop health and variability.

• Efficient irrigation planning and nutritional management using NDVI data.
• Optimized field operations with georeferenced maps.
This tool promotes sustainability, precision agriculture, and profitability, directly benefiting farmers.

Objective 2 activities, under Outcome R3, enable an intuitive web interface that integrates the optimized point samples resulting from Objective 1. They also enable efficient management and analysis of agricultural data to support decision-making and optimize crop management.

Key Results:
• Georeferenced maps with optimized point samples on farms.
• Data integration that includes zonal soil maps, vegetation indices obtained through drones and satellites, and weather forecasts.

Benefits or Added Values for Farmers:
• Precision Agriculture: Allows for targeted interventions using vegetation indices to monitor crop behavior and moisture levels.
• Reduce costs by focusing on critical growth stages and optimizing inputs such as water and fertilizers.
• Increased Productivity: Facilitates the identification of optimal sampling points and homogeneous areas for better performance prediction.

Practical Applications:
• Real-time monitoring of crop health and variability using interactive maps.
• Crop management planning based on data trends.
• Navigation tools guide users to sampling points or problem areas using GPS integration. Comparison tools allow for parallel analysis of vegetation indices across different dates or treatments.
This platform promotes sustainability, precision agriculture, and profitability by providing farmers with practical information tailored to their specific needs.

A demonstration platform for simulating citrus yield prediction at the supra-orchard scale has been developed, integrated into the project platform, and developed as an interactive application in Google Earth Engine (https://molto-enr.users.earthengine.app/view/citriaforo). This platform demonstrates how large-scale planning and decision-making can be supported by accessible and visual yield estimates.

The platform uses over 3,000 anonymized orchards with scaled data. It predicts total yield based on individual orchard area, agroclimatic variables (radiation, evaporation, air and soil temperature, and precipitation), and Sentinel-2 spectral indices: NDRE (related to vegetation vigor), NDMI (related to water content), and PSRI (related to senescence). The tool can be tested on three varieties. Total yield prediction error is also calculated.

The simulations were trained with reduced data samples (15% if there were more than 700 orchards, 25% otherwise). Ninety percent of the simulations showed total yield errors of less than 10%.

The tool's adaptable design meets the needs of agricultural businesses and organizations, facilitating field planning and yield forecasting. Its open-access implementation encourages knowledge sharing and collaboration among stakeholders.

Outcome 2 of Objective R4 enabled the integration of multiple data sources from other activities into a single visualization platform, improving the user experience for efficient agricultural planning.
Key Results:

• Creation of a digital cloud platform that integrates data from various sources and other project activities (satellite, ground, and air).
  

• Added Value for Farmers:

  This result provides farmers with an accessible tool that optimizes their experience by reducing the number of requests and data accesses, improving market competitiveness, and encouraging technology adoption in the citrus sector. It promotes generational renewal and facilitates the inclusion of young people and women.

• Practical Applications for Farmers:
  Centralization of harvest forecasting at different levels of data acquisition, facilitating organization and query strategies.