Conference Agenda

Organic beekeeping in the EU remains limited, with organic hives accounting for roughly 2%, indicating substantial growth potential. A key constraint is selecting apiary locations distant from intensive agriculture. This study explores how Geographic Information System (GIS) tools can identify suitable sites for organic beekeeping in Latvia by integrating spatial land data with organic status indicators. Agricultural field data from the Latvian Rural Support Service, including area, crop types, and attributes, were analyzed. A Web GIS application was developed in R and deployed on an open-source Shiny Server under Ubuntu. The system, BeeLand Bio, is publicly available at: https://imaginarium.lbtu.lv/beeland_bio and demonstrates how national scale agricultural datasets can support decision making in organic beekeeping. Users can select potential or existing apiary sites, define a foraging radius, and evaluate surrounding land for organic management and crop suitability for honey bees. Results confirm the feasibility of combining agricultural spatial data with interactive geospatial analysis to support compliance with organic standards. The platform provides a scalable, interdisciplinary approach for precision beekeeping and strengthens the integration of ICT tools into apiary planning and management. It also encourages data-driven collaboration among researchers, beekeepers, and policymakers to promote sustainable land use and biodiversity conservation across regions.

Tomato Brown Rugose Fruit Virus (ToBRFV) is a highly infectious Tobamovirus that threatens global food security, primarily spreading through the international seed trade. This research proposes a noninvasive, real-time detection technique using Hyperspectral Imaging (HSI) and machine learning techniques to identify the virus in seeds during the asymptomatic stage. Utilizing a dataset of 30,000 tomato seeds, the study employs VNIR (400–1000 nm) and SWIR (1000–2350 nm) sensors. An optimized XGBoost-based Sequential Forward Selection (SFS) methodology was developed to identify a minimal set of informative spectral bands. Results demonstrate that while geometrical features alone achieved only 74% accuracy, the full spectral range reached a benchmark accuracy of 98.6%. The refined SFS methodology identified an optimal 9-band configuration (5 VNIR and 4 SWIR) that maintained 98.0% accuracy, representing a significant reduction in data complexity. Furthermore, a cost-effective 6-band solution (5 VNIR and 1 SWIR) achieved 96.4% accuracy. These findings were validated through a multispectral camera simulation, which showed a minimal performance drop of approximately 0.1%. the research showed that infected tomato seeds can be detected by VNIR and SWIR HSI. The proposed band-selection strategy enables the transition from expensive hyperspectral systems to affordable, high-throughput multispectral sensors for effective seed health management.

Pest dynamics arise from intricate interactions between climate and landscape, producing spatiotemporal patterns that often elude conventional parametric models. Reliable forecasting of these dynamics is critical for sustainable monitoring and management in intensive fruit-production regions. In this work, we present a hybrid warning system for codling moth (Cydia pomonella), combining machine learning with spatiotemporal geostatistics. The dataset comprised over 400 weekly observations (April-September, 2022–2024) collected from Catalonia’s principal fruit-producing area (Lleida, NE Spain). Predictors included climatic variables and GIS-derived landscape metrics, encompassing spatial configuration indices of host crops (apple, pear and walnut). Feature selection was optimized via recursive feature elimination (RFE). Global deterministic trends were captured using a Random Forest (RF) model, calibrated through 5-fold cross-validation on threshold-based binary capture data, while residual spatiotemporal autocorrelation was modeled using a sum-metric spatiotemporal variogram. Map accuracy was evaluated via leave-one-out cross-validation. RF model captured the global deterministic trend with high accuracy, identifying climatic variables as the primary drivers, whereas landscape metrics were not selected. Residuals exhibited spatiotemporal autocorrelation up to 15 km and 3 weeks, reflecting localized outbreak clustering. Integrating this structure enabled the generation of accurate maps of seasonal hotspots.

Agricultural mechanization in developing nations like Nigeria relies on effective farm machinery management. This study developed and validated a computer-based Decision Support System (DSS), FUOYEFMMS, to optimize tractor-implement matching and evaluate equipment performance and costs. Built with Python and Tkinter, the DSS integrates ensemble machine learning models using ASABE and Brixius equations to predict parameters such as fuel consumption, implement draft, and drawbar pull. A comprehensive machinery database was compiled from Nebraska Tractor Testing Laboratory (NTTL),

ASABE publications, and field data from ten farms in Southwest Nigeria. Ensemble learning methods, including Stacking with Random Forest, Extra Tree, and KNN regression, were applied. Extra Tree and Decision Tree regressors achieved the highest accuracy, with R² values of 0.9738 for drawbar pull and 0.8543 for implement draft. Validation involved three tractor models (MF 8680, MF 6499, Case-IH JX75) and three implements (disc plough, offset disc harrow, disc ridger) tested at depths of 100–300 mm across sandy, loamy, and clay soils. Predicted draft values (3.26–13.26 kN) closely matched field data, confirmed by a p-value of 0.31. With its user-friendly interface and deployable design, FUOYEFMMS offers a strategic tool for sustainable mechanized farming in sub-Saharan Africa.

HEALTHYTOMATO is an international project that aims to develop a greenhouse tomato disease risk warning and detection system for medium and small-sized tomato growers. The system consists of two main elements: i) a warning model that will assess the risk of disease spread in a greenhouse based on sensor data, weather forecasts, multispectral (MS) cameras, and manually entered information; ii) an early disease detection model based on the analysis of the Hyperspectral (HS) image dataset. During the project, tomato cultivar ‘Encore’ was inoculated with Alternaria protenta, which causes early blight, and Botrytis cinerea/pseudocinerea, which causes grey mold. The dynamics of disease development were recorded with an HS camera (spectral range 400 – 1000 nm) and assessed visually. During Alternaria protenta inoculation trials, a total of 1000 HS images were collected. LabelStudio was used for annotation; the leaves were categorized into three classes: i) control, ii) healthy, and iii) diseased regions of inoculated leaves. Preliminary principal component analysis (PCA), based on 348 HS images, showed that the largest differences between healthy and diseased regions were observed in the spectral ranges of chlorophyll absorption peaks, as well as in the infrared region, with the maxima at 687 and 702 nm.

The primary objective of this study is the development of a low-cost early warning system to monitor Black Sigatoka in banana plantations. Developed under the Banalytic project and supported by Innovate UK, the methodology involved the installation of sensor prototypes in Ghana and Brazil to collect essential climatic variables, focusing initially on temperature and relative humidity. These real-time environmental data were integrated with an extensive six-year historical monitoring database provided by UNESP. This combined dataset was used to train Artificial Neural Networks (ANN) aiming to predict disease outbreaks with higher precision. The system utilizes LoneWhisper satellite-IoT connectivity to ensure data transmission from remote tropical areas that lack conventional internet infrastructure. Preliminary results demonstrate that the synergy between low-cost hardware and machine learning models can optimize fungicide application schedules and reduce operational costs. By uniting accessible sensing technology and advanced computational intelligence, the Banalytic project provides a scalable and sustainable tool for phytosanitary management, increasing the climate resilience of banana producers in emerging markets.