ASHS 2025 Conference

Forecasting yield is a timely opportunity to make anticipated harvesting decisions on the grown crop and understand field variability. Such information is a remarkable contribution to the precision agriculture context. However, developing such an approach is challenging for perennial crops such as pecan. These crops present slight canopy changes, which often do not reflect the upcoming yield. Consequently, waiting for the harvest date is the only approach to obtain yield data. Conversely, the advent of image-based data and artificial intelligence techniques has proven their applicability in addressing this issue. Therefore, our objective was to analyze whether UAV multispectral images and AI-based data analysis are suitable for developing forecasting models for yield in pecan trees. Hence, we began collecting multispectral images approximately five months before the harvesting date. Each data collection date had an interval of fifteen days, totaling ten multispectral image sets. Subsequently, we processed the images to generate ten orthomosaics (one for each date). The orthomosaics were used to calculate numerous vegetation indices, texture data, and the canopy area to be used as inputs for the forecasting models. At the harvest date, we measured the yield of 78 individual plants across two pecan fields. Before developing the forecasting models, we performed a correlation analysis to better understand the relationship between the image data and yield. Afterward, we developed the forecasting models using machine learning algorithms, namely, multiple linear regression, decision tree, support vector machine, and random forest. The dataset was split into 70% (n = 55) for training and 30% (n = 23) for testing. The training dataset was used to train the forecasting models, while the testing dataset was used to assess the models’ effectiveness regarding precision (coefficient of determination, R²) and accuracy (mean absolute error, MAE; and root mean squared error, RMSE). All the models produced interesting results and could be implemented to forecast yield in pecan trees. However, random forest outperformed the others (high precision and accuracy) and, therefore, was the remaining model for this study. Forecasting yield in pecan trees presented increased effectiveness, improving the models’ performance early on and establishing higher accuracies closer to the harvesting date. We also performed a feature importance analysis, where predominantly the texture data contributed better to the models’ performance. Certainly, our findings are timely and support pecan growers and stakeholders in making better decisions for harvesting with anticipated and accurate yield data without waiting for the harvesting date.

Non-destructive estimation models often require cultivar-specific calibrations due to spectral differences arising from genetic variability. Integrating diverse cultivars into a single model can reduce costs and simplify data collection. However, in persimmons, the abundant and variable proanthocyanidins in the fruit overlap with spectral regions used for sugar estimation, rendering accurate prediction with a single model challenging. In this study, we attempted sugar estimation in diverse persimmon cultivars using near-infrared (NIR) spectroscopy and hyperspectral imaging. A total of 989 spectral measurements were acquired from 34 persimmon cultivars. Regression models employing various pre-processing and modeling techniques achieved a maximum R² of 0.786, indicating the feasibility of modeling sugar content across diverse cultivars with a unified approach. Furthermore, by designing a cover for the NIR sensor and combining it with SNV pre-processing, we demonstrated that stable spectra for sugar estimation can be obtained under outdoor conditions. With further improvements in accuracy, this approach is expected to facilitate rapid fruit quality evaluation and contribute to optimized production.

Combining green roofs with solar modules can protect plants and produce energy in cities. Growing crops in this system is called rooftop agrivoltaics (RAV) and can complement current urban agriculture efforts. We evaluated a group of five leafy green crops (arugula, kale, lettuce, spinach, and Swiss chard) under different solar modules over two years at two locations. Data measurements were taken for fresh and dry weight (FW, DW) stomatal conductance (SC), plant size at harvest (PSH), and microclimate data. Treatments included a polycrystalline opaque silicon module, a cadmium telluride (CdTe) frameless opaque module, a 40% semi-transparent CdTe module, and a full sun control. Four of the five leafy greens produced higher FW and DW under the 40% semi-transparent modules compared to other treatments and the full sun control, except spinach. Most species also produced larger PSH under the PV module treatments compared to the full sun control. Leafy greens under the module treatments resulted in lower SC, however, lettuce and Swiss chard grown under the semi-transparent module treatment produced higher SC compared to all other treatments. This research shows that incorporating photovoltaics on rooftop gardens influences the yield and stomatal conductance of select leafy green crops. While FW and DW mostly decreased under the deep shade treatments (opaque module, frameless module, and bifacial module) SC decreased, possibly due to less solar radiation on the leafy greens, reducing water use. Understanding the growth characteristics and growing environment of high value crops like leafy greens will increase understanding of what food crops are suitable for RAV systems.

Real-time monitoring of crop health is pivotal for advancing precision agriculture, enabling timely interventions to mitigate abiotic stress impacts. This study presents a novel and non-destructive approach for detecting salt stress in hydroponically grown arugula (Eruca sativa L.) using low-cost MQ gas sensors. Arugula seedlings, 11 days post-germination, were transplanted to a deep-water culture (DWC) hydroponic system in the greenhouse facility of The Pennsylvania State University. Salt stress was induced 9 days after planting by supplementing a modified Hoagland nutrient solution with sodium chloride (NaCl) at three concentrations: 0 mM (control), 40 mM, and 80 mM. Electrical conductivity (EC), pH and temperature parameters were regularly monitored during the cultivation period. Three MQ gas sensors—MQ2, MQ135, and MQ137—were integrated into a dome-shaped enclosure positioned over individual net pots, each containing four plants. To achieve a comprehensive volatile organic compound (VOC) profile, sensor units were strategically positioned on multiple plants within each treatment group. A total of 144 plants per treatment were cultivated, and two sets of sensor units recorded VOC emissions for 8 consecutive days. Salt stress significantly influenced plant growth, with fresh weight (FW) and leaf area decreasing as salinity increased. The 80 mM treatment exhibited the lowest FW and leaf area (61.69 ± 2.7 g, p-value = 0.015; 1434.25 ± 58 cm², p-value = 0.003), followed by the 40 mM treatment. All three MQ sensor responses revealed distinct VOC emission patterns correlating with salt stress levels. These sensor outputs were leveraged to train three machine learning models—K-Nearest Neighbors (KNN), Support Vector Machine (SVM), and Random Forest—to classify stress-induced VOC signatures. Among these, the cubic KNN model demonstrated superior predictive performance, achieving 98.73% accuracy, 98.74% precision, 98.73% recall, and an F1 score of 0.98 for the test dataset. These findings underscore the potential of cost-effective MQ gas sensors for real-time, non-invasive stress detection in crops, offering a promising tool for precision agriculture and early stress diagnosis. The integration of VOC sensing with machine learning models provides a scalable solution for enhancing crop management strategies in controlled environments.

Land use is a major concern for our society, which impacts governmental regulations, industry and agriculture, and individual property owners. Agrivoltaics (APV) includes the combination of agriculture and photovoltaics energy production on a single plot of land. APV has been proposed as a way to integrate agriculture into large-scale photovoltaic arrays or integrate energy production into established agricultural operations. The objectives of this project were to establish replicated APV research trials with fixed vertical panel photovoltaic arrays and investigate the feasibility of growing various warm- and cool-season crops between them. Warm-season crops included tomato, watermelon, bush bean, and zucchini. Cool-season crops included fall-planted lettuce and beets as well as spring-planted spinach and lettuce. Vertical panels were constructed with four replications that examined systems effects of APV compared to the open-field. A split-plot randomized complete block design was utilized, whereby main plots included system and sub-plots were the crops. For each crop, yield, marketability, quality, and economic data were collected. Crop quality parameters tested included: visual quality, color, firmness, titratable acidity, and total soluble solids. Additionally, PAR sensors were located within crop rows to characterize light availability at the replicated site as well as at another solar array. Sensors were placed approximately 2.5’ above the ground surface to generally mimic mature plant canopy height, transversely across the expected light-treatment area. From the first year of study, similarities among the two systems (APV and open-field) were more numerous than significant differences. Only one statistically-significant treatment effect was found on crop yield, among one of two lettuce varieties grown in the fall (P

Blueberries are extremely susceptible to drought due to their shallow root systems and limited water regulation capabilities. Climate change exacerbates drought stress in major blueberry production regions, which affect key physiological traits, such as leaf water content (LWC), photosynthesis (A), stomatal conductance (gs), electron transport rate (ETR), photosystem II efficiency (φPSII) and transpiration rate (E). Current phenotyping methods for measuring these physiological traits are time-consuming and labor-intensive as well as limited by the need for specialized equipment. To address this, a high-throughput phenotyping (HTPP) platform integrated with hyperspectral camera and a novel graph convolutional network (GCN)-based model, Plant-GCN, was developed to predict physiological traits of blueberry plants under drought stress. Spectral reflectance obtained from the hyperspectral images were transformed into a graph representation, with each plant represented as a node, spectral reflectance as node features, and edges defined by spectral similarities. The Plant-GCN model utilizes graph convolutional layers that aggregate information from neighboring nodes, effectively capturing complex interactions in the spectral signature and enhancing the prediction of physiological traits. Plant-GCN achieved a coefficient of determination (R²) of 0.89 for LWC, 0.94 for A, 0.89 for gs, 0.92 for ETR, 0.93 for φPSII and 0.89 for E on the test dataset. The performance of the proposed Plant-GCN model was compared with multilayer perceptron (MLP), partial least squares regression (PLSR), support vector regression (SVR), and random forest (RF), and it consistently outperformed all these models as well as data published in other reports. The high-throughput phenotyping system enabled efficient large-scale data collection, while the Plant-GCN model captured long-range spectral relationships significantly improved the prediction of physiological traits. The high predictability of the models could facilitate the screening of blue-berry cultivars for the specified traits allowing the selection and breeding of new drought tolerant cultivars in the future.

Bacterial leaf scorch is a harmful disease for pecan production, which can cause premature canopy defoliation, reduced kernel weight, and significant yield losses. The disease can cause a 10-13% reduction in shell nut weight and a 14-19% decrease in kernel weight, adversely affecting the quality and quantity of pecan yields. Early detection and precise management are important for minimizing economic losses and sustainable disease management. Current methods, such as manual scouting and conventional imaging, are inadequate for pecan orchards due to the trees' height and their inability to capture temporal changes or disease transmission patterns. Spatio-temporal modeling is a useful technique that enables tracking disease spread across time and location, identifying hotspots and transmission patterns to prioritize areas needing urgent intervention. This study aimed to develop a spatio-temporal model to visualize and evaluate the progression of bacterial leaf scorch disease in pecan orchards. High-resolution multispectral images of pecan trees were collected using a drone- mounted multispectral camera. A 3D point cloud was generated using Pix4D to create a reference NIR point cloud, and other vegetation indices point clouds were then aligned to provide sufficient data for model training. Ten vegetation indices, including the Normalized Difference Vegetation Index (NDVI), Green Normalized Difference Vegetation Index (GNDVI), Normalized Difference Red Edge Index (NDRE), Modified Red Edge Simple Ratio (MRESR), Enhanced Vegetation Index (EVI), Soil-Adjusted Vegetation Index (SAVI), Optimized Soil-Adjusted Vegetation Index (OSAVI), Atmospherically Resistant Vegetation Index (ARVI), Chlorophyll Index - Red Edge (CI_RE), and Simple Ratio Index (SR), were used to evaluate disease sensitivity. Individual tree canopies were segmented using the 3D DBSCAN algorithm for each index. A point transformer deep learning model was trained with 3D vegetation indices of each tree with the ground truth to develop the prediction ability of the model for classifying individual each tree disease severity level. The early results indicate that the model can classify disease severity in the 3D point clouds, capturing the disease stress in the individual tree. Additionally, integrating a temporal embedding layer into the trained model enables the transformer block to track changes in vegetation indices and quantify disease progression over time. The findings of this study facilitate the monitoring of disease progression and support site-specific management decisions, thereby enhancing the sustainability of pecan production.
Keywords: Artificial intelligence, Spatio-temporal modeling, Multispectral imaging, Spectral analysis, Drone-based Scouting.

Precise and accurate quantification of blackberry flowers are essential for yield forecasting, phenotypic assessment, and enhancing management techniques in precision agriculture. Threshold segmentation of images for blackberry feature quantification could be challenging due to shadows and background variability, while manual flower estimation is subjective and time-consuming. The objective of this research was to apply artificial intelligence and computer vision to identify and quantify blackberry flowers from Unmanned Aerial Vehicle (UAV) remote sensing. A computer vision algorithm You Only Look Once (YOLO) was trained with 1142 image datasets of blackberry flowers to develop an image processing workflow to quantify UAV captured images. A performance analysis was conducted with YOLO variants (YOLOv8s–YOLO12s) for quantifying blackberry flowers. YOLOv10s achieved the best performance with a mAP@0.5 of 58%, precision of 60%, and recall of 58%. Input resolution had a notable impact, performing better at 1024×1024 pixels (mAP@0.5 = 55%) than at 640×640 (mAP@0.5 = 30%). Increasing the training dataset from 250 to 1,142 images progressively improved detection accuracy, highlighting the value of data volume for model generalization. Additionally, flower counts predicted by YOLOv10s showed a strong correlation with flower-to-vegetation ratio (FVR; r = 0.71, p < .001), supporting FVR as a practical proxy for estimating floral density in the field. A contribution to computer assisted agriculture integration in the blackberry industry has been made by investigating the performance of computer vision algorithms on blackberry flower detection. Challenges such as small and uneven sized flowers, overlapping occlusion, and plot-wise analysis must be further investigated. Keywords. UAV, automation, blackberry, phenotyping, artificial intelligence.

A forum for discussion of potential collaborations with regards to technology in horticulture – i.e. biotechnology, UAVs, cameras, sensors, artificial intelligence, etc.

The Technology Applications in Horticulture Interest Group will meet at this session.