Conference Agenda

Soil available phosphorus (AP) is crucial for crop growth and ecological stability. Conventional detection methods are often cumbersome and lack sufficient precision. This research proposed a high-precision AP detection method using multi-sensor data fusion of an electronic nose (E-nose) and Ultra Violate-Visible (UV-vis) spectroscopy based on soil pyrolysis gas. Firstly, a self-developed soil pyrolysis gas collection system integrated with an Arduino-based precision control module was constructed, with pyrolysis parameters optimized using Response Surface Methodology (RSM), identifying the optimal conditions as pyrolysis temperature of 800°C, pyrolysis time of 3 min 30 s, and soil weight of 4 g. Subsequently, the E-nose and UV-vis signals of the soil pyrolysis gas were collected. Olfactory and spectral features were extracted using Principal Component Analysis (PCA) and Competitive Adaptive Reweighted Sampling based on Savitzky-Golay (SG-CARS), respectively, and integrated through a self-attention-based feature-level fusion strategy. Finally, Partial Least Squares Regression (PLSR), Support Vector Regression (SVR), and Multi-Layer Perceptron (MLP) were established for the prediction and validation of AP. Results of PLSR demonstrated that the fusion feature achieved superior accuracy (R² = 0.89) compared to single feature spectral (R² = 0.83) or olfactory (R² = 0.85). This research provides an advanced technical solution for precise AP characterization.

Soil Water Content (SWC) is essential for irrigation scheduling and crop management in precision agriculture. Low-cost RGB cameras are field-deployable, but SWC estimates can vary strongly with illumination color temperature and intensity. Hyperspectral imaging (HSI) offers high spectral fidelity, yet its cost and operational complexity limit routine field use.

We propose an HSI-guided calibration framework that improves illumination robustness while keeping RGB-only operation in deployment. Sandy soil samples spanning silt contents of 0–40% and gravimetric water contents of 0–39% were imaged with an RGB camera and a VNIR hyperspectral sensor under halogen, LED, and mixed lighting. A physics-based RAW correction pipeline reduced inter-illumination color differences (CIEDE2000) by 74.1%, increasing optical consistency across lighting conditions.

HSI measurements were used as a spectral reference during development to calibrate the RGB feature space; the final estimator runs solely on RGB images. In independent RGB-only testing, performance remained stable under illumination shifts (R² ≈ 0.91, RMSE ≈ 2.1%) and systematic bias decreased by ~46% versus a conventional RGB-only model.

Overall, treating HSI as a development-stage reference—not a real-time requirement—provides an engineering pathway to low-cost, illumination-robust soil moisture monitoring for precision agriculture.

Soil quality assessment is fundamental for sustainable agricultural management but is complicated by the variability, complexity, and long-term dynamics of soil properties. This study developed a soil quality index using a data-driven class-modelling approach based on the Data-Driven Soft Independent Model of Class Analogy (DD-SIMCA). A reference model was created from a synthetic dataset representing typical ranges of nine physico-chemical parameters of agricultural soils. The model was then applied to a real dataset of approximately 9,800 soil samples. Classification performance was excellent (sensitivity = 1), enabling clear categorization into distinct soil quality classes. To enhance usability, results were converted into a colour-coded QR code, providing an intuitive visual summary of soil quality for practitioners. Further validation will use harmonized datasets from the Varda Foundation’s SoilHive platform, supporting scalability across regions and pedo-climatic contexts. Integration within SoilHive will ensure continuous model refinement as new data are contributed. Overall, the framework demonstrates the potential of machine-learning-based class modelling to support robust, scalable, and user-friendly soil quality assessment in agricultural systems.

Accurate soil moisture monitoring underpins water-efficient irrigation and plant health in agriculture, horticulture, and building-integrated greening systems (e.g., green facades and roofs).

Conventional capacitive probes often suffer from long-term drift due to changing soil contact (root growth, compaction) and sample only a small volume around the electrodes. To overcome these limitations, we present a non-contact sensing approach that infers volumetric water content from radio-frequency (RF) attenuation between antennas. Unlike traditional probes, our method integrates information over a larger sensing volume.

Custom-matched antennas were designed, characterized from 496 MHz to 2.4 GHz, and evaluated in controlled stepwise wetting experiments. Based on the observed penetration–sensitivity trade-off, our findings demonstrate that 2.4 GHz provides high sensitivity, while 868 MHz offers improved depth penetration.

The sensor is implemented on a Texas Instruments CC1352P system-on-chip, enabling low-power, IoT-ready wireless connectivity and remote data acquisition. The data reveals a strong, repeatable relationship between the Received Signal Strength Indicator and water content. Overall, the proposed dual-band RF sensor offers a scalable, low-maintenance alternative for long-term monitoring in biologically and mechanically dynamic substrates.

Water scarcity in small islands and remote agricultural regions has accelerated deployment of decentralized Reverse Osmosis (RO) desalination units powered by Renewable Energy Sources (RES), particularly Photovoltaic (PV) systems. Although direct PV coupling enhances sustainability and reduces fossil fuel dependency, it induces intermittent operation with fluctuating feed pressure (±15–25%) and frequent start–stop cycles. These conditions affect membrane compaction, fouling dynamics, and long-term permeability stability. This study evaluates the performance of small-scale PV-driven RO systems under realistic variable irradiance profiles, targeting potable and irrigation water production in island environments. Results demonstrate that variable pressure events can lead to water permeability decline of up to 12–18% compared to steady-state operation. However, implementation of control strategies such as Variable Frequency Drive (VFD)-based pressure smoothing and optimized intermittent flushing reduced irreversible water permeability losses by approximately 35% and stabilized normalized flux within ±5% of baseline values. Salt rejection remained above 98%, confirming product water suitability for irrigation purposes under moderate salinity conditions. The findings highlight that control-oriented system design significantly enhances membrane resilience and operational reliability under RE variability. The proposed approach supports climate-resilient, low-emission water supply systems tailored for island communities, strengthening the water–energy nexus within sustainable biosystems engineering.