Rapid urbanisation and shrinking arable land have necessitated the exploration of alternative agricultural paradigms to ensure food security under space and resource constraints. Conventional ground-based agriculture is challenged by population growth and climate uncertainty, accelerating the need for alternatives, including controlled cultivation technologies. Hence, vertical farming and Internet of things (IOT)-enabled Artificial Intelligence (AI) approaches have emerged as viable and sustainable solutions. The study aims to create a roadmap for next-generation vertical farming systems and bridge the gap between technology and policies by providing a technical blueprint that supports scalable and sustainable agricultural transformation. The pillars around which the paper is structured are (i) Infrastructure & Systems: A detailed analysis of hydroponic, aeroponic, aquaponic, and hybrid technologies, analysing their operational parameters, environmental requirements, and sustainability metrics (ii) Digital Integration using IoT and AI: An architectural breakdown of multilayered IoT frameworks, integrating sensor networks, robotics, and automation with advanced Machine Learning, Deep Learning, and Digital Twin technologies for precise environmental control (iii) policy and techno-economic frameworks: A comparative analysis of the UK and Indian landscapes, assessing how their respective regulatory frameworks, energy infrastructures, and market structures can be harmonised to support large-scale adoption. A comparative techno-economic and policy analysis of vertical farming adoption in the United Kingdom and India is presented, considering climatic conditions, energy infrastructure, market structures, and regulatory frameworks. It also integrates UK’s high-tech agricultural innovation with India’s massive scale and policy frameworks to develop an adaptable, sustainable solution. The paper also proposes a next-generation cyber–physical vertical Farming architectureand outlines the policy roadmaps that must be rolled out to support scalable, resilient, and climate-adaptive agricultural systems. This review follows the PRISMA framework, with 246 studies included after systematic screening. The significance of the proposed framework is that it offers actionable and impactful guidance for vertical farming design, digital integration, and policy alignment that supports resilient and climate-adaptive food systems
In protected island ecosystems, invasion risk may be associated with disturbance history, anthropogenic influences, propagule pressure, and local habitat conditions. This study assessed the spatial pattern and environmental correlates of current invasion risk for the invasive alien tree Triadica sebifera (Euphorbiaceae) on Miyajima Island, Japan, using field-based presence–absence data and predictors derived from satellite imagery, topography, disturbance history, and anthropogenic factors. A classification random forest model was evaluated using nested five-fold spatial cross-validation, providing spatially explicit validation in a protected island ecosystem. Under the 500-m configuration, pooled out-of-fold predictions achieved an AUC of 0.977 and an accuracy of 0.922. Predicted invasion risk was highest along the coast and in anthropogenically modified areas in northeastern Miyajima Island, whereas the mountainous forest interior, including much of the area affected by the 1984 wildfire, showed relatively low predicted risk. Cross-validated permutation importance identified distance to anthropogenic patches as the most important predictor, followed by autumn and spring normalized difference moisture index (NDMI) and the summer solstice illumination model; fire severity and tourist-facility count had little predictive importance. The current observed invasion risk was therefore more strongly associated with proximity to anthropogenic patches, vegetation moisture, canopy structure, and light availability than with mapped historical fire severity. These findings suggest that current invasion risk reflects interactions between environmental opportunities and local habitat conditions associated with anthropogenic edges and coastal environments. This spatially validated framework provides a practical basis for prioritizing early detection, monitoring, and management of invasive plants in protected island forest ecosystems.
Converting unmanned aerial vehicle (UAV) multispectral imagery and ground measurements into field-scale trait estimates and management maps requires several linked processing steps. We developed AgriMapper, a WebUI that integrates geospatial data preparation, spectral-feature calculation and selection, regression modelling, pixel-wise estimation, and semi-automated prescription-map generation. A winter-wheat case study used 240 sampling records from 16 plots observed at three growth stages to estimate plant nitrogen concentration (PNC). Among 15 models evaluated using common shuffled five-fold partitions, AutoGluon achieved the highest mean coefficient of determination (R2; 0.79 ± 0.09), with a root mean square error (RMSE) of 2.00 ± 0.31 g kg−1. Values are means ± sample standard deviations across validation folds. Tiled estimation reduced memory requirements in the tested configurations. The generated prescription maps illustrate a pathway from trait estimation to agronomist-reviewed management, but their agronomic and environmental benefits require field validation. The project repository is https://github.com/George-Horus/AgriMapper_by_ZYH.
Federated Learning (FL) enables geographically dispersed edge devices to jointly train a shared model without transmitting raw data, which is essential in precision agriculture, where bandwidth constraints, sensor heterogeneity, and data sovereignty preclude centralized processing. Yet FL remains vulnerable to malicious or faulty clients injecting corrupted data, a threat amplified in open-field deployments where nodes can be physically compromised or adversarially controlled. We address this problem in the context of Kenyan Arabica coffee leaf disease detection, using the JMuBEN2 dataset. We propose WaSA (Wavelet Scattering Autoencoder), a fully unsupervised, proactive framework that screens clients before training begins. Each client transmits a single spectral embedding, constructed via truncated Singular Value Decomposition and a first-order Wavelet Scattering Transform, to the server once at initialization; a denoising autoencoder trained on clean auxiliary data then scores each client by reconstruction error. WaSA requires no attack labels, generalizes to unknown threat types, and adds negligible communication overhead per client. Under 90% client corruption, WaSA recovers Oracle-level accuracy (up to 100.00%), outperforming all baselines by over 17 percentage points in a regime where every existing defense collapses.
