In this study we have evaluated three advanced water treatment technologies in laboratory conditions, electrochemical (EC), fluidized bed (FB) and nanocomposite-based systems. The performance of the three technologies were evaluated based on several characteristics, such as pollutant removal efficiency, operating cost (USD/m$^3$), specific energy consumption (kWh/kg), throughput (kg/h), space-time yield (STY, kg/m$^3$$\cdot$h) and energy utilization efficiency (kg/kWh). The results show that the nanocomposite system offers the best treatment efficiency (93.17% removal efficiency and very low variability (standard deviation = 0.78%), showing good stability and reliability of the process. We found that nanocomposite system had moderate operating cost of 0.109−0.116 USD/m$^3$ and specific energy consumption of 3.60−6.52 kWh/kg, with an average value of 4.70 kWh/kg. Also, it has the highest STY (0.94 kg/m$^3$$\cdot$h) and high energy utilization efficiency (0.2776 kg/kWh). In contrast, the FB system has the lowest average operating cost (0.1016 USD/m$^3$), lowest average specific energy consumption (4.20 kWh/kg) and the best energy utilization efficiency (0.2493 kg/kWh) and is the most economical option even with the lowest pollutant removal efficiency. The EC system provided the best removal efficiency (91.32%), but the highest operating cost (0.1242 USD/m$^3$) and energy consumption (6.50 kWh/kg) of the other technologies. In analysis of variance (ANOVA) and Tukey’s Honestly Significant Difference (HSD) tests, there was significant difference between all the technologies ($p$ $<$ 0.05). The nanocomposite system achieved 5.39% removal efficiency and the FB system was able to have better energy utilization than the EC and nanocomposite technology. In general, the nanocomposite technology was the best in terms of treatment efficiency, energy efficiency, and operational cost optimization and the FB system is the best choice for large-scale applications.

University students, as a key youth consumer demographic, will play a vital role in shaping sustainable purchasing behavior in the future. This study aims to uncover the factors influencing students’ intention to minimize food waste at universities using the extended norm activation theory. An online survey of 664 students examined intentions to reduce food waste on campus. Of these, 245 students used online food delivery (OFD), while 419 engaged in in-canteen dining (IC). To evaluate the empirical data, this study utilized a partial least squares structural equation model and executed measurement invariance testing within the composite model. The empirical results demonstrate that the activation of personal norms is driven by awareness of consequence and the ascription of responsibility, which consequently has a direct impact on the intention to reduce food waste. Personal norms also indirectly influence the intention to minimize food waste. Students who purchased meals only reported weaker personal norms and lower intention to reduce food waste than those who ate in the canteen. However, the OFD group showed greater awareness of consequence, which supported their efforts to reduce food waste, compared with the IC group. Overall, this study provides further insight into the psychological mechanisms underlying sustainable food consumption among university students.

The expansion of the shipbuilding industry in coastal areas contributes substantially to economic development while simultaneously posing significant risks to air quality and community health. This study analyzed the concentration and spatial distribution of air pollutants and assessed non-carcinogenic health risks in the shipyard industrial area of Batam City. Air quality measurements were conducted for PM$_{2.5}$, SO$_2$, NO$_2$, CO, and Pb parameters at multiple receptor points at different distances from the emission source. The field measurement data were then integrated with dispersion modeling using the American Meteorological Society/Environmental Protection Agency Regulatory Model (AERMOD) based on local meteorological conditions. Health risk was evaluated using the Hazard Quotient (HQ) approach with reference to national air quality standards. Pollutant concentrations decreased consistently with increasing distance from emission sources, with PM$_{2.5}$ exhibiting the widest and most persistent spatial distribution. Although most pollutant concentrations remained below regulatory thresholds, PM$_{2.5}$ yielded HQ values exceeding 1.0 across all receptor distances up to 2000 m, indicating significant non-carcinogenic health risk at all observed distances. Model validation demonstrated strong spatial agreement between measured and simulated concentrations ($R^2$ $>$ 0.84), with a consistent tendency toward underestimation of absolute values. The integration of spatial dispersion modeling with health risk assessment offers a comprehensive analytical framework for air quality management and public health protection in coastal industrial settings.

An Environmental, Social, and Governance (ESG) report is an essential information source for evaluating a company’s performance in sustainability practices. Organizations structure their environmental impacts, social responsibilities, and governance practices within a defined framework. This standardization is provided by the Global Reporting Initiative (GRI), which constitutes an internationally recognized guideline for sustainability reporting. Traditional reporting workflows are time-consuming for organizations and prone to data-entry errors, which limits the reliability of disclosed information. In this context, leveraging the capabilities of Large Language Models (LLMs) offers significant time and resource savings. This study uses the Llama-3.1-8B-Instruct model under two scenarios, Retrieval-Augmented Generation (RAG) and Low-Rank Adaptation (LoRA) fine-tuning, to analyze 30 food-sector ESG reports and produce ESG summaries, SWOT analyses, and GRI-aligned recommendations. The two approaches are evaluated on a stratified hold-out set of 6 unseen test reports (24 reports used for training) under a fair, matched-budget setup in which RAG retrieves the target report at inference. On four quality metrics, LoRA achieved higher mean scores than RAG; however, statistically significant differences were observed in only 4 of the 12 task–metric comparisons. Token usage was comparable, whereas RAG was substantially faster at inference. Rather than favoring one approach over the other, these findings reveal a trade-off between output quality and computational efficiency: LoRA yields quality gains on specific metrics, whereas RAG is substantially more efficient at inference. Given the limited size of the held-out test set, these results should be interpreted with caution.