NILU Brage
Not a member yet
3619 research outputs found
Sort by
Investigating lightweight and interpretable machine learning models for efficient and explainable stress detection
No full textStress is a common human reaction to demanding circumstances, and prolonged and excessive stress can have detrimental effects on both mental and physical health. Heart rate variability (HRV) is widely used as a measure of stress due to its ability to capture variations in the time intervals between heartbeats. However, achieving high accuracy in stress detection through machine learning (ML), using a reduced set of statistical features extracted from HRV, remains a significant challenge. In this study, we aim to address these challenges by proposing lightweight ML models that can effectively detect stress using minimal HRV features and are computationally efficient enough for IoT deployment. We have developed ML models incorporating efficient feature selection techniques and hyper-parameter tuning. The publicly available SWELL-KW dataset has been utilized for evaluating the performance of our models. Our results demonstrate that lightweight models such as k-NN and Decision Tree can achieve competitive accuracy while ensuring lower computational demands, making them ideal for real-time applications. Promisingly, among the developed models, the k-nearest neighbors (k-NN) algorithm has emerged as the best-performing model, achieving an accuracy score of 99.3% using only three selected features. To confirm real-world deployability, we benchmarked the best model on an 8 GB NVIDIA Jetson Orin Nano edge device, where it retained 99.26% accuracy and completed training in 31 s. Furthermore, our study has incorporated local interpretable model-agnostic explanations to provide comprehensive insights into the predictions made by the k-NN-based architecture.publishedVersio
Non-target and suspect screening of volatile organic compounds from Scots pine and Norway spruce building materials
No full textWood building materials can be a source of volatile organic compounds (VOCs) in the indoor environment and increasing focus is put on classification and regulation of the use of wood building materials in Europe. The main wood related VOCs such as monoterpenes rarely pose adverse health effects for humans, but as analytical procedures become more sensitive new hazardous VOCs are detected in low concentration. There is a need for comprehensive identification of VOCs emitting from different wood building materials for indoor use. This study performed a first semi-quantitative non-target and suspect screening of VOC emissions from three important wood-based building materials in Europe. Air samples collected from emission chambers were analyzed using comprehensive two-dimensional gas chromatography coupled to time-of-flight mass spectrometry and resulting mass spectra were classified into confidence groups. A total of 84, 133 and 197 compounds were found to emit from cross-laminated timber, untreated spruce panel and untreated pine panel, respectively. Pine panel was found to emit a higher number of VOCs as well as higher concentrations of most VOCs compared to the spruce building materials. Several new VOCs were detected in the emission profile of pine and spruce. However, they were mostly structurally similar to previously reported wood VOCs. Two compounds of concern emitting from all three wood building materials were furfural and (E)-2-octenal, as these have been classified as group 2 carcinogen and potent eye irritant, respectively.publishedVersio
Multi-year black carbon observations and modeling close to the largest gas flaring and wildfire regions in the Western Siberian Arctic
No full textThe influence of aerosols on the Arctic system remains associated with significant uncertainties, particularly concerning black carbon (BC). The polar aerosol station “Island Bely” (IBS), located in the Western Siberian Arctic, was established to enhance aerosol monitoring. Continuous measurements from 2019 to 2022 revealed the long-term effects of light-absorbing carbon. During the cold period, the annual average light-absorption coefficient was 0.7 ± 0.7 Mm−1, decreasing by 2–3 times during the warm period. The interannual mean showed a peak in February (0.9 ± 0.8 Mm−1) then 10 times the lower minimum in June and exhibited high variability in August (0.7 ± 2.2 Mm−1). An increase of up to 1.5 at shorter wavelengths from April to September suggests contribution from brown carbon (BrC). The annual mean equivalent black carbon (eBC) demonstrated considerable interannual variability, with the lowest in 2020 (24 ± 29 ng m−3). Significant difference was observed between Arctic haze and Siberian wildfire periods, with record-high pollution levels in February 2022 (110 ± 70 ng m−3) and August 2021 (83 ± 249 ng m−3). Anthropogenic BC contributed 83 % to the total for the entire study period, and gas flaring, domestic combustion, transportation, and industrial emissions dominated. During the cold season, > 90 % of surface BC was attributed to anthropogenic sources, mainly gas flaring. In contrast, during the warm period, Siberian wildfires contributed to BC concentrations by 48 %. In August 2021, intense smoke from Yakutian wildfires was transported at high altitudes during the region's worst fire season in 40 years.publishedVersio
Indian Land Carbon Sink Estimated from Surface and GOSAT Observations
Full text linkThe carbon sink over land plays a key role in the mitigation of climate change by removing carbon dioxide (CO2) from the atmosphere. Accurately assessing the land sink capacity across regions should contribute to better future climate projections and help guide the mitigation of global emissions towards the Paris Agreement. This study estimates terrestrial CO2 fluxes over India using a high-resolution global inverse model that assimilates surface observations from the global observation network and the Indian subcontinent, airborne sampling from Brazil, and data from the Greenhouse gas Observing SATellite (GOSAT) satellite. The inverse model optimizes terrestrial biosphere fluxes and ocean-atmosphere CO2 exchanges independently, and it obtains CO2 fluxes over large land and ocean regions that are comparable to a multi-model estimate from a previous model intercomparison study. The sensitivity of optimized fluxes to the weights of the GOSAT satellite data and regional surface station data in the inverse calculations is also examined. It was found that the carbon sink over the South Asian region is reduced when the weight of the GOSAT data is reduced along with a stricter data filtering. Over India, our result shows a carbon sink of 0.040 ± 0.133 PgC yr−1 using both GOSAT and global surface data, while the sink increases to 0.147 ± 0.094 PgC yr−1 by adding data from the Indian subcontinent. This demonstrates that surface observations from the Indian subcontinent provide a significant additional constraint on the flux estimates, suggesting an increased sink over the region. Thus, this study highlights the importance of Indian sub-continental measurements in estimating the terrestrial CO2 fluxes over India. Additionally, the findings suggest that obtaining robust estimates solely using the GOSAT satellite data could be challenging since the GOSAT satellite data yield significantly varies over seasons, particularly with increased rain and cloud frequency.publishedVersio
Investigating the impact of climate change on PCB-153 exposure in Arctic seabirds with the nested exposure model
No full textAt the same time Arctic ecosystems experiences rapid climate change, at a rate four times faster than the global average, they remain burdened by long-range transported pollution, notably with legacy polychlorinated biphenyls (PCBs). The present study investigates the potential impact of climate change on seabird exposure to PCB-153 using the established Nested Exposure Model (NEM), here expanded with three seabird species, i.e. common eider (Somateria mollissima), black-legged kittiwake (Rissa tridactyla) and glaucous gull (Larus hyperboreus), as well as the filter feeder blue mussel (Mytulis edulis). The model's performance was evaluated using empirical time trends of the seabird species in Kongsfjorden, Svalbard, and using tissue concentrations from filter feeders along the northern Norwegian coast. NEM successfully replicated empirical PCB-153 concentrations, confirming its ability to simulate PCB-153 bioaccumulation in the studied seabird species within an order of magnitude. Based on global PCB-153 emission estimates, simulations run until the year 2100 predicted seabird blood concentrations 99% lower than in year 2000. Model scenarios with climate change-induced altered dietary composition and lipid dynamics showed to have minimal impact on future PCB-153 exposure, compared to temporal changes in primary emissions of PCB-153. The present study suggests the potential of mechanistic modelling in assessing POP exposure in Arctic seabirds within a multiple stressor context.publishedVersio
Unprecedented shifts in aerosol pollution sources in China under a decade of clean air actions
No full textChina is a major hotspot of black carbon (BC) emissions, contributing to climate warming and risk to public health. Here, our dual-isotope-constrained observations indicate stringent air pollution controls have drastically reduced coal-burning in North China over the past decade, marking a transition to a “post-coal” era compared to earlier 2012–2014. However, biomass-burning fraction (fbb) for north/central/east winter hazes has doubled from earlier (north/east) ~20%, with significantly higher fbb during polluted winters. Comparisons between observation and transport modelling show good alignment in BC concentrations but substantial discrepancies in source attribution (i.e., fbb). Leveraging radiocarbon measurements, advanced atmospheric modelling, and a Bayesian approach, our study identifies biases stemming from misallocated residential fuel types in emission inventories. These findings underscore the untapped potential to mitigate BC emissions by targeting rural biomass burning, while providing critical insights into BC source evolution to refine emission inventories and formulate effective air quality policies for China and other nations facing severe air pollution.publishedVersio
Stress management with HRV following AI, semantic ontology, genetic algorithm and tree explainer
Full text linkHeart Rate Variability (HRV) serves as a vital marker of stress levels, with lower HRV indicating higher stress. It measures the variation in the time between heartbeats and offers insights into health. Artificial intelligence (AI) research aims to use HRV data for accurate stress level classification, aiding early detection and well-being approaches. This study’s objective is to create a semantic model of HRV features in a knowledge graph and develop an accurate, reliable, explainable, and ethical AI model for predictive HRV analysis. The SWELL-KW dataset, containing labeled HRV data for stress conditions, is examined. Various techniques like feature selection and dimensionality reduction are explored to improve classification accuracy while minimizing bias. Different machine learning (ML) algorithms, including traditional and ensemble methods, are employed for analyzing both imbalanced and balanced HRV datasets. To address imbalances, various data formats and oversampling techniques such as SMOTE and ADASYN are experimented with. Additionally, a Tree-Explainer, specifically SHAP, is used to interpret and explain the models’ classifications. The combination of genetic algorithm-based feature selection and classification using a Random Forest Classifier yields effective results for both imbalanced and balanced datasets, especially in analyzing non-linear HRV features. These optimized features play a crucial role in developing a stress management system within a Semantic framework. Introducing domain ontology enhances data representation and knowledge acquisition. The consistency and reliability of the Ontology model are assessed using Hermit reasoners, with reasoning time as a performance measure. HRV serves as a significant indicator of stress, offering insights into its correlation with mental well-being. While HRV is non-invasive, its interpretation must integrate other stress assessments for a holistic understanding of an individual’s stress response. Monitoring HRV can help evaluate stress management strategies and interventions, aiding individuals in maintaining well-being.publishedVersio
Transformation Product Formation and Removal Efficiency of Emerging Pollutants by Three-Dimensional Ceramic Carbon Foam-Supported Electrochemical Oxidation
Full text linkThis study evaluated galvanostatic three-dimensional electrolysis using ceramic carbon foam anodes for the removal of emerging pollutants from wastewater and assessed transformation product formation. Five pollutants (paracetamol, triclosan, bisphenol A, caffeine, and diclofenac) were selected based on their detection in wastewater treatment plant effluents. Electrochemical oxidation was carried out on artificial wastewater spiked with these compounds under galvanostatic conditions (50, 125, and 250 mA) using a stainless steel tube electrolyzer with three ceramic carbon foam anodes and a stainless steel cathode. Decreasing pollutant concentrations were observed in all of the experiments. Nontarget chemical analysis using liquid chromatography coupled to a high-resolution mass spectrometer detected 338 features with increasing intensity including 12 confirmed transformation products (TPs). Real wastewater effluent spiked with the pollutants was then electrolyzed, again showing pollutant removal, with 9 of the 12 previously identified TPs present and increasing. Two TPs (benzamide and 2,4-dichlorophenol) are known toxicants, indicating the formation of a potential toxic by-product during electrolysis. Furthermore, electrolysis of unspiked real wastewater revealed the removal of five pharmaceuticals and a drug metabolite. While demonstrating electrolysis’ ability to degrade pollutants in wastewater, the study underscores the need to investigate transformation product formation and toxicity implications of the electrolysis process.publishedVersio
Microplastic pellets in Arctic marine sediments: a common source or a common process?
Full text linkPlastic consumption is increasing, and millions of tonnes of plastic are released into the oceans every year. Plastic materials are accumulating in the marine environment, especially on the seafloor. The Arctic is contaminated with plastics, including microplastics (MPs, <5 mm) but occurrences, concentrations and fate are largely unknown. This study aimed at assessing whether MPs accumulate at greater water depths in the Barents Sea, and close to the Longyearbyen settlement, and at understanding the ubiquity and source of a specific type of collected pellets. Surface sediments were collected at seven stations around Svalbard with a box-corer, and three replicates were taken at each station. MPs were extracted through density separation with saturated saltwater. Many pellets were found, and their composition was assessed by pyrolysis-GC/MS. Procedural blanks were performed using field blanks as samples to assess the overall contamination. The composition of all extracted particles was then analysed by μRaman spectroscopy. On average, 3.61 ± 1.45 MPs/100 g (dw) were found. The sea ice station, after blank correction, was more contaminated and displaying a different profile than the other stations, and the deepest station did not show the highest MP concentrations but rather the opposite. Sediments close to Longyearbyen were not more contaminated than the other stations either. Dark pellets of similar aspect were found at all stations, raising the question about a possible common source or process. These pellets were made of several plastic polymers which varied in proportion for each pellet, suggesting a common process was at the origin of those pellets, potentially marine snow formationpublishedVersio