Pub Date : 2026-04-01Epub Date: 2026-01-29DOI: 10.1016/j.coldregions.2026.104851
Pravin Punde , Yngve Birkelund , Muhammad Shakeel Virk , Xingbo Han , Trude Eidhammer
This study investigates in-cloud ground icing over Fagernesfjellet, a complex terrain site in northern Norway, using new kilometer-scale simulations from the Weather Research and Forecasting model (WRF) during the year 2022–2023. The WRF-derived icing results and meteorological parameters relevant to in-cloud icing are validated using measurement data (MEAS), with focus on model resolution and terrain height. Our findings indicate that WRF effectively represented the temporal evolution of icing events, with higher altitudes indicating more severe icing and an increased number of annual icing hours. However, in the current model setup, WRF underestimates the magnitude and variability of icing loads; an improvement in icing load amount is found when accurate terrain height point is used. The highest spatial resolution improved the simulation of key meteorological parameters, such as temperature and wind speed, but struggled with relative humidity and wind direction. Our study shows that high-resolution simulation and accurate terrain height are essential for improving atmospheric in-cloud ground icing simulations over complex terrain.
{"title":"Kilometer-scale simulation of atmospheric in-cloud ground icing over complex terrain in Norwegian Arctic","authors":"Pravin Punde , Yngve Birkelund , Muhammad Shakeel Virk , Xingbo Han , Trude Eidhammer","doi":"10.1016/j.coldregions.2026.104851","DOIUrl":"10.1016/j.coldregions.2026.104851","url":null,"abstract":"<div><div>This study investigates in-cloud ground icing over Fagernesfjellet, a complex terrain site in northern Norway, using new kilometer-scale simulations from the Weather Research and Forecasting model (WRF) during the year 2022–2023. The WRF-derived icing results and meteorological parameters relevant to in-cloud icing are validated using measurement data (MEAS), with focus on model resolution and terrain height. Our findings indicate that WRF effectively represented the temporal evolution of icing events, with higher altitudes indicating more severe icing and an increased number of annual icing hours. However, in the current model setup, WRF underestimates the magnitude and variability of icing loads; an improvement in icing load amount is found when accurate terrain height point is used. The highest spatial resolution improved the simulation of key meteorological parameters, such as temperature and wind speed, but struggled with relative humidity and wind direction. Our study shows that high-resolution simulation and accurate terrain height are essential for improving atmospheric in-cloud ground icing simulations over complex terrain.</div></div>","PeriodicalId":10522,"journal":{"name":"Cold Regions Science and Technology","volume":"245 ","pages":"Article 104851"},"PeriodicalIF":3.8,"publicationDate":"2026-04-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146184768","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-04-01Epub Date: 2026-01-19DOI: 10.1016/j.coldregions.2026.104832
Yong Wook Lee , Mingjian Wu , Tae J. Kwon
Adverse winter weather significantly compromises driving safety and mobility in regions such as Canada and the northern United States. This study addresses these challenges by utilizing stationary Road Weather Information Systems (RWIS) equipped with cameras. These images capture complex scenes, making automated road surface condition (RSC) classification systems particularly challenging. Unlike previous studies that required manual cropping of main road pavement, we applied convolutional neural networks (CNNs) directly to full stationary RWIS imagery to validate their effectiveness and generalizability for real-world winter road maintenance (WRM) applications. Our study focused on four key aspects: (1) rigorously validating CNN performance on stationary RWIS images without manual cropping, (2) systematically analyzing the influence of camera angles using explainable artificial intelligence (XAI) techniques, (3) evaluating the effect of image resolution on model accuracy, and (4) exploring data-quantity trade-offs, including the impact of adding or removing camera feeds, to develop robust and deployable CNN models. The developed CNN achieved excellent performance metrics, all exceeding 98%. Our findings indicate that optimizing camera orientation substantially enhances the model's focus on relevant features that align with human interpretation. Reducing background complexity and increasing road captures from different perspectives further enhanced model focus. Furthermore, increasing image resolution up to 224 × 224 improved performance, although gains were marginal beyond this point while computational costs rose substantially. This comprehensive evaluation demonstrates the high potential of using stationary RWIS imagery for RSC classification with CNNs, suggesting significant improvements in WRM efficiency and traffic safety during winter.
{"title":"Integrating convolutional neural networks and explainable AI for enhanced winter road surface conditions classification using stationary RWIS imagery","authors":"Yong Wook Lee , Mingjian Wu , Tae J. Kwon","doi":"10.1016/j.coldregions.2026.104832","DOIUrl":"10.1016/j.coldregions.2026.104832","url":null,"abstract":"<div><div>Adverse winter weather significantly compromises driving safety and mobility in regions such as Canada and the northern United States. This study addresses these challenges by utilizing stationary Road Weather Information Systems (RWIS) equipped with cameras. These images capture complex scenes, making automated road surface condition (RSC) classification systems particularly challenging. Unlike previous studies that required manual cropping of main road pavement, we applied convolutional neural networks (CNNs) directly to full stationary RWIS imagery to validate their effectiveness and generalizability for real-world winter road maintenance (WRM) applications. Our study focused on four key aspects: (1) rigorously validating CNN performance on stationary RWIS images without manual cropping, (2) systematically analyzing the influence of camera angles using explainable artificial intelligence (XAI) techniques, (3) evaluating the effect of image resolution on model accuracy, and (4) exploring data-quantity trade-offs, including the impact of adding or removing camera feeds, to develop robust and deployable CNN models. The developed CNN achieved excellent performance metrics, all exceeding 98%. Our findings indicate that optimizing camera orientation substantially enhances the model's focus on relevant features that align with human interpretation. Reducing background complexity and increasing road captures from different perspectives further enhanced model focus. Furthermore, increasing image resolution up to 224 × 224 improved performance, although gains were marginal beyond this point while computational costs rose substantially. This comprehensive evaluation demonstrates the high potential of using stationary RWIS imagery for RSC classification with CNNs, suggesting significant improvements in WRM efficiency and traffic safety during winter.</div></div>","PeriodicalId":10522,"journal":{"name":"Cold Regions Science and Technology","volume":"245 ","pages":"Article 104832"},"PeriodicalIF":3.8,"publicationDate":"2026-04-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146036118","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-04-01Epub Date: 2026-01-08DOI: 10.1016/j.coldregions.2026.104824
Motoki Sakaguchi , Kanta Ishida , Koichiro Tanaka , Moeka Tsukamoto , Chao Kang , Yu Kurokawa
Ice accretion on solid surfaces at low temperatures causes serious problems in numerous engineering applications, such as aircraft, wind turbines, and power lines. To develop effective anti-icing and de-icing technologies, understanding the mechanism of residual stress evolution within the ice film during the accretion process is essential. In this study, experiments were conducted to clarify the residual stress evolution during the solidification of a single water droplet dropped onto a low-temperature stainless steel substrate. In-situ observations of the droplet impact, spreading, and solidification were performed, varying the substrate temperature as a key parameter. Simultaneously, strain was measured on the substrate's backside using a strain gauge, showing that tensile strain develops on the substrate as the droplet solidifies. Furthermore, the residual strain increased with the decreasing substrate temperature. Cracking occurred immediately at lower substrate temperatures, whereas higher temperatures required additional cooling for cracking. Numerical analysis reproduced these experimental observations, quantifying the stress evolution within the ice film. Temperature-dependent material properties were incorporated in the analysis, as well as a creep constitutive equation based on stress relaxation tests, thereby capturing the time-dependent residual strain on the substrate and elucidating the distribution and evolution of tensile residual stress within the ice film. Ultimately, tensile stress developed during solidification and increased with cooling, reaching approximately 7–8 MPa at the time of cracking. These findings provide a fundamental understanding of the thermo-mechanical processes during ice accretion, crucial for developing robust anti-icing and de-icing strategies.
{"title":"Residual stress evolution during ice accretion from a single water droplet","authors":"Motoki Sakaguchi , Kanta Ishida , Koichiro Tanaka , Moeka Tsukamoto , Chao Kang , Yu Kurokawa","doi":"10.1016/j.coldregions.2026.104824","DOIUrl":"10.1016/j.coldregions.2026.104824","url":null,"abstract":"<div><div>Ice accretion on solid surfaces at low temperatures causes serious problems in numerous engineering applications, such as aircraft, wind turbines, and power lines. To develop effective anti-icing and de-icing technologies, understanding the mechanism of residual stress evolution within the ice film during the accretion process is essential. In this study, experiments were conducted to clarify the residual stress evolution during the solidification of a single water droplet dropped onto a low-temperature stainless steel substrate. In-situ observations of the droplet impact, spreading, and solidification were performed, varying the substrate temperature as a key parameter. Simultaneously, strain was measured on the substrate's backside using a strain gauge, showing that tensile strain develops on the substrate as the droplet solidifies. Furthermore, the residual strain increased with the decreasing substrate temperature. Cracking occurred immediately at lower substrate temperatures, whereas higher temperatures required additional cooling for cracking. Numerical analysis reproduced these experimental observations, quantifying the stress evolution within the ice film. Temperature-dependent material properties were incorporated in the analysis, as well as a creep constitutive equation based on stress relaxation tests, thereby capturing the time-dependent residual strain on the substrate and elucidating the distribution and evolution of tensile residual stress within the ice film. Ultimately, tensile stress developed during solidification and increased with cooling, reaching approximately 7–8 MPa at the time of cracking. These findings provide a fundamental understanding of the thermo-mechanical processes during ice accretion, crucial for developing robust anti-icing and de-icing strategies.</div></div>","PeriodicalId":10522,"journal":{"name":"Cold Regions Science and Technology","volume":"245 ","pages":"Article 104824"},"PeriodicalIF":3.8,"publicationDate":"2026-04-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146036533","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-04-01Epub Date: 2026-02-08DOI: 10.1016/j.coldregions.2026.104857
Xiyue Wang , Wanjun Ye , Qianqian Ma , Hao Yang , Minghao Zhang
In seasonal frozen regions, loess is susceptible to mechanical damage under the action of freeze-thaw (F-T) cycles and moisture content changes. Fiber reinforcement is recognized as an effective technique for enhancing its engineering performance. As a critical index for assessing fiber-reinforced loess, shear strength is governed by multiple interactive factors, including the number of F-T cycles, fiber content, and confining pressure, rendering its accurate prediction challenging. To address this challenge, a series of triaxial shear tests on fiber-reinforced loess subjected to wet-dry freeze-thaw (WDFT) cycles was performed. The influence mechanisms of cycle number, fiber content, and confining pressure on shear strength indicators were revealed through Spearman correlation analysis. On this basis, to construct a high-precision prediction model, multiple algorithms including support vector regression (SVR), random forest (RF), and generalized additive model (GAM) were compared. Extreme gradient boosting (XGBoost) was finally selected as the benchmark model. Subsequently, the pelican optimization algorithm (POA) was introduced for automatic hyperparameter optimization of the model. A hybrid machine learning model (POA-XGBoost) was thus proposed. The results indicate that the POA-XGBoost model achieved superior predictive performance. On the test set, the model attained a coefficient of determination (R2) of 0.9767 and a mean absolute percentage error (MAPE) of 0.0406. Compared with the benchmark XGBoost model, this represents an improvement in prediction accuracy exceeding 32%. This study provides a reliable and novel tool for predicting the mechanical properties of fiber-reinforced loess in seasonal frozen regions.
{"title":"Study on shear strength prediction model of glass fiber-improved loess in seasonal frozen regions based on POA-XGBoost","authors":"Xiyue Wang , Wanjun Ye , Qianqian Ma , Hao Yang , Minghao Zhang","doi":"10.1016/j.coldregions.2026.104857","DOIUrl":"10.1016/j.coldregions.2026.104857","url":null,"abstract":"<div><div>In seasonal frozen regions, loess is susceptible to mechanical damage under the action of freeze-thaw (F-T) cycles and moisture content changes. Fiber reinforcement is recognized as an effective technique for enhancing its engineering performance. As a critical index for assessing fiber-reinforced loess, shear strength is governed by multiple interactive factors, including the number of F-T cycles, fiber content, and confining pressure, rendering its accurate prediction challenging. To address this challenge, a series of triaxial shear tests on fiber-reinforced loess subjected to wet-dry freeze-thaw (WDFT) cycles was performed. The influence mechanisms of cycle number, fiber content, and confining pressure on shear strength indicators were revealed through Spearman correlation analysis. On this basis, to construct a high-precision prediction model, multiple algorithms including support vector regression (SVR), random forest (RF), and generalized additive model (GAM) were compared. Extreme gradient boosting (XGBoost) was finally selected as the benchmark model. Subsequently, the pelican optimization algorithm (POA) was introduced for automatic hyperparameter optimization of the model. A hybrid machine learning model (POA-XGBoost) was thus proposed. The results indicate that the POA-XGBoost model achieved superior predictive performance. On the test set, the model attained a coefficient of determination (R<sup>2</sup>) of 0.9767 and a mean absolute percentage error (MAPE) of 0.0406. Compared with the benchmark XGBoost model, this represents an improvement in prediction accuracy exceeding 32%. This study provides a reliable and novel tool for predicting the mechanical properties of fiber-reinforced loess in seasonal frozen regions.</div></div>","PeriodicalId":10522,"journal":{"name":"Cold Regions Science and Technology","volume":"245 ","pages":"Article 104857"},"PeriodicalIF":3.8,"publicationDate":"2026-04-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146184773","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-04-01Epub Date: 2026-01-15DOI: 10.1016/j.coldregions.2026.104830
Huayang Sun , Yanlin Huo , Xiaobing Ma , Xuesi Ji , Zhichao Xu , Zhitao Chen , Yingzi Yang
Early-age frozen damage poses significant challenges to concrete durability and structural integrity in cold-region construction. This study investigates the long-term mechanical properties of early-age frozen sulphoaluminate cement-engineered cementitious composites (SAC-ECC). Mechanical tests, single-fiber pullout tests, micromechanical modeling, and microstructural analyses were conducted to evaluate the effects of pre-curing times and frozen temperatures. The experimental results demonstrated that SAC-ECC with shorter pre-curing times (0.75 h), both compressive and flexural strengths increased as frozen temperatures decreased from 0 °C to −10 °C. In contrast, for SAC-ECC with longer pre-curing times (1.5 h and 3 h), the compressive and flexural strengths were aggravated with decreasing freezing temperatures. In addition, longer pre-curing times and lower frozen temperatures reduced tensile strength but significantly enhanced tensile ductility. For SAC-ECC frozen at −10 °C after 3 h of pre-curing, the tensile strain increased by 92.94% compared with the unfrozen group. Micromechanical and microstructural analyses revealed that for early-frozen SAC-ECC, a shorter pre-curing time improved the pore structure and fiber matrix interface, whereas a longer pre-curing time increased porosity and weakened interfacial bonding. The TOPSIS analysis can effectively balance the mechanical properties and time cost, thus providing valuable guidance for the application of SAC-ECC in cold-region construction.
{"title":"New insight into mechanical evolution and micro-mechanisms of early-age frozen engineered cementitious composites","authors":"Huayang Sun , Yanlin Huo , Xiaobing Ma , Xuesi Ji , Zhichao Xu , Zhitao Chen , Yingzi Yang","doi":"10.1016/j.coldregions.2026.104830","DOIUrl":"10.1016/j.coldregions.2026.104830","url":null,"abstract":"<div><div>Early-age frozen damage poses significant challenges to concrete durability and structural integrity in cold-region construction. This study investigates the long-term mechanical properties of early-age frozen sulphoaluminate cement-engineered cementitious composites (SAC-ECC). Mechanical tests, single-fiber pullout tests, micromechanical modeling, and microstructural analyses were conducted to evaluate the effects of pre-curing times and frozen temperatures. The experimental results demonstrated that SAC-ECC with shorter pre-curing times (0.75 h), both compressive and flexural strengths increased as frozen temperatures decreased from 0 °C to −10 °C. In contrast, for SAC-ECC with longer pre-curing times (1.5 h and 3 h), the compressive and flexural strengths were aggravated with decreasing freezing temperatures. In addition, longer pre-curing times and lower frozen temperatures reduced tensile strength but significantly enhanced tensile ductility. For SAC-ECC frozen at −10 °C after 3 h of pre-curing, the tensile strain increased by 92.94% compared with the unfrozen group. Micromechanical and microstructural analyses revealed that for early-frozen SAC-ECC, a shorter pre-curing time improved the pore structure and fiber matrix interface, whereas a longer pre-curing time increased porosity and weakened interfacial bonding. The TOPSIS analysis can effectively balance the mechanical properties and time cost, thus providing valuable guidance for the application of SAC-ECC in cold-region construction.</div></div>","PeriodicalId":10522,"journal":{"name":"Cold Regions Science and Technology","volume":"245 ","pages":"Article 104830"},"PeriodicalIF":3.8,"publicationDate":"2026-04-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146036532","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Snow avalanches pose a major hazard in mountainous regions, but their sporadic occurrence and remote locations hinder consistent regional monitoring. Automated remote sensing techniques, particularly those using Synthetic Aperture Radar (SAR), offer promising solutions for systematic data collection. However, validating SAR-based avalanche detections remains challenging due to the limited availability of ground truth data, spatial mismatches, temporal inconsistencies between reference datasets, and uncertainties associated with the relatively simple hypotheses underlying detection algorithms. This study assesses the performance and reliability of SAR-based avalanche debris detection across seven massifs in the French Alps over two winter seasons (2017–2018 and early 2020). The SAR-derived detections are evaluated against multiple indicators of avalanche activity, including avalanche inventories, snow cover simulation models, and hazard levels from official French avalanche bulletins. The findings demonstrate that, overall, SAR-based methods effectively capture the spatial and temporal patterns of ground-observed avalanche activity and align well with reported hazard levels, particularly during periods of elevated avalanche risk. Notably, for the Beaufortain massif during the 2020 season, SAR detections achieved a Pearson correlation coefficient of 0.65 with ground-based observations. Nevertheless, performance varies significantly across massifs and seasons, with strong correlations in some areas and weaker associations in others. The topographic characteristics (slope, elevation, aspect) of detected debris also show good agreement with other indicators. Despite inherent biases in each reference dataset, the results highlight the potential of SAR imagery for capturing regional-scale spatiotemporal dynamics of avalanches. While SAR offers valuable insights, detection remains far from perfect, underscoring the continued need for direct field observations and further refinement of detection algorithms to improve accuracy and validation.
{"title":"Can Sentinel-1 reliably provide regional-scale information on avalanche activity","authors":"Suvrat Kaushik , Fatima Karbou , Nicolas Eckert , Léo Viallon-Galinier , Adrien Mauss","doi":"10.1016/j.coldregions.2026.104822","DOIUrl":"10.1016/j.coldregions.2026.104822","url":null,"abstract":"<div><div>Snow avalanches pose a major hazard in mountainous regions, but their sporadic occurrence and remote locations hinder consistent regional monitoring. Automated remote sensing techniques, particularly those using Synthetic Aperture Radar (SAR), offer promising solutions for systematic data collection. However, validating SAR-based avalanche detections remains challenging due to the limited availability of ground truth data, spatial mismatches, temporal inconsistencies between reference datasets, and uncertainties associated with the relatively simple hypotheses underlying detection algorithms. This study assesses the performance and reliability of SAR-based avalanche debris detection across seven massifs in the French Alps over two winter seasons (2017–2018 and early 2020). The SAR-derived detections are evaluated against multiple indicators of avalanche activity, including avalanche inventories, snow cover simulation models, and hazard levels from official French avalanche bulletins. The findings demonstrate that, overall, SAR-based methods effectively capture the spatial and temporal patterns of ground-observed avalanche activity and align well with reported hazard levels, particularly during periods of elevated avalanche risk. Notably, for the Beaufortain massif during the 2020 season, SAR detections achieved a Pearson correlation coefficient of 0.65 with ground-based observations. Nevertheless, performance varies significantly across massifs and seasons, with strong correlations in some areas and weaker associations in others. The topographic characteristics (slope, elevation, aspect) of detected debris also show good agreement with other indicators. Despite inherent biases in each reference dataset, the results highlight the potential of SAR imagery for capturing regional-scale spatiotemporal dynamics of avalanches. While SAR offers valuable insights, detection remains far from perfect, underscoring the continued need for direct field observations and further refinement of detection algorithms to improve accuracy and validation.</div></div>","PeriodicalId":10522,"journal":{"name":"Cold Regions Science and Technology","volume":"245 ","pages":"Article 104822"},"PeriodicalIF":3.8,"publicationDate":"2026-04-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146036534","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-04-01Epub Date: 2026-02-02DOI: 10.1016/j.coldregions.2026.104853
Jianing Pan , Xin Zhang , Yibo Xi , Kuigeng Lin , Zhenyu Wang
Blade icing poses a major threat to the operational reliability and power generation of wind turbines in cold regions. To overcome the limitations of high computation demand and complicated modelling procedures in blade icing prediction, this study proposes an image-based deep learning model for efficient prediction of ice shape and accretion process. Firstly, systematic computational fluid dynamics (CFD)-based icing simulations were conducted on the blade of a 5 MW onshore wind turbine to investigate the effects of environmental parameters on ice accretion. The results revealed distinct characteristics of typical rime ice and glaze ice. A SegFormer-based model was developed for ice shape prediction, achieving superior performance compared with other models. This model was further enhanced by introducing a novel curvature consistency-based loss function, which significantly improved the geometric fidelity of predicted ice shapes, especially for fine-grained features such as sharp protrusions. Comprehensive evaluations verified the proposed model's robustness under varied environmental conditions, dynamic ice accretion processes, and different airfoil geometries. Furthermore, the predicted ice shapes of individual blade elements were integrated to reconstruct the full three-dimensional ice morphology and estimate the total ice mass across the entire blade. This study establishes a robust and efficient framework for ice shape prediction, which lays a solid foundation for the subsequent precise aerodynamic performance analysis of iced wind turbine blades.
{"title":"Image-based ice shape and accretion process prediction on wind turbine blades via deep learning with curvature consistency constraint","authors":"Jianing Pan , Xin Zhang , Yibo Xi , Kuigeng Lin , Zhenyu Wang","doi":"10.1016/j.coldregions.2026.104853","DOIUrl":"10.1016/j.coldregions.2026.104853","url":null,"abstract":"<div><div>Blade icing poses a major threat to the operational reliability and power generation of wind turbines in cold regions. To overcome the limitations of high computation demand and complicated modelling procedures in blade icing prediction, this study proposes an image-based deep learning model for efficient prediction of ice shape and accretion process. Firstly, systematic computational fluid dynamics (CFD)-based icing simulations were conducted on the blade of a 5 MW onshore wind turbine to investigate the effects of environmental parameters on ice accretion. The results revealed distinct characteristics of typical rime ice and glaze ice. A SegFormer-based model was developed for ice shape prediction, achieving superior performance compared with other models. This model was further enhanced by introducing a novel curvature consistency-based loss function, which significantly improved the geometric fidelity of predicted ice shapes, especially for fine-grained features such as sharp protrusions. Comprehensive evaluations verified the proposed model's robustness under varied environmental conditions, dynamic ice accretion processes, and different airfoil geometries. Furthermore, the predicted ice shapes of individual blade elements were integrated to reconstruct the full three-dimensional ice morphology and estimate the total ice mass across the entire blade. This study establishes a robust and efficient framework for ice shape prediction, which lays a solid foundation for the subsequent precise aerodynamic performance analysis of iced wind turbine blades.</div></div>","PeriodicalId":10522,"journal":{"name":"Cold Regions Science and Technology","volume":"245 ","pages":"Article 104853"},"PeriodicalIF":3.8,"publicationDate":"2026-04-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146184770","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-04-01Epub Date: 2025-12-22DOI: 10.1016/j.coldregions.2025.104800
Benjamin M. Roberts-Pierel , Christopher J. Crawford , Steven W. Brown , Raymond F. Kokaly , Kelly E. Gleason , Anne W. Nolin , Edward H. Bair , Brenton A. Wilder , Anton J. Surunis , S. McKenzie K. Skiles , Joachim Meyer , Allyson E. Fitts , Jeremy M. Johnston , Adam G. Hunsaker , Martin Stuefer , Trond Løke
Seasonal snow is an integral component of global hydrological systems, global energy budget and Earth's climate. As an important part of many Earth systems, seasonal snow is also an essential source of water for many human populations and ecosystems around the world. As such, the measurement of seasonal snow and characterization of uncertainty in those measurements is crucial. To elucidate potential uncertainty attributable to commonly used field spectrometers (and to a lesser extent imaging spectrometers) and associated reference panels, this work presents results from an intercalibration experiment conducted synchronously with the NASA 2023 Snow Experiment (SnowEx) Albedo campaign near Fairbanks, Alaska USA. Three sets of experiments were carried out under controlled laboratory conditions to characterize the radiometric and spectral wavelength consistency of the instruments as well as the white reference panels used to calculate reflectance from field measurements. Although there was generally good agreement between the instruments, panels, and the references, there were also some notable differences. One instrument showed an average − 74 % change from the reference for radiance, and multiple instruments exceeded the suggested 0.5 nm threshold for spectral wavelength scale. The Discussion section highlights how some of these findings and their implications could improve future field campaigns and general use/maintenance of these high-precision scientific instruments.
{"title":"A laboratory-based spectrometer intercomparison for the measurement of snow spectra","authors":"Benjamin M. Roberts-Pierel , Christopher J. Crawford , Steven W. Brown , Raymond F. Kokaly , Kelly E. Gleason , Anne W. Nolin , Edward H. Bair , Brenton A. Wilder , Anton J. Surunis , S. McKenzie K. Skiles , Joachim Meyer , Allyson E. Fitts , Jeremy M. Johnston , Adam G. Hunsaker , Martin Stuefer , Trond Løke","doi":"10.1016/j.coldregions.2025.104800","DOIUrl":"10.1016/j.coldregions.2025.104800","url":null,"abstract":"<div><div>Seasonal snow is an integral component of global hydrological systems, global energy budget and Earth's climate. As an important part of many Earth systems, seasonal snow is also an essential source of water for many human populations and ecosystems around the world. As such, the measurement of seasonal snow and characterization of uncertainty in those measurements is crucial. To elucidate potential uncertainty attributable to commonly used field spectrometers (and to a lesser extent imaging spectrometers) and associated reference panels, this work presents results from an intercalibration experiment conducted synchronously with the NASA 2023 Snow Experiment (SnowEx) Albedo campaign near Fairbanks, Alaska USA. Three sets of experiments were carried out under controlled laboratory conditions to characterize the radiometric and spectral wavelength consistency of the instruments as well as the white reference panels used to calculate reflectance from field measurements. Although there was generally good agreement between the instruments, panels, and the references, there were also some notable differences. One instrument showed an average − 74 % change from the reference for radiance, and multiple instruments exceeded the suggested 0.5 nm threshold for spectral wavelength scale. The Discussion section highlights how some of these findings and their implications could improve future field campaigns and general use/maintenance of these high-precision scientific instruments.</div></div>","PeriodicalId":10522,"journal":{"name":"Cold Regions Science and Technology","volume":"245 ","pages":"Article 104800"},"PeriodicalIF":3.8,"publicationDate":"2026-04-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145976385","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-04-01Epub Date: 2026-01-17DOI: 10.1016/j.coldregions.2026.104831
Francine Hematang , Jonni Marwa , Anton Sinery , Meliza Worabai , Dominggas Renwarin , Evelin Tanur , Obed Lense , Dina Arung Padang , Alexander Rumatora , Elieser Sirami , Ana Tampang , Christian Imburi , Petrus Dimara
Climate change continues to occur, so that the chrysosphere ecosystem has been heavily affected. This study aimed to investigate changes of glacier area in Puncak Jaya in response to ongoing global climate change issue. Landsat 2–9 satellite image series was used to see the dynamics of glacier change using digitized on-screen method. An important finding is that the tropical glaciers of Papua will survive until 2024, but melt and disappear faster than researchers previously predicted. Results showed that the glacier area has decreased by 7.28 km2 (97%) in over 44 years. In 1980, Puncak Jaya and Idenburg glaciers covered 7.46 km2 and then decreased to 0.19 km2 in 2024. Research also shows that only two glaciers remain, while four others Ngga Pilimsit Glacier, Meren Glacier, Southwall Hanging Glacier and West Northwall Firn Glacier have disappeared. In 2024, the Carstensz Glacier covered 0.050 km2 and the East Firn Northwall Glacier 0.136 km2. Another important finding is that the East Firn Northwall Glacier is predicted to disappear faster in 2028–2029 than the Carstensz Glacier in 2029–2030. In the future, it will be necessary to collect all the information from the field, estimate the impact on the ecosystem if the glacier completely disappears, and find out why East Northwall Firn Glacier melts faster than Carstensz Glacier.
气候变化不断发生,使温圈生态系统受到严重影响。本研究旨在探讨punak Jaya冰川面积的变化对全球气候变化问题的响应。利用Landsat 2-9卫星图像序列,采用数字化屏幕显示方法观察冰川变化动态。一项重要的发现是,巴布亚的热带冰川将存活到2024年,但融化和消失的速度比研究人员先前预测的要快。结果表明:44年来冰川面积减少了7.28 km2 (97%);1980年,Puncak Jaya和Idenburg冰川覆盖面积为7.46 km2, 2024年减少至0.19 km2。研究还表明,只有两个冰川仍然存在,而其他四个冰川,阿嘎皮利姆斯特冰川,梅伦冰川,南墙悬挂冰川和西北墙芬冰川已经消失。2024年,Carstensz冰川覆盖面积为0.050 km2, East Firn Northwall冰川覆盖面积为0.136 km2。另一个重要的发现是,预计2028-2029年东芬-诺斯沃尔冰川的消失速度将超过2029-2030年卡斯滕斯冰川的消失速度。在未来,有必要收集所有来自现场的信息,估计冰川完全消失对生态系统的影响,并找出为什么东北壁芬冰川比卡斯滕斯冰川融化得更快。
{"title":"Rapid retreat of tropical glaciers in Puncak Jaya, Papua: Four decades of change observed from Landsat Imagery, 1980–2024","authors":"Francine Hematang , Jonni Marwa , Anton Sinery , Meliza Worabai , Dominggas Renwarin , Evelin Tanur , Obed Lense , Dina Arung Padang , Alexander Rumatora , Elieser Sirami , Ana Tampang , Christian Imburi , Petrus Dimara","doi":"10.1016/j.coldregions.2026.104831","DOIUrl":"10.1016/j.coldregions.2026.104831","url":null,"abstract":"<div><div>Climate change continues to occur, so that the chrysosphere ecosystem has been heavily affected. This study aimed to investigate changes of glacier area in Puncak Jaya in response to ongoing global climate change issue. Landsat 2–9 satellite image series was used to see the dynamics of glacier change using digitized on-screen method. An important finding is that the tropical glaciers of Papua will survive until 2024, but melt and disappear faster than researchers previously predicted. Results showed that the glacier area has decreased by 7.28 km<sup>2</sup> (97%) in over 44 years. In 1980, Puncak Jaya and Idenburg glaciers covered 7.46 km<sup>2</sup> and then decreased to 0.19 km<sup>2</sup> in 2024. Research also shows that only two glaciers remain, while four others Ngga Pilimsit Glacier, Meren Glacier, Southwall Hanging Glacier and West Northwall Firn Glacier have disappeared. In 2024, the Carstensz Glacier covered 0.050 km<sup>2</sup> and the East Firn Northwall Glacier 0.136 km<sup>2</sup>. Another important finding is that the East Firn Northwall Glacier is predicted to disappear faster in 2028–2029 than the Carstensz Glacier in 2029–2030. In the future, it will be necessary to collect all the information from the field, estimate the impact on the ecosystem if the glacier completely disappears, and find out why East Northwall Firn Glacier melts faster than Carstensz Glacier.</div></div>","PeriodicalId":10522,"journal":{"name":"Cold Regions Science and Technology","volume":"245 ","pages":"Article 104831"},"PeriodicalIF":3.8,"publicationDate":"2026-04-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146036531","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-04-01Epub Date: 2026-01-23DOI: 10.1016/j.coldregions.2026.104845
Ye Zhang, Yufeng Zhan, Hongru Chen, Yongcan Zhu, Long Zhao, Yi Tian
The transmission line icing scenario is complex, with limited samples for different icing types and high sim-ilarity between them. Moreover, there is currently a lack of non-contact methods for identifying icing types. Therefore, this paper proposes a transmission line icing classification algorithm based on BiTex-ResNet34. Firstly, a quantitative analysis of the texture features of icing images is conducted to identify the most significant feature parameters that highlight the mean differences between different icing types, thereby enhancing the discriminability between them. Secondly, the model adopts a dual-branch architecture, with each branch containing a complete ResNet34 convolutional backbone network to parallelly extract both the raw features and texture features of icing images, thereby enhancing the model's feature representation. Finally, the Second-order Feature Fusion Module (SK-FM) module is embedded at different layers of the model's dual-branch architecture. This module integrates second-order features and concatenates the Selective Kernel (SK) attention mechanism to capture the correlations between different icing feature information, thereby improving the model's ability to distinguish between three types of icing: rime, hard rime, and soft rime. Experimental results show that BiTex-ResNet34 can accurately identify the three types of icing—glaze, hard rime, and soft rime—under complex environments, achieving precision, recall, and F1-score of 94.7%, 91.07%, and 92.85%, respectively, providing a new approach for transmission line icing type recognition.
{"title":"Research on transmission line icing classification and recognition algorithm based on BiTex-ResNet34","authors":"Ye Zhang, Yufeng Zhan, Hongru Chen, Yongcan Zhu, Long Zhao, Yi Tian","doi":"10.1016/j.coldregions.2026.104845","DOIUrl":"10.1016/j.coldregions.2026.104845","url":null,"abstract":"<div><div>The transmission line icing scenario is complex, with limited samples for different icing types and high sim-ilarity between them. Moreover, there is currently a lack of non-contact methods for identifying icing types. Therefore, this paper proposes a transmission line icing classification algorithm based on BiTex-ResNet34. Firstly, a quantitative analysis of the texture features of icing images is conducted to identify the most significant feature parameters that highlight the mean differences between different icing types, thereby enhancing the discriminability between them. Secondly, the model adopts a dual-branch architecture, with each branch containing a complete ResNet34 convolutional backbone network to parallelly extract both the raw features and texture features of icing images, thereby enhancing the model's feature representation. Finally, the Second-order Feature Fusion Module (SK-FM) module is embedded at different layers of the model's dual-branch architecture. This module integrates second-order features and concatenates the Selective Kernel (SK) attention mechanism to capture the correlations between different icing feature information, thereby improving the model's ability to distinguish between three types of icing: rime, hard rime, and soft rime. Experimental results show that BiTex-ResNet34 can accurately identify the three types of icing—glaze, hard rime, and soft rime—under complex environments, achieving precision, recall, and F1-score of 94.7%, 91.07%, and 92.85%, respectively, providing a new approach for transmission line icing type recognition.</div></div>","PeriodicalId":10522,"journal":{"name":"Cold Regions Science and Technology","volume":"245 ","pages":"Article 104845"},"PeriodicalIF":3.8,"publicationDate":"2026-04-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146075302","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}