PSI - Issue 73

Available online at www.sciencedirect.com Available online at www.sciencedirect.com ScienceDirect Structural Integrity Procedia 00 (2025) 000–000 Available online at www.sciencedirect.com Structural Integrity Procedia 00 (2025) 000–000

www.elsevier.com/locate/procedia www.elsevier.com/locate/procedia

ScienceDirect

Procedia Structural Integrity 73 (2025) 138–145

23rd International Conference on Modelling in Mechanics 2025 Automated ensemble machine learning models for roof snow load estimation of heated flat roofs Lenganji Simwanda a* , Jana Markova a , Miroslav Sykora a , Thomas Kringlebotn Thiis b a Klokner Institute, Czech Technical University in Prague, Šolínova 7, Prague, Czech Republic b Norwegian University of Life Sciences, Department of Building and Environmental Technology, NO-1432 Ås, Norway Abstract Accurate prediction of roof snow loads is essential for structural safety in cold climates. This study applies an automated machine learning (ML) framework to estimate snow loads on heated flat roofs with thermal transmittance U = 1.0 W/m²K (no sliding) using long-term hourly climate data for Oslo, Norway. Four ensemble machine learning models—Random Forest (RF), Gradient Boosting Machine (GBM), Categorical Boosting (CatBoost), and extreme gradient boosting (XGBoost)—were trained to predict roof snow load from meteorological inputs such as temperature, precipitation, humidity, wind, and solar radiation. The models were evaluated on an independent test set, achieving high accuracy with 2 values exceeding 99.6% and root-mean-square errors below 0.05 kN/m². Shapley additive explanation (SHAP) analysis confirmed the dominant influence of ground snow load, snow depth, air and soil temperatures, surface pressure and thermal radiation on roof snow load formation. Local SHAP analysis also revealed the nonlinear effects and interactions between meteorological variables on roof snow load. The models effectively captured both accumulation and melt events, demonstrating their utility for fast, reliable snow load estimation without the need for complex physical simulations. The approach is scalable to other insulation levels and climates, offering a promising tool for data driven snow load assessment and structural design optimization. © 2025 The Authors. Published by ELSEVIER B.V. This is an open access article under the CC BY-NC-ND license (https://creativecommons.org/licenses/by-nc-nd/4.0) Peer-review under responsibility of 23rd International Conference on Modelling in Mechanics 2025 organizers Keywords: Roof snow load; Ensemble machine learning; Climate data; Flat roofs; No Sliding; SHAP. 1. Introduction Snow accumulation on building roofs is a critical structural safety concern in cold regions. Excessive snow loads have caused numerous roof failures; for example, during the 2005–2006 winter, over 200 roof collapses were reported in Europe due to heavy snowfall (Croce et al., 2018). Traditionally, building codes estimate the roof snow load by a* a a a Klokner Institute, Czech Technical University in Prague, Šolínova 7, Prague, Czech Republic b Norwegian University of Life Sciences, Department of Building and Environmental Technology, NO-1432 Ås, Norway Abstract Accurate prediction of roof snow loads is essential for structural safety in cold climates. This study applies an automated machine learning (ML) framework to estimate snow loads on heated flat roofs with thermal transmittance U = 1.0 W/m²K (no sliding) using long-term hourly climate data for Oslo, Norway. Four ensemble machine learning models—Random Forest (RF), Gradient Boosting Machine (GBM), Categorical Boosting (CatBoost), and extreme gradient boosting (XGBoost)—were trained to predict roof snow load from meteorological inputs such as temperature, precipitation, humidity, wind, and solar radiation. The models were evaluated on an independent test set, achieving high accuracy with 2 values exceeding 99.6% and root-mean-square errors below 0.05 kN/m². Shapley additive explanation (SHAP) analysis confirmed the dominant influence of ground snow load, snow depth, air and soil temperatures, surface pressure and thermal radiation on roof snow load formation. Local SHAP analysis also revealed the nonlinear effects and interactions between meteorological variables on roof snow load. The models effectively captured both accumulation and melt events, demonstrating their utility for fast, reliable snow load estimation without the need for complex physical simulations. The approach is scalable to other insulation levels and climates, offering a promising tool for data driven snow load assessment and structural design optimization. © 2025 The Authors. Published by ELSEVIER B.V. This is an open access article under the CC BY-NC-ND license (https://creativecommons.org/licenses/by-nc-nd/4.0) Peer-review under responsibility of 23rd International Conference on Modelling in Mechanics 2025 organizers Keywords: Roof snow load; Ensemble machine learning; Climate data; Flat roofs; No Sliding; SHAP. 1. Snow accumulation on building roofs is a critical structural safety concern in cold regions. Excessive snow loads have caused numerous roof failures; for example, during the 2005–2006 winter, over 200 roof collapses were reported in Europe due to heavy snowfall (Croce et al., 2018). Traditionally, building codes estimate the roof snow load by © 2025 The Authors. Published by ELSEVIER B.V. This is an open access article under the CC BY-NC-ND license (https://creativecommons.org/licenses/by-nc-nd/4.0) Peer-review under responsibility of the scientific committee of the event organizers

2452-3216 © 2025 The Authors. Published by ELSEVIER B.V. This is an open access article under the CC BY-NC-ND license (https://creativecommons.org/licenses/by-nc-nd/4.0) Peer-review under responsibility of 23rd International Conference on Modelling in Mechanics 2025 organizers 2452-3216 © 2025 The Authors. Published by ELSEVIER B.V. This is an open access article under the CC BY-NC-ND license (https://creativecommons.org/licenses/by-nc-nd/4.0) Peer-review under responsibility of 23rd International Conference on Modelling in Mechanics 2025 organizers

2452-3216 © 2025 The Authors. Published by ELSEVIER B.V. This is an open access article under the CC BY-NC-ND license (https://creativecommons.org/licenses/by-nc-nd/4.0) Peer-review under responsibility of the scientific committee of the event organizers 10.1016/j.prostr.2025.10.022