In the heart of Wrocław, Poland, a groundbreaking study is challenging the status quo of building performance simulations, with significant implications for the energy sector. Krzysztof Cebrat, a researcher from the Faculty of Architecture at Wrocław University of Science and Technology, has led a team that has developed a sophisticated machine-learning model to predict urban microclimates, potentially revolutionizing how we approach energy modelling in cities.
The team’s work, published in the journal ‘Buildings & Cities’ (translated from Polish as ‘Budynki i Miasta’), highlights the inadequacy of standardised meteorological datasets for urban environments. “We’ve found that a static correction factor just doesn’t cut it when adjusting meteorological data from suburban stations to city centres,” Cebrat explains. “The temperature dynamics near buildings are far too complex for that.”
To tackle this complexity, Cebrat and his team analysed air temperature variability using data from a city-centre weather station, 32 facade-mounted sensors, and thermographic imaging. They focused on locations representative of approximately 25% of Wrocław’s housing stock, comparing this data with standard meteorological data from Wrocław II, located at an airport.
The results were striking. The team’s machine-learning model, which accounts for factors like building type, location, facade orientation, and materials, achieved high accuracy in predicting thermal variability. “Our model achieved an R2 of over 0.93 in training and 0.92 in validation,” Cebrat reveals. “This underscores the need for microclimate-informed meteorological adjustments in building simulations.”
So, what does this mean for the energy sector? According to Cebrat, implementing corrections to outdoor temperature in urban environments can significantly enhance the accuracy of energy modelling, particularly for natural ventilation strategies addressing urban overheating. “However, due to the complexity of temperature dynamics near buildings, defining a single static correction factor for climate parameters such as temperature is unlikely to be effective,” he cautions.
The team’s method enables the generation of dynamic, high-spatial-resolution temperature data, incorporating facade orientation and specific building floor heights. These refined datasets can be used to adjust meteorological station measurements, improving the accuracy of simulations. When applied to a typical meteorological year dataset, this approach may offer a way to better represent thermal variability in urban environments, supporting more context-sensitive building-energy modelling.
This research could shape future developments in the field by encouraging a shift towards more nuanced, data-driven approaches to energy modelling. As cities continue to grow and urban overheating becomes an increasingly pressing issue, the ability to accurately predict and respond to microclimates will be invaluable. Cebrat’s work is a significant step in that direction, offering a compelling example of how machine learning and advanced data analysis can drive innovation in the energy sector.