Improving Outdoor Thermal Comfort Forecasts
Author Information
Author(s): Kuzmanović Danijela, Banko Jana, Skok Gregor
Primary Institution: Faculty of Mathematics and Physics, University of Ljubljana
Hypothesis
Can machine learning methods improve the accuracy of operational forecasts of the Universal Thermal Climate Index (UTCI)?
Conclusion
Both linear regression and neural networks significantly improved the accuracy of UTCI forecasts, reducing the mean error to almost zero.
Supporting Evidence
- Both machine learning methods reduced the daily mean error from about 2.6°C to almost zero.
- The daily mean absolute error decreased from 5°C to 3°C for the neural network and 3.5°C for linear regression.
- Neural networks outperformed linear regression in terms of mean absolute error.
Takeaway
This study shows that using smart computer programs can help predict how comfortable it will feel outside by making weather forecasts more accurate.
Methodology
The study used seven years of hourly data from 42 meteorological stations and applied linear regression and neural networks for post-processing UTCI forecasts.
Potential Biases
Potential biases may arise from the reliance on meteorological station data and the specific machine learning models used.
Limitations
The study is limited to Slovenia and may not generalize to other regions with different climatic conditions.
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website