Subject

When considering the impact of meteorological conditions (such as heat waves) on human health, it is not enough to only look at the individual meteorological variables. It is often the combination of these meteorological variable with the physical factors surrounding a person that will determine the thermal comfort that this person feels. e.g., the perceived higher temperature and thermal discomfort due to standing in the clear sun during summer or perceived lower temperature due to wind blowing during the winter (wind chill).

To quantify the thermal comfort, certain indices, such as the Universal Thermal Climate Index (UTCI) or the Physiological Equivalent Temperature (PET) score, were proposed [1]. With these indices (both expressed in °C), we can easily determine whether a person is subject to heat stress or cold stress. Prior research has already shown that there is a direct link between these indices and excess mortality [2]. It is, therefore, important to understand how these indices will evolve, especially in the context of climate change, the intensification of heat waves and the increasing urbanisation.

The computation of these indices is done by computing a human body energy balance model between the person and the surrounding physical environment. Especially the impact of radiation tends to be quite involved, as it is influenced by many environmental factors. This causes the model to be very complex but also computationally heavy. It is, therefore, interesting to look at alternative methods to obtain these thermal comfort indices at a lower computational cost.

With this master's thesis, the aim would be to investigate the usage of machine learning models to emulate thermal comfort scores, such as the UTCI or the PET score [3,4]. Here, you will train and compare machine learning models by using a variety of meteorological variables (temperature, wind speed …) and environmental factors (such as Sky view factor, land cover fractions ...) as input variables and targeting the corresponding thermal comfort scores. In particular we will make use of the VLINDER network for the meteorological variables. This is a network consisting of 70+ standardised weather stations placed in varying environment types.

Kind of work

Objectives

• Literature study on the thermal comfort scores and the usage of machine learning for emulation.
  
• Constructing a working machine learning model that outputs the desired thermal comfort score.
  
• Comparing and evaluating different types of machine learning models.
  
• (If enough time) Calculating the thermal comfort scores for an entire city during an interesting meteorological event (e.g., a heatwave).
  

Framework of the Thesis

[1] Zare, Sajad, et al. "Comparing Universal Thermal Climate Index (UTCI) with selected thermal indices/environmental parameters during 12 months of the year." Weather and climate extremes 19 (2018): 49-57.
br
[2] Di Napoli, Claudia, Florian Pappenberger, and Hannah L. Cloke. "Assessing heat-related health risk in Europe via the Universal Thermal Climate Index (UTCI)." International journal of biometeorology 62 (2018): 1155-1165.
br
[3] Ketterer, C., & Matzarakis, A. (2016). Mapping the Physiologically Equivalent Temperature in urban areas using artificial neural network. Landscape and Urban Planning, 150, 1-9.
br
[4] Philippopoulos, Kostas, et al. "A novel artificial neural network methodology to produce high-resolution bioclimatic maps using Earth Observation data: A case study for Cyprus." Science of The Total Environment (2023): 164734.

Expected Student Profile

Programming background (e.g., Python)
br
Background in machine learning
br
Experience with machine learning frameworks (e.g., TensorFlow or PyTorch)
br
Strong interest in weather and climat