Development and application of artificial neural network models to estimate values of a complex human thermal comfort index associated with urban heat and cool island patterns using air temperature data from a standard meteorological station

• Published in: International journal of biometeorology Vol. 62; no. 7; pp. 1265 - 1274
• Main Authors: Moustris, Konstantinos, Tsiros, Ioannis X., Tseliou, Areti, Nastos, Panagiotis
• Format: Journal Article
• Language: English
• Published: Berlin/Heidelberg Springer Berlin Heidelberg 01.07.2018; Springer Nature B.V
• Subjects: Air temperature; Animal Physiology; Artificial neural networks; Biological and Medical Physics; Biophysics; Cities; Clusters; Data processing; Discomfort; Earth and Environmental Science; Environment; Environmental Health; Extreme heat; Extreme weather; False alarms; Greece; Hot Temperature; Hot weather; Humans; Humidity; Islands; Long wave radiation; Mathematical models; Meteorology; Model testing; Neural networks; Neural Networks (Computer); Original Paper; Plant Physiology; Radiation; Relative humidity; Statistical analysis; Temperature; Temperature data; Temperature effects; Temperature requirements; Thermal comfort; Thermosensing; Weather stations; Wind speed; Physiologically Equivalent Temperature (PET) index; Urban microclimate; Thermal climate indices; Neural network architecture; Thermal sensation; Performance criteria
• ISSN: 0020-7128; 1432-1254
• DOI: 10.1007/s00484-018-1531-5

The present study deals with the development and application of artificial neural network models (ANNs) to estimate the values of a complex human thermal comfort-discomfort index associated with urban heat and cool island conditions inside various urban clusters using as only inputs air temperature data from a standard meteorological station. The index used in the study is the Physiologically Equivalent Temperature (PET) index which requires as inputs, among others, air temperature, relative humidity, wind speed, and radiation (short- and long-wave components). For the estimation of PET hourly values, ANN models were developed, appropriately trained, and tested. Model results are compared to values calculated by the PET index based on field monitoring data for various urban clusters (street, square, park, courtyard, and gallery) in the city of Athens (Greece) during an extreme hot weather summer period. For the evaluation of the predictive ability of the developed ANN models, several statistical evaluation indices were applied: the mean bias error, the root mean square error, the index of agreement, the coefficient of determination, the true predictive rate, the false alarm rate, and the Success Index. According to the results, it seems that ANNs present a remarkable ability to estimate hourly PET values within various urban clusters using only hourly values of air temperature. This is very important in cases where the human thermal comfort-discomfort conditions have to be analyzed and the only available parameter is air temperature.