How Swedish Students Help Map Temperature Perception in Real-Time
A citizen science project in Swedish schools reveals marked regional differences in how people perceive cold and heat. By combining human observations, satellite data, and AI, researchers pave the way for more personalized forecasts.
Over 500 students across Sweden spent a year recording how they feel when temperatures drop or rise, generating over 30,000 responses that revealed previously invisible perception gaps in classical models.
What the study reveals: Precise results and key figures
Analyses show that northern residents perceive ambient temperatures as an average of 2 to 3°C colder than their southern counterparts, even after correcting for humidity and wind differences. In urban areas, the sensation of heat is amplified by about 1°C compared to rural zones, an effect attributed to urban heat islands. Responses also vary by age: adolescents report feeling cold more intensely than adults, although the exact number of points difference remains unconfirmed information at this stage.
These perception gaps, previously masked by objective temperature measurements, are now quantified through a predictive model that associates subjective responses with official weather data. The model, trained on over 30,000 observations, achieves 78% accuracy in predicting thermal sensation at the communal level.
How: Research method and data used
The "Kallt och svettigt" project engaged classes from 6th to 12th grade who weekly recorded their thermal sensation via a mobile app. Each entry included locally measured temperature, humidity rate, wind speed, and a sensation note ranging from -1 (very cold) to +5 (very hot). These human data were then coupled with Copernicus satellite data, ECMWF reanalyses, and Météo France observations to enrich the variable set.
Researchers built a convolutional neural network capable of ingesting both physical variables (temperature, humidity, radiation) and subjective responses. The model was validated by cross-validation, and its performance was compared to GraphCast and Pangu-Weather, two purely physical forecasting models, showing notable improvement in predicting local sensations.
Scientific explanation of the phenomenon
The difference between measured temperature and perceived sensation is explained by several physiological and environmental processes. The human body regulates its internal temperature through cutaneous vasodilation or vasoconstriction, sweating, and shivering. In northern regions, prolonged exposure to cold winters leads to acclimatization that increases sensitivity to cold, resulting in a colder perception for the same ambient temperature value.
