The Language of Weather: Social Media Reactions to Weather Accounting for Climatic and Linguistic Baselines

TL;DR

The study addresses how public mood on social media responds to weather by incorporating local climatic baselines and regional language variation. It combines UK Twitter data with E-OBS weather observations and uses CIDER-based, dictionary-driven sentiment and lexicon analyses to build weather-aware, explainable measures, defined with location-specific $z$-scores $z_C(x,t) = \frac{C(x,t) - \mu_C(x)}{\sigma_C(x)}$. The findings show that baseline normalization reveals a consistent, non-linear mood response to weather and significant multivariate interactions, with regional patterns aligning when baselines are accounted for. These insights have practical implications for impact-based forecasting and risk communication by enabling context-sensitive interpretation of social-media signals about weather and climate change.

Abstract

This study explores how different weather conditions influence public sentiment on social media, focusing on Twitter data from the UK. By considering climate and linguistic baselines, we improve the accuracy of weather-related sentiment analysis. Our findings show that emotional responses to weather are complex, influenced by combinations of weather variables and regional language differences. The results highlight the importance of context-sensitive methods for better understanding public mood in response to weather, which can enhance impact-based forecasting and risk communication in the context of climate change.

Figures (5)

• Figure 1: Linguistic variation of tweets as a result of temperature, precipitation, humidity, wind speed, and barometric pressure fluctuations, compared to sentiment fluctuations. Spectra created using CIDER young_cider_2024.
• Figure 2: Change in sentiment at different weather conditions. Z-score conditions for every tweet have been calculated using historical gridded weather data Cornes2018.
• Figure 3: Hexbin plot comparing the average sentiment of tweets at different combinations of weather conditions.
• Figure 4: North vs South sentiment responses to temperature in the UK. Conditions and Sentiment have been normalised by z-scores on the right-hand side.
• Figure 5: North vs South sentiment responses to precipitation, wind, humidity, and pressure variations in the UK. Conditions and sentiment have been normalised by z-scores on the right-hand side.