Drought perceived impacts via text mining of social media

Droughts have complex, multifaceted consequences for society, yet drought indices often focus on drought’s physical characteristics, overlooking its societal impacts, while impact studies are usually published several months later or annually. In this context, social media provides a unique lens into public perceived impacts, capturing first-hand experience that conventional monitoring approaches might miss in a timely way. We investigate the consequences of the 2022 drought in Italy by analyzing Twitter posts. Leveraging large language models (LLMs), we classify the reported impacts into seven impact categories and analyze their spatio-temporal patterns. We observe that public discourse on drought is primarily concentrated in Northern Italy, where economic activity relies more heavily on agriculture and industry compared with the rest of the country. Discourse across all sectors peaks in summer, indicating a collective increase in concern across sectors. We find strong associations between the tweet-derived impacts and socioeconomic indicators such as crop prices and hydropower generation. Yet, Twitter data is weakly aligned with the standardized precipitation index. These findings suggest that meteorological conditions alone do not directly drive public response; rather, public perception aligns more closely with the tangible consequences of drought. We demonstrate the potential of using social media to enhance drought monitoring by capturing near-real-time, ground-level public responses. As such, it serves as a valuable complement to physical drought indicators and traditional statistical sources, supporting more informed impact assessment and management strategies.