Spatial Analysis of Drought Perceived Impacts Using Social Media Text Mining

Droughts affect diverse sectors, including water resources, agricultural productivity, and ecosystem stability. While indices like the Standardized Precipitation Index (SPI) and the Standardized Precipitation Evapotranspiration Index (SPEI) are widely employed to measure the intensity of droughts, they tend to focus on meteorological and hydrological aspects instead of social and economic dimensions. Notably, droughts of comparable meteorological severity can have vastly different outcomes, influenced by disparities in infrastructure, economic resilience, and community preparedness. Recent drought studies have highlighted the potential of integrating text mining and natural language processing to enhance drought impact assessments. However, many of these studies rely on official reports or newspapers, which often face limitations in temporal and spatial resolution due to the constraints of available data sources. In contrast, social media platforms like Twitter (X) host and disseminate real-time text data from individuals experiencing drought events, providing more granular and dynamic information about drought impacts that traditional methods may struggle to capture.

This study seeks to develop a framework for assessing perceived drought impacts through a set of sectoral impact scores generated from social media data by leveraging text mining techniques. Furthermore, the research compares these social media-derived scores with severity data from the report-based European Drought Impact Database (EDID) and physical drought indices to identify similarities and discrepancies between public perceived impacts, officially reported impacts, and meteorological drought intensity. To our knowledge, this is the first study to convert social media text into indicators of drought impacts across multiple categories, offering an innovative complement to traditional indices and enhancing our understanding of how affected communities perceive drought events.

Focusing on the 2022 Italian drought, we analyzed location-specific tweets using sentence embedding and large language models to identify sector-specific topics. We then examined the spatial and temporal patterns of perceived sectoral impact scores across Italy based on each tweet's relevance to the identified impact sectors. Our analysis revealed that Twitter activity about droughts peaked in the summer, with water availability and societal responses drawing the most attention in Northern Italy. This activity pattern closely aligned with the seasonality identified by SPI metrics, with areas of extreme drought conditions expanded during the summer months. On the other hand, comparisons with the report-based EDID showed inconsistencies, as EDID emphasized more severe impacts on agriculture. This suggests that while social media captures timely public perceived impacts, it may fail to fully reflect the depth or breadth of impacts in certain sectors due to the underrepresentation of specific groups on these platforms.