Drought impact profiles: Analyzing multivariate socio-economic drought impacts using nonlinear dimensionality reduction

Abstract

Socio-economic drought impacts often occur concomitantly across multiple sectors, leading to more severe consequences than if they affected single sectors. Improved management of such disasters requires cross-sectoral impact assessments and analyses. As such, analyzing how regions are affected by multiple impacts can provide crucial information for mitigating their consequences. Here, we characterize the multivariate distributions of socio-economic drought impacts. Our aim is to understand patterns by which diverse drought impacts co-occur. We introduce the concept of drought impact profiles, which describe characteristic distributions of co-occurring impacts. To this end, we use a unique spatio-temporal dataset generated with text mining and machine learning applied to newspaper articles. This dataset describes reported socio-economic drought impacts along seven categories (agriculture, forestry, fires,  social, aquaculture, livestock, waterways) in Germany between 2000-2022. We combine several dimensionality reduction algorithms (PCA, ISOmap, self-organizing maps) to generate robust and interpretable representations of the drought impacts. Our results show characteristic patterns for both particular drought events and regions. Also, the applied methods provide a low-dimensional representation of the multivariate socio-economic drought impacts. This research provides a methodological contribution to the holistic, empirical investigation of co-occurring drought impacts. The proposed methods can inform risk models, and policy-makers on the urgency of cross-sectoral governance approaches. Also, the proposed method could apply to other hazards or compound events.