From indices to impacts: Understanding the dynamics of drought impacts through socio-economic clustering

Over the past decade, the Horn of Africa (HoA) has been plagued by recurrent drought events that have had devastating impacts on the population. The frequency, duration and severity of these droughts are expected to increase in the wake of global warming, leading to higher losses and damages if the vulnerability of the population is not reduced. Monitoring and early warning systems for droughts are based on various drought hazard indicators. However, assessments of how these indicators are linked to impacts are rare. For adequate drought management, it is essential to understand and characterise the drivers of drought impacts, especially in the HoA, where most studies focus either on meteorological droughts, agricultural droughts or the propagation of droughts through the hydrological cycle, without considering the relationship between hazard and impact. Drought hazard indices alone cannot capture the vulnerability of the system. In this study, we identify meaningful indices for the occurrence of region- and sector-specific impacts. We assess the effectiveness of socio-economic clustering in categorising counties based on common characteristics and their correlation with historical drought impacts (malnutrition, milk production and trekking distances to water sources). Using Random Forest (RF) and Spearman correlation analyses, we examine the link between drought indices (Standardised Precipitation Index, Standardised Precipitation Evapotranspiration Index, Standardised Soil Moisture Index, Standardised Streamflow Index and Vegetation Condition Index) with different accumulation periods and the impact data. We find that clustering regions based on vulnerability proxies significantly improves the hazard-impact relationship, emphasising the importance of considering vulnerability factors in drought risk assessment. Our results indicate an impact-specific relationship that is strongly influenced by the vulnerability of the region. In particular, household and livestock distance to water is most strongly associated with medium- to long-term precipitation-based indices (2-10 months), while milk production can be associated with a variety of indices with different accumulation periods (5-24 months), and malnutrition is correlated with precipitation- and streamflow-based indices (5-24 months). Household and livestock distance to water is well modelled by clusters reflecting low access to improved sanitation and safe water sources, high poverty, aridity and gender disparities. Malnutrition was well modelled by clusters related to aridity, average precipitation, food consumption score, access to water sources, improved sanitation and poverty levels. The type of clustering used in modelling the impact of drought on milk production does not have a major impact on the performance of the models. We then apply this relationship to hindcast drought indices to obtain impact data on individual counties for periods when no impact monitoring was done yet. With that information we estimate the associated risk under specific climatic conditions. By recognising the drivers and vulnerability factors that influence the sensitivity of counties to drought, communities can better prepare and mitigate the impacts of drought.