Multivariate analysis of spatio-temporal droughts: a way to advance drought planning and management

Abstract

Drought is a complex, multifaceted phenomenon that crosses political and geographical borders. However, it is usually studied using one-dimensional time series of drought variables, ignoring the multivariate characteristics of droughts and their interdependence. On the other hand, increasing the complexity of drought analysis by considering multiple variables can be difficult to incorporate into management. Impediments to using multivariate analysis in this context may management. A framework to perform multivariate frequency analysis at a spatial level that can be related to socioeconomic impacts was proposed. The 2012-2018 drought in northeast Brazil was studied using this framework. Improved risk assessment was achieved by simultaneously considering drought duration and severity at the hydrographic region level. This framework is already being used as a planning instrument in new proactive drought plans that are being constructed in the studied region. (iii) The multivariate characterization of droughts can present the following dilemma: to oversimplify and ignores other sides of the problem or to increase complexity and make it difficult for decision-makers to have a clear view of the event. To improve our understanding of the compounding effects of spatio-temporal relationships it is proposed a simple way to analyze the drought dynamics over time and space. Intra-event analysis of drought dynamics and searching for patterns and relationship between mean drought characteristics are presented. The first part proposes the use of growth curve, growth rate and acceleration to understand how drought evolves in time and space inside the drought event. The second analyses the mean characteristics of all drought events in search for patterns and relationships that can be helpful for decision-makers in drought monitoring and early warning. It was found that central part of the Northeast region developed longer, more severe, and more widespread droughts than any other area. This result is important for preparing for upcoming events that presents its centroid’s onset at this region. This thesis presents strategies and frameworks that can increase our knowledge of drought events by considering multiple characteristics but still making them simple and straightforward for decision-makers. This way we can link science and management, advancing our understanding of drought events and improving our ability to cope with future droughts. include high entrance barriers for analysts to perform multivariate analysis and difficulty presenting the results to decision-makers. This gap prevents us from adequately forecasting, planning, and managing to cope with the adverse effects of droughts. Therefore, finding a balance between complexity and simplicity is something that must be sought. In this thesis, I argue that the shift to understanding drought as a multivariate event provides enough gain of information to justify its use in drought planning and management if it is possible to make a link between science and management. To do so, I study droughts' compounding relationships in three different dimensions of drought risk: monitoring and early warning, multivariate frequency analysis, and spatio-temporal characterization. (i) Droughts can be anticipated and mitigated using a Continuous Drought Probability Monitoring System (CDPMS). This system uses the conditional probability theory to monitor the occurrence probability of drought during the rainy season, as the rainy season advances. The likelihood that the rainy season is defined as drought is updated every month inside this period. The concept of precipitation thresholds is used in CDPMS to simplify the drought analysis for decision-makers. The model was assessed in mainland Portugal and demonstrated the capacity to anticipate drought by modeling the complex dependence structures between total and given precipitation. The proposed framework can be used to quantify the risk of drought and mitigate its impacts; (ii) Traditional drought frequency analysis uses only one characteristic of drought (e.g., the duration), which is an incomplete representation of the event and ignores mutual dependence. Also, it frequently ignores the spatial extent of droughts or considers it without a relationship with drought