Introducing robustness evaluation and archetype analysis in drought risk assessments

Droughts have huge negative impacts on livelihoods and economies throughout the world, and climate change is expected to increase their future frequency and severity. For an effective drought management, drought risk assessment is considered of major importance. However, despite the high number of studies, shared and clear guidelines to perform drought risk assessments are missing, undermining the overall reliability of this procedure. A significant limitation common to most drought risk assessments is the lack of any form of validation. Moreover, checking the robustness of the assessment tools is of paramount importance, but appropriate data are usually not available for external validation; hence, internal validations are in many cases the only option. For this scope, we propose a simple but robust uncertainty analysis, using the methodology presented in the “Handbook on constructing composite indicators” of OECD (2008). An additional deficiency of most drought risk assessments is the missing link between the results and possible adaptation strategies. To address this limitation, we propose to use archetype analysis, which is an emerging approach for identifying recurrent patterns within cases and supporting a context-specific generalization of insights.  

The innovations introduced were applied to a drought risk assessment performed for the agricultural systems of five coastal watersheds of central and southern Tuscany, Italy. These watersheds are particularly prone to drought impacts because of the high concurrent water demand for domestic and agricultural uses during the summer months. To allow a better discretization, municipalities were selected as units of analysis. A total of 42 indicators were used to represent drought hazard, exposure, and vulnerability. Multiple drought hazard indicators were selected to estimate both past and future drought hazards, using ready-to-use data from public institutions. Overall, the southern part of Tuscany showed to be the most at risk, in particular the Grosseto province. For the robustness evaluation, we (1) excluded individual exposure and vulnerability indicators, (2) included the excluded indicators with the multicollinearity analysis, (3) assigned different weights, and (4) used an alternative aggregation method to calculate the composite risk indicator. Results in terms of average shifts in rankings and new rankings assigned revealed that the most uncertain parts were the selection of exposure indicators and the assignment of weights, but overall, the rankings were confirmed. The archetype analysis yielded as result seven clusters of municipalities; their characteristics were analysed and tailored adaptation strategies were proposed according to their specific drought risk profiles.