Integration of clustering and principal component analyses in MCDA to identify feasible regions for managed aquifer recharge

• Semi-automatic estimations of criteria weights via multivariate statistical methods. • MAR feasibility depends on intrinsic suitability, water demand and availability. • Multiple realizations of criteria weights are used to conduct variability analysis • The results were in good agreement with those of previous studies at the demo site. • This approach is directly implemented in software for broad usage by practitioners. Managed aquifer recharge (MAR) provides a nature-based solution to water scarcity issues, contributing to the design of effective water management policies. The novelty of this study lies in the integration of multivariate statistical methods within the framework of multicriteria decision analysis to provide nonsubjective, semi-automatic estimations of MAR feasibility based on three thematic layers, particularly intrinsic (hydrogeological, topographical, meteorological) features, water availability and demand for MAR. The concept of MAR typology is used to define the MAR problem and select a set of criteria for each thematic layer that are relevant for the Sado River Basin (southern Portugal). The hierarchical clustering algorithm (HCA) is used to partition the study area into distinct subregions based on selected criteria. For each subregion and thematic layer, a set of weight coefficients for the criteria is generated via products of the cumulative variance of the principal scores with the corresponding eigenvector matrix, which are then used to generate multiple realizations of the thematic maps. The results reveal that the majority of the study area (60%) has mean feasibility scores less than 0.54, whereas the remaining 40% (highest values) vary within a smaller interval (0.54–0.67). The present results are consistent with the previous study, with both identifying the same high-suitability regions. Additionally, it allows exploration of the variability in criteria weights across multiple realizations, providing a preliminary indication of how this variability can influence the results. This approach can easily be implemented in software and ultimately automated, supporting the unsupervised streamlining of the decision-making process.