Abstract The spatial mapping of social-ecological system (SES) archetypes constitutes a fundamental tool to operationalize the SES concept in empirical research. Approaches to detect, map, and characterize SES archetypes have evolved over the last decade towards more integrative and comparable perspectives guided by SES conceptual frameworks and reference lists of variables. However, hardly any studies have investigated how to empirically identify the most relevant set of indicators to map the diversity of SESs. In this study, we propose a data-driven methodological routine based on multivariate statistical analysis to identify the most relevant indicators for mapping and characterizing SES archetypes in a particular region. Taking Andalusia (Spain) as a case study, we applied this methodological routine to 86 indicators representing multiple variables and dimensions of the SES. Additionally, we assessed how the empirical relevance of these indicators contributes to previous expert and empirical knowledge on key variables for characterizing SESs. We identified 29 key indicators that allowed us to map 15 SES archetypes encompassing natural, mosaic, agricultural, and urban systems, which uncovered contrasting land sharing and land sparing patterns throughout the territory. We found synergies but also disagreements between empirical and expert knowledge on the relevance of variables: agreement on their widespread relevance (32.7% of the variables, e.g. crop and livestock production, net primary productivity, population density); relevance conditioned by the context or the scale (16.3%, e.g. land protection, educational level); lack of agreement (20.4%, e.g. economic level, land tenure); need of further assessments due to the lack of expert or empirical knowledge (30.6%). Overall, our data-driven approach can contribute to more objective selection of relevant indicators for SES mapping, which may help to produce comparable and generalizable empirical knowledge on key variables for characterizing SESs, as well as to derive more representative descriptions and causal factor configurations in SES archetype analysis.