GeoAI systems leverage models of geospatial data to inform decision making. Visual explanations of GeoAI model recommendations (XGeoAI) require combining algorithmic transparency with spatial representations. Our exploration and co-design study identified the
importance of model validation, uncertainty visualisation, and trust mechanisms in the renewable energy context. Yet our 14 participants’ interface sketches contained exclusively map-based visualisations without algorithmic elements. Our early prototype integrated LIME/SHAP within map interfaces to bridge these parallel conceptualisations.

Further validation (16 participants) identified four important research directions for XGeoAI research: spatial data bias, demand-supply dynamics, optimal configuration, and climate projections. A second prototype pursued the bias theme and was evaluated through cognitive walkthrough interviews with four domain experts, uncovering how visual design mediates trust in AI recommendations.

Our work highlights the complexity of integrating algorithmic explanations into geographic decision support systems and sets an agenda for XGeoAI research to close this gap. In the future, our work will lead to systems that leverage XGeoAI to help decision makers deploy renewal energy infrastructure effectively.