In the complex decision-making process of Deep Neural Networks (DNNs), eXplainable Artificial Intelligence (XAI) techniques have been employed to glassbox the blackbox of Artificial Intelligence (AI). As yet, there is little research examining the special challenges to explainability imposed by Geospatial Artificial Intelligence (GeoAI), chief among them spatial scale. We employ a case study using DNN for land use classification to illustrate four specific scale challenges in integrating GeoAI and XAI, namely, spatial resolution, spatial extent, semantic scales, and audience scale. Spatial resolution represents the level of details in the explanation of DNN. Spatial extent defines spatial boundaries in the application of XAI. Semantic scale implies the fitness of labelling and also suggests whether additional knowledge is required to infuse geography in the explanation. Audience scale indicates the appropriate delivery of XAI to its user base, which connects social and ethical issues of XAI with its application. These four scale challenges highlight important research directions in GIScience to achieve an explainable and interpretable GeoAI.
