A short-term probabilistic forecasting method for distributed photovoltaic (PV) clusters based on spatial encoding maps is proposed in this study. First, the spatial layout of the distributed PV clusters is encoded into a two-dimensional matrix, forming PV output feature maps and weather feature maps. Then, a conditional generative adversarial network is used as the forecasting model, transforming the short-term power forecasting problem of the cluster into a problem of generating future multi-period PV output feature maps under given conditions. By changing the input random vectors, multiple forecast scenarios can be generated, thus constructing probabilistic forecasting results. Finally, the effectiveness of the forecast results is comprehensively evaluated using three metrics: point-to-point error, interval coverage rate, and data distribution similarity. Case study results based on the Alice Spring distributed PV project at the Australian Solar Energy Centre indicate that the proposed method can fully exploit the spatiotemporal correlations within the cluster, simultaneously obtaining forecast results for all PV units in the cluster with good prediction accuracy and computational efficiency.