Open-loop groundwater heat pumps (GWHP) offer a climate-friendly solution for the energyintensive task of cooling and heating buildings. They extract groundwater from the aquifer to cool or heat buildings, before reinjecting the water back into the subsurface. As a result, heat plumes form around the injection point. In areas with a high density of GWHP systems, optimizing the placement of extraction and injection wells is crucial for ensuring high efficiency, which requires highly accurate estimations of heat plumes in the aquifer. Previous work primarily relies on a two-stage approach using convolutional neural networks (CNNs) to approximate groundwater temperatures. In contrast, this work explores an iterative single-stage estimation approach. To achieve this, the input parameters are expanded to provide the CNN with additional temperature information, allowing it to estimate the global domain iteratively. The results of this study demonstrate that a single-stage approach is feasible for estimating the global domain. However, it comes with a significant increase in computational cost and slightly reduced accuracy. Further research is necessary to fully explore the potential of this alternative method and optimize its performance.