Accurate weather forecasts are crucial to planning disaster management, protecting existing infrastructure, and sustaining federal mission areas. Geospatial foundation models (GFMs) have recently been developed as impactful tools for various stakeholders as vision transformers increasingly outperform numerical weather prediction (NWP) models. The state-of-the-art for machine learning-based weather foundation models has been recently fueled by the vision transformer architecture and may be finetuned for regional or local weather predictions. The vision transformer architecture uses patches representing regions of space and time over the input data. However, weather reanalysis datasets, such as ERA5, provide weather data as projections of the Earth’s spheroidal topology onto a rectangle. This introduces both position-dependent warping and an arbitrary seam at the 180° line of longitude. This may be eliminated by considering the topology of the ERA5 dataset as a cylinder, padding certain longitudes to effectively achieve a wrap-around patch embedding that is not affected by a seam, analogous to previous work developing cylindrical convolutional layers. We demonstrate promising results on smaller vision transformers and propose that this method improves performance on regional finetuning tasks in the vicinity of the 180° line of longitude. To mitigate the warping effect of projection and reduce overreliance on positional embedding, we employ an inverse projection of the source data back to spherical coordinates. We then rotate the sphere so that data are projected back onto the source projection, where the patches are centered and minimally warped. This ensures consistent patch representation across latitudes. By selecting patches in spherical coordinates, the model weights geospatial weather data equally across the globe. We propose these techniques to improve the model’s generalization and are likely to scale to larger models. Further work is required to fully evaluate the impact on state-of-the-art GFMs. We implement these techniques on Microsoft’s ClimaX foundation model, finetuning to make forecasts local to government installations in the Pacific, working towards improved mitigation of weather uncertainty for operational assets near the 180° line of longitude.