The Traveling Observer Model: Multi-task Learning Through Spatial Variable Embeddings