Geospatial data representation has evolved significantly over the years, from basic points, lines, and polygons to more complex embeddings. Geospatial embeddings, a technique used in spatial analysis, map geographic locations to vectors of real numbers, enabling the integration of diverse data types and facilitating advanced spatial analysis tasks. By mapping geographic entities to vectors of real numbers, embeddings capture not only the spatial coordinates but also the semantic meaning and relationships embedded in the data. This transformation enables the integration of diverse spatial data types, such as satellite imagery, GIS layers, textual descriptions, and sensor data, into a unified representation that preserves the unique characteristics, underlying patterns, and relationships between data. These embeddings enable machine learning algorithms to perform tasks such as location prediction, change detection, and semantic analysis with unprecedented accuracy. These representation methods facilitate the integration of geospatial data into deep learning models and provide a mechanism for efficiently comparing, indexing, and classifying geometric entities. This paper explores various spatial embedding techniques, their applications, challenges, and future directions. The paper also provides a comparative analysis of different approaches and discusses their effectiveness in diverse geospatial domains. Finally, we identify key insights, research gaps, and research scope in the field of geospatial embeddings.