Despite the significant growth in geospatial data applications, there has been limited progress on building a scientific foundation for geospatial data science, making it difficult to develop reliable and trustworthy geospatial models and tools. The specific properties of geospatial data, as well as their volume, variety, and velocity, and the implicit but complex spatial relationships limit the applicability of traditional data science methods. We propose that geospatial data science be formally defined as a scientific process of extracting valuable information from raw geospatial data with reasonable effort. In this chapter, we define the emerging field of geospatial data science from a transdisciplinary perspective across the three closely related scientific disciplines of statistics, mathematics, and computer science. Our proposed definition aims to reduce the redundant work across siloed disciplines and to promote better understanding of the limits of geospatial data science as well as the expectations via examples.