Spatial data, which describes the position or shape of objects in space, plays a central role in numerous scientific and analytics applications. However, there is a significant mismatch between the requirements of modern spatial applications and the capabilities of current data management solutions. First, traditional spatial query operators evaluate spatial relations with time-consuming geometric tests that oppose the interactivity expected from exploratory applications, creating an undue overhead. Furthermore, they rely on rough spatial approximations that are inaccurate and do not capture the complexity and diversity of today’s spatial data. In this talk, I will present a novel family of spatial data management tools that aim to enhance the performance of modern spatial applications. First, I will show that spatial query operators can be decomposed into primitive graphics operations that are efficiently executed by graphics hardware (GPU). Furthermore, I will introduce the notion of distance-bounded spatial approximations and explain how they enable us to leverage modern hardware and trade accuracy for interactivity. Finally, I will discuss some open challenges in managing heterogeneous spatial data for earth observation data analytics.