Spatial data are being collected at unprecedented scales and variety: the volume of Earth observation data at NASA alone is projected to reach 250-PB by 2025, and the number of GPS receivers has surpassed 6-billion in 2021. Such datasets are at the core of decision-making across many critical sectors including agriculture, transportation and public health, and are broadly used to tackle some of the most pressing challenges including climate change and the COVID-19 pandemic. While machine learning is important for automating the analysis of such gigantic datasets, direct applications of these methods often fall short due to the unique challenges posed by spatial data. This talk will focus on the common heterogeneity problem, where not only data distributions are non-stationary in space, but also the footprints of the distributions are unknown. We will discuss two model-agnostic frameworks to address the challenge from two different perspectives: performance (e.g., F1 scores) and fairness. The talk will conclude with a brief discussion on other challenges and emerging opportunities.