When experiments are not feasible due to financial or ethical concerns, investigators turn to observational and increasingly natural experimental methods, where some exogenous source of variation in the treatment is used to improve the credibility and transparency of causal claims. In this context, determining whether key assumptions of natural experiments are or are not satisfied remains a critical task, and one that we argue in this paper Earth Observation data can help address. We focus on geographic regression discontinuity designs, where observational units on one side and another side of a border are seen as intervened upon by some policy implemented on one side, but not the other, of the border. We develop a series of evaluation criteria for empirically establishing whether units on one side or another of the boundary are empirically separable from the other using EO data streams. After testing this approach in a simulation study, we apply the approach to two geo-RDD designs.