Remote Sensing: Measurement and Causal Inference

• In Event: Remote Sensing—Methods and Applications

Abstract

Satellite imagery offers researchers the unprecedented ability to measure outcomes on the ground at high spatial and temporal frequency. This paper develops and validates a set of methods which take advantage of three sources of variation in the remotely sensed signal to measure outcomes and infer potential outcomes. Spectral variation in which wavelengths of light are reflected or absorbed by different materials can help categorize the primary contents of a single pixel at one moment in time. Temporal variation is useful for distinguishing land cover types with seasonal variation (especially vegetation) of a pixel over time. Spatial variation allows measurement of features which are larger than one pixel by characterizing spatial patterns in spectral and temporal variation. I show how traditional time series methods as well as neural networks with recurrent and convolutional structures can accurately classify pixels and objects. I show how these measurement strategies can be combined with treatment modeling (eg. propensity score methods) and outcome modeling (eg. Matrix completion methods) to perform causal inference. Finally I demonstrate these methods by testing whether a land-titling reform program in Benin influenced the behavior of those who received titles and the degree to which ISIS provided public goods during its occupation of Mosul.

Author

• Luke Sanford, Yale University