Knowledge of high disparities in developmental indicators, such as per capita income, on a local level, is crucial for policy design. However this data is scarce and difficult to obtain. Moreover, this valuable and critical data is often the most challenging to obtain in the areas that need it the most. Through this project, we provide a low-cost, scalable method to predict the spatial distribution of average yearly income levels within a nation or subnational division from satellite imagery. Using satellite imagery of the Los Angeles Metropolitan Area we provide a proof-of-concept model, leveraging convolutional neural networks that can be trained to identify features indicating variance in average income and hence predict income levels of a locality. We hope that this model can generalize to other archetypes of geographic development in developing nations.