The building light and radar data are joined by selecting building centroids within a 200 foot radius of each radar observation and associating average lumination in that area with a measure of bird density in the atmosphere. This is done for each temporal record we have - the radar scans once every 10 minutes, each night. We encountered two main issues. First,  some buildings are not visible to our cameras, such as buildings "behind" the Empire State Building. If such an occluded building fell within 200 feet of a radar observation, we picked the nearest building with a similar number of floors, and used its light measuremnet instead. If a set of spatial radar observations had no visible buildings within 200 feet, we did not use that observation in our analysis. Second, areas of radar scans with reflectivity below a certain amount lacked records in the dataset. To address this, we imputed bird density values of "0" for each such missing record. This allowed us to maintain spatial and temporal consistency as we tracked building lights over time, over our study area.