Back to Data Consulting

Orbital Insight

Orbital Insight is a geospatial analytics startup that analyzes satellite, drone, aerial vehicle, and geolocation data to provide strategic business insights based on changes to human activity. They aim to extract information from aerial imagery to identify patterns in natural disaster damage, retail inventory and revenue, supply chain blockages, and foreign military surveillance, to be provided to a broad spectrum of influential clients including hedge funds and government defense agencies.

Problem

With such high stakes applications and consumers, Orbital Insight utilizes high resolution imagery for their computer vision algorithms and offers high accuracy trend analysis. However, these forecasts and insights can be compromised by cloud cover, which often obscures parts of the satellite imagery and leads to masked or blocked geographical regions. These concealed details can cost clients a fortune in terms of time, safety, and financial outcomes.

Methodology

We performed spatiotemporal imputation to help estimate the hidden data. Given historical records and geospatial information about objects of interest across a variety of regions, we developed time series forecasts rooted in spatial statistical techniques to model the missing objects in the images.

Results

Our final prediction pipelines utilized temporally-binned Kernel Density Estimation and Kriging (spatial linear regression). We also delivered a re-usable and scalable validation pipeline that enables Orbital Insight to evaluate the performance of future models by generating random clouds on clear imagery.