A Philosophy and Keys to Creating a Thriving Aerial Data Operation — Part 2

The three main sources of aerial imagery — drones, aircraft, and satellites.

Recap of Part 1: In the last post, we spoke about the ubiquitous nature of aerial imagery data in geospatial analytics and the three main sources of collecting this data (drones, aircraft, and satellites). We also touched on two important keys in running the data collection side of the operation:

  1. Understanding the spatial and spectral resolution needed for the product.
  2. Understanding the frequency of the capture needed for the product.

In this post, we will transition from the data collection aspect of the operation over to the data processing side of the operation and introduce three more keys.


1) Understand the Data Processing Operation the Customer Needs

The Golden State Warriors have dominated the NBA over the past few years, winning championships in two of the last three seasons. A large part of their success has been fueled by the play making ability of their players to pass the ball to give other teammates easy baskets. However, for all those great passes, if the likes of Draymond, Klay, or Durant were not able to finish by scoring, those great passes might as well not matter, since they did not translate to points.

I think this anecdote translates well to an efficiently run imagery data operation — once an image is acquired it needs to be processed. Depending on intended use and the customer’s needs, this processing can include everything from image mosaicing (stitching together individual images), to orthorectification and georeferencing, contrast-stretch algorithms and even in some cases, analysis. At the same time, any processing must be completed and delivered with relevant insights to the customer within an appropriate time frame. A captured image (in basketball, a great pass) is no use without proper processing (in basketball, the shot that was made).

Looping back to agriculture, a farm agronomist will incur diminishing returns if he or she must wait three weeks to utilize imagery of the crop. Weather and soil conditions can change in time intervals faster than three weeks, and much of the utility of having imagery would be significantly reduced.

On the contrary, if a company like Uber or Mapbox is attempting to build up a basemap of data, they may not be held to any customer in need of data immediately and are probably not concerned with variables like quickly changing soil conditions. With this in mind, the time to process the data can be increased to reduce costs. A thorough cost optimization model should be created to dictate the amount of processing time needed.

The key question to ask here is: What types of investments should be made on the processing side to ensure a turnaround time appealing to the end client? Proper investments need to be made on the amount of processing horsepower allocated to image processing to ensure a delivery time that creates an appealing product. Another critical part of the processing pipeline are well placed Quality Control checks, which brings me to my next point.

2) Understand Where to Set Up Quality Control Checks

Given the highly variable nature of the real world as it relates to imagery capture (think clouds, smog, haze, storms), not all imagery will be usable for the product. Additionally, image processing steps can produce erroneous results, which will decrease product quality. Correctly placed quality control checks during image acquisition and subsequent processing can translate into thousands of dollars of savings in labor, aerial asset usage, or data processing. Therefore, it is imperative to have quality control checks at strategic points during the overall data operation.

In general, it is useful to have an initial quality check or some sort of mechanism to inform an operations team of a camera failure or excessive cloud cover in imagery — that would be the first hindrance to extracting valuable insights from captured data and would prevent processing data with little added value to the product.

Further down the data pipeline, other efforts should be made to quality check data before time intensive processes, such as a large scale stitch of images over a significant geographic region. Finding “bad” data and either replacing it or parsing it out before plugging it into a resource intensive processing environment will allow the data pipeline to focus efforts on low risk imagery, which would increase the efficiency of the entire data operation.

Another significant benefit of timely quality control checks is the ability to re-capture data at a lower cost. Obviously, costs may not go up with a geosynchronous satellite, as the same area can be captured without relocating a satellite, or with serendipitous routing of drones or aircraft. But for other use cases, it is logical to assume that an aircraft, drone, or non-geosynchronous satellite will be further from the area of interest on the ground the longer it has been since the original capture. Therefore, quickly finding poor data should drive down re-acquisition costs, as well as help mitigate missing the target deadline or goal. Of course, depending on the type of business, one could acquire imagery from another provider and have a contractual agreement for certain imagery specifications, and in this case you may not have to worry about the monetary cost of re-acquisition of data.

The key question to ask here is: When is my data most prone to distortion that it would delay integration into the final product or would drive acquisition and processing costs up?

Satellite image of Earth from June 19th, 2017 (via NASA/GSFC/JPL-Caltech/EOSDIS/MODIS). As one can see, cloud cover and weather can be a significant hurdle to overcome in processing imagery. The quality controls in the data pipeline should address this.

3) Understand the Regulatory, Legal, and Privacy aspects of your Data and the Implications of the Operation on Other People’s Livelihoods

This key is a loaded point and can easily be glanced over, however, a lack of awareness about the regulatory, legal, and privacy aspects of the operation can be costly. On the regulatory front, a best practice would be to stay abreast of FAA, ATC, and perpetually changing drone laws. The nascent state of the industry makes it vulnerable to shifts and the laws should be reviewed in the context of your operation.

Moving on to privacy, the United States and Canada in general are relaxed with regard to image acquisition, but some international countries have a data review process for captured imagery. There should be an understanding of nuances of international capture, and there might be a need for different data operation strategies for different countries in the event that export of data is not allowed or highly regulated.

Additionally, there is always a possibility of data or asset loss. A well thought out contract with partners (if there are partnerships) should have a contingency plan for this.

Lastly, we are in an age of automated data processing which is only expected to speed up in the coming decades. Insights that used to take physical labor to obtain can now be extracted algorithmically. When running the operation, one should be aware if the product takes away from “traditional” jobs or increase unemployment in certain industries, such as agriculture, oil, or energy. If it does, awareness can bring empathy when communicating with people in the industry.

The key question to ask here is: What are the regulations surrounding imagery capture and relevant data privacy laws in the various countries of operation?


  • This two part post mainly focuses on passive sensors to acquire data, but there are also other types of aerial data that can be captured using active sensors, such as LiDar or radar sensors. Depending on the use case, it could make sense to use those types of sensors instead of cameras that capture images.
  • “Traditional” data such as economic or industry specific indices used in combination with the aerial data can potentially add more value to the operation. Therefore, there is no reason to solely rely on aerial imagery data alone.
  • There are certainly benefits and drawbacks to each type of aerial asset (drone, aircraft, or satellite) used to capture your data. The entire roll out of the operation, from image resolution to frequency of data acquisition to the data processing turn-around should be customer centric. At the end of the day, using this new technology has to add value in the industry, whether it be insurance, energy, retail, agriculture, or whatever new application this new technology will impact. Otherwise, you will just be taking some really expensive, nice looking pictures.

Alim can be reached via message on LinkedIn. He is open to discussing the subject matter and exploring new opportunities in the industry. Alim previously led 3D Aerial Operations at Google, however, the views of this article only reflect Alim’s personal philosophy on the topic.