This is the first in a series of three posts examining how organizations are adopting Earth-observation (EO) data in pursuit of their goals. The series will review the current state of the industry, consult with traditional academic models of technology adoption, and offer a new framework that can guide firms who want to climb the EO learning curve.
Machine learning has been changing our world for years and continues to do so in ways we humans are still learning. Across industries, the application of artificial intelligence to massive datasets is revealing insights that are changing the way business gets done.
For example, manufacturers used to wait until their machines broke before they fixed them; now, they proactively scan relationships between streams of data to predict component failure. They fix the break before it happens, and in so doing ship more product than they could without machine learning.
The most powerful enabling force in the rise of machine learning has been cloud computing. Less than ten years ago, most computer users stored and utilized data on the machine in front of them; the notion of a “local” device hadn’t come close to entering the common vocabulary. Today, the cloud is the default. Cisco predicts that 94 percent of work will be done in the cloud by 2021. Between those points was explosive growth in the cloud market.
Cisco Global Cloud Index: Forecast and Methodology, 2016-2021 White Paper
Contents What you will learn Forecast overview Regional cloud readiness Top seven data center and cloud networking…
Perhaps no company has profited as handsomely from this explosion as Amazon Web Services, who currently holds a majority share of the cloud services market. Amazon has succeeded in harnessing the exponential growth of cloud computing, attested by the 8x growth in revenue they’ve experienced in the last five years.
For every Amazon, there have been countless firms who failed to adapt. Many enterprises waited too long to deploy cloud technology. Others rushed in ad hoc without a cohesive strategy, leading to fragmented environments. Still other companies never built the connections across cloud instances that allow systems to scale.
Similar patterns are emerging with respect to the production, processing, and interpretation of Earth-observing (EO) satellite imagery. Over the last few years, the amount of Earth-observing satellites and the resulting data flows have exploded. Thousands of EO satellites of varying sizes will be launched in the next several years; petabytes of valuable data will be streamed down for analysis each month.
Growing Data for a Shrinking World
Astraea’s EarthAI platform helps you navigate the oncoming EO data deluge.
What sparked this explosion? And where is it headed? The remaining sections of this article will examine those questions.
Convergence of Trends
An earlier reflection on our blog identified three trends that we believe will be critical enablers of EO technology:
- Lower costs to produce, launch, and operate EO satellites will allow satellite operators to produce more data.
- Developments in cloud computing will make storing and accessing EO data cost-effective for companies and individuals, not just governments.
- Machine learning tools will become increasingly applicable to EO data.
To that core list, I’ll add two more: computer vision and the commercialization of space.
Advances in parallel technologies like autonomous vehicles and facial recognition have driven the rapid evolution of computer vision. The same (or at least similar) Convolutional Neural Networks and other processing algorithms used in these cases are similar to those deployed by remote sensing scientists making sense of satellite imagery.
Object detection, image classification, and pixel-based segmentation are beginning to prove as useful in utilizing geospatial data as they are in helping to drive a car.
The growth of computer vision is now converging with a new era of space technology, sometimes called “New Space” or “Space 2.0”, where commercial enterprises principally drive innovation. Since the U.S. Commercial Space Launch Act of 1984, private companies have participated meaningfully in humanity’s quest to reach beyond our atmosphere. Today, companies like SpaceX can launch, and re-launch, private vehicles, and NASA has announced plans to commercialize the International Space Station.
NASA releases ISS commercialization plan - SpaceNews.com
WASHINGTON - NASA unveiled a multi-pronged effort June 7 to increase commercial use of the International Space Station…
A critical force in the unfolding of this new era has been the rise of corporate and venture-backed Earth-observing satellite operators. NASA and ESA have operated Earth-observing satellites for decades for scientific purposes, and data from these public providers are freely available. These data are often limited, though, in their instrumentation, spatial resolution, and temporal revisit rates. To fill this gap, defense contractors like Airbus, established players such as Digital Globe and Planet Labs, as well as a slew of venture-backed startups, now offer private imagery that provides better spatial, temporal, or spectral resolution than government data.
This has led to a massive proliferation of data. As our CTO Daniel Bailey shared, at the end of 2018, NASA had cumulatively produced 27+ petabytes of Earth-observing data since the first Landsat mission in 1979. NASA alone will generate close to 27 petabytes of data in 2019, and corporate and startup operators will deepen those volumes.
Significantly more EO data will be produced this year than in each year of human history, combined. We are reaching a critical point at the base of the hockey stick.
The State of the Industry
We think of the EO industry in terms of the value chain that delivers insight to the market. This value chain can be segmented in the following steps:
- Produce pixels: launch and operate Earth-observing satellites.
- Ingest & store data: manage the vast stream of downlinked bits.
- Provision cluster: set up the cloud infrastructure needed for computing.
- Prep & fuse data: process raw data into analytics-ready data in a tabular form.
- Create AI models: train machine learning models for a given task (e.g. identify utility-scale solar farms) over an Area and Time of Interest with known attributes.
- Deploy models: score models on new Area and/or Time of Interest.
- Insight: interpret results of models to better understand the Earth.
The current economics of the industry has led firms to specialize in one part of this chain. Because of the complexities involved in steps five through seven, which differ based on the industry and use case, insight-centric firms have focused on answering a specific problem for a specific audience. Orbital Insights’ GO product suite proves a potent example. Satellite operators have taken note of this trend, and many are now “pushing toward insight” in the value chain, offering a proprietary portal (e.g. Digital Globe’s GBDX) for customers to analyze data themselves.
Atop this value chain are these firms’ customers. Increasingly, organizations across sectors are eager to explore the power of geospatial data.
As is typical with emerging technologies, each organization faces a unique set of challenges in working with Earth-observation data. Some struggle to identify relevant use cases, while others lack the machine learning and/or EO expertise to build useful models. Still others face technical challenges in handling these massive data sets.
These challenges differ across industries — use cases for big agriculture and asset management are far better-defined and observed than use cases for impact investment. These challenges also diverge within a given industry — some firms’ leaders are more willing to adopt emerging technologies than other firms’ leaders. Even within a single organization, EO expertise and tools are often scattered, leaving some groups able to capitalize on EO opportunities while others look on.
Projecting the Path Forward
At Astraea, while we do not have a crystal ball or perfect foresight, we do have an opinion on how converging trends will unfold.
We’ll highlight three broad trends that, coupled with the influences mentioned above, will accelerate the extent to which individuals and businesses adopt EO.
First, individuals are now using satellite data ubiquitously, navigating the Earth using applications like Google Maps. Less than a decade ago, most people didn’t think about their world from a birds-eye view. Today, Google Maps has over a billion users globally.
Second, our planet is in peril, and people increasingly care about making a difference. The recent rise in the prevalence and severity of natural disasters and the increasingly dire warnings from groups like the United Nations are broadly raising awareness of the need to enact change. EO data, a reflectance of the state of our planet, provides an unrivaled means of understanding where and how that change should occur.
Finally, supportive economics bolster these trends. The cost of compute will continue to decline as cloud technology advances. The cost of pixel procurement will decline as more private companies achieve scale and expertise in launch and operation. The cost of analysis will decline as EO algorithms become better-known and increasingly commoditized.
Combining these forces, we foresee a burgeoning need for people, organizations, and companies to investigate the planet, with the basic familiarity and supportive economics required to do so.
The convergence of these factors, added to the technological influences mentioned above, in today’s data-rich environment, has led organizations to explore the power of EO data.
Interestingly, the adoption of EO data by the market has been anything but uniform. As mentioned above, the challenges of harnessing EO may differ across industries, across organizations within an industry, and across individuals or groups within an organization.
In the next post in this series, we will examine these issues in the broader light of the EO adoption curve. We will review traditional models for technology adoption and demonstrate why the adoption of EO data is breaking these paradigms.