We talked to Dr. Hamed Alemohammad of Radiant Earth Foundation on applying machine learning for Earth observation to meet the Sustainable Development Goals.
First of all, how are you and your family doing in these COVID-19 times?
Dr. Hamed Alemohammad: Thank you for asking. We are largely keeping healthy by following the CDC guidelines, particularly social distancing, and wearing a mask when we go out for a walk or a grocery run.
Tell us about you, your career, how you joined Radiant Earth Foundation.
Dr. Hamed Alemohammad: I specialize in remote sensing and imagery techniques and statistical and machine learning models for geospatial and big data analytics. My interest in learning more about mapping and spatial analysis stemmed from growing up in a semi-arid region of the world and experiencing water and environmental challenges first-hand. I was passionate to change the business-as-usual paradigm. I started delving deeper into remote sensing during my masters’ studies, which focused on monitoring the temporal dynamics of water balance in a river basin using geospatial data from a NASA mission called GRACE. But during my doctoral studies, I became fascinated by the amount of data being collected by these satellites on a regular basis. I started incorporating computer vision and machine learning (ML) techniques and how we can apply them to remote sensing imagery on a large scale.
I first joined Radiant Earth in 2017 as a Senior Geospatial Data Scientist, and later in 2019, was promoted to Chief Data Scientist leading our technology team. In August of this year, I took on the additional role of Executive Director. I was drawn to Radiant Earth because of its focus on using the power of Earth observation data for social good. When you think about Earth observation, they are available globally with the same frequency and quality. But when it comes to using them to address international development challenges, there are still barriers that need to be removed to democratize access to these data and ML applications of them. Radiant Earth is on a mission to develop tools and datasets that facilitate ML applications on Earth observation.
How does Radiant Earth Foundation innovate?
Dr. Hamed Alemohammad: Training data are the building blocks of ML models. At the same time, generating high-quality and geographically diverse training data is an extensive and expensive process. Geographical diversity is necessary; otherwise, the models built on Earth observation will have biased outcomes in regions that were not represented in the training data.
We at Radiant Earth are working on innovative solutions to address bottlenecks for geospatial training datasets by a) providing a hub for discovery and accessing these training datasets, b) developing new methodologies for augmenting limited training datasets that might be available in a region for a specific problem.
Radiant MLHub is our open-access repository of training data (and soon ML models) that anyone can use to discover training data or share their own data. To establish Radiant MLHub, we have also convened the community to define a data cataloging specification that enables hosting geospatial data using a standard metadata format for increased interoperability. SpatioTemporal Asset Catalog (STAC) is a unique specification that allows all data providers to have a simple way of exposing their data through a static catalog or a dynamic API.
With Radiant MLHub, data scientists around the world can consume satellite imagery and build local ML applications to address challenges in their region.
How the coronavirus pandemic affects your business, and how are you coping?
Dr. Hamed Alemohammad: We are very lucky not to have been affected as much as other companies. But what this pandemic has shown us is the importance of Earth observation to help us understand how different elements of the Earth system interact with each other. There are great examples from the EO community working on spatial data visualization and analysis to support the emergency response to COVID-19. For example, in Kenya and Columbia, researchers used spatial data to develop a platform identifying vulnerable populations and high-risk populations while several other countries used EO to identify risk areas for transmission. . What we did to support these initiatives was to aggregate these various initiatives under the hashtag #GeoCOVID and publish monthly updates for the community to learn about the potential opportunities.
How do you deal with stress and anxiety, how do you project yourself and Radiant Earth Foundation in the future?
Dr. Hamed Alemohammad: I believe we have two types of stress, the ones that can be turned around to help you progress and be more productive, and the ones that can deter you from your work and cause distraction. Think about a deadline coming up for an opportunity, and you learn about it in the 11th hour. This can be stressful, without a doubt. On a personal level, I try to use this as a motivation and make sure I’m more productive and efficient in doing the job. And I do the same with our team and try to make sure we focus on the most important priorities.
Broadly speaking, some of these stresses are caused when we try to be too broad and shallow, which is not a good thing for a small organization, particularly a nonprofit. At the heart of what we do, is empowering those individuals and organizations who are providing solutions to decision-makers on the ground. So we are successful when they are. We aim to be focused and deep on problems that they face and provide technologies and services that can help them build local solutions for their local problems efficiently.
Who are your competitors? And how do you plan to stay in the game?
Dr. Hamed Alemohammad: As a nonprofit, Radiant Earth Foundation is a neutral organization that exists as a social good. We work with various partners and organizations within the ML for EO ecosystem. While we primarily focus on the open training data and models, we also provide data analysis and services. We see ourselves as a catalyzer and enabler, and our work is not possible without collaborating with a lot of other players in both the commercial and non-commercial sectors. From large established organizations that provide satellite imagery such as NASA to commercial companies, including start-ups that will use our data to validate the accuracy of their own models.
Our neutral status also allows us to be a convener. We bring different voices from our community together so that we can collectively remove bottlenecks that stifle innovation.
Your final thoughts?
Dr. Hamed Alemohammad: Data is fundamental to building ML applications that can produce accurate results. If you have a project that generates data, especially training data that can enable an ML application, please be open to sharing it. Open data sharing is essential to make sure your results are reproducible by others. It can empower many other users to build on top of your applications and be innovative. If you are starting with a project, do reach out, and we can help you plan on how to design your project with an open data framework.