Satellite images play a major role in harnessing sustainable agriculture. Today, we meet Jyotsna Budideti, co-founder and CEO of SpaceSense, a Paris-based startup that combines the power of satellite imagery and AI to help monitor 4 million acres over 10 countries daily.
Caroline Lair: Hi Jyotsna, what has been your inspiration to become a machine-learning engineer and take the entrepreneurial road?
Actually, when I was young, I was obsessed with becoming a pilot, but, when I turned 15, I started to need glasses, which automatically disqualified me to be a pilot. Since I won’t be able to drive these things, I’ll learn how to build them and I’ll study space engineering.
During my Bachelor, I had the opportunity to work at TIDE, a nonprofit based in Bangalore working on technological solutions that help the rural market, the villages, and the under-served. I was in charge of building a modular biomass stove for dissemination and this was a key moment for me as I really enjoyed working on a product with a user-centric focus. In addition, I’ve realized that I need to work on something that can be useful for people, that solves their problems right now and not in 10 or 20 years.
So after my Bachelor, I decided to do a master's in Innovation and Technology at Polytechnique in France, to have that mix of engineering and entrepreneurship. That took me to Berkeley as well for one semester in California, where I started learning about machine learning. It just opened up the possibilities of so many innovations, so that was super exciting. That made me realize that it was worth spending another year studying only that and I applied for another master’s in Applied Mathematics for Machine Learning.
At the end of my master’s degree, I started a job at Airbus as a Data Scientist, then a Machine-Learning engineer. Among others, I was doing research on a project, named ATOLL, around complete autonomous flight, from take-off to landing. And that was the time when I started thinking about Space Sense. The idea is that satellite images are basically one of the greatest technologies to understand and mitigate climate change. However, it is very complex to extract information from these images. The idea behind SpaceSense is to make satellite image analysis accessible to all industries, starting with agriculture.
When I was sure about the approach and the business model, I started looking for a co-founder and that’s when I found Samy. Then I quit my job and went full-time on SpaceSense.
Caroline Lair: How remote sensing is changing agriculture and helping farmers to work more sustainably?
First, remote sensing brings an affordable and scalable solution for precision agriculture. Indeed, every generation comes with new technology, ground sensors, IoT, drones, and now remote sensing. With ground sensors, we had the issue of scalability, and the same for the drones. Not everyone could afford these technologies as well. As a result, that limited the possibility of using precision agriculture in only a few scenarios. What we were able to do with remote sensing is complementary to these existing technologies but it does also enable affordability and scalability, not only to regions but to the world.
That is crucial if we want to make agriculture more sustainable because the world population is constantly growing, meaning we need to grow more while maintaining the quality of the soil and not hurting the environment and biodiversity. That means we need to use fewer resources, less water, and fewer fertilizers and this is what precision agriculture helps farmers do. I can give you an example. With satellite images, we can look at how the crop is growing everywhere in the field but also how much water is available, or what are the different types of soil in that type of field. So, the farmer is able to combine this information and use it to make a decision on optimizing how much and where fertilizer should be applied in their field. That is super important because, in the end, the goal is to increase yields, but you have to use fertilizer and at the same, if you do not optimize it this way you will end up using much more than what is actually required for your yield.
Second, it is also a way to tackle climate change, as the climate is changing way too fast and is much more unpredictable, so any kind of tools you can use to adapt and inform farmers so they can adapt quickly in a way that does not hurt their profitability, and beyond the whole economy, is needed.
We are also enabling businesses within the agriculture ecosystem to better advise the farmers they’re working with. For example, they can advise farmers about the type of seed variety that is the most suitable given the changes that have happened because of climate, the soil, and different dimensions, then moving forward, what type of crop variety, what fertilizer, and what kind of protection to use. Because today, it is hard to rely on farmers’ intuitions, and being able to provide this type of information to the experts will also allow them to make better advice to the farmers, and consequently empower them in their own partnerships with the farmers.
Finally, another important point is in insurance and the availability of capital. For example, there are a few companies that are trying to provide finance to smallholder farmers that never had access to any kind of capital so far. These farmers usually have a very high debt cycle and it is very hard for them to get any type of loan. Remote sensing can be used to measure and model the risks of these smallholder farmers and make them eligible for loans, changing their lives completely, but also, at a larger scale, the production of countries that have many small farmers.
Caroline Lair: One of the promises of SpaceSens is ‘democratizing space intelligence’, so, talking about developing countries, is it really affordable for them, are there any specificities in the way these farmers use remote sensing in addition to facilitating their access to credit?
It is a pretty good question. The technology is available but the way it is used is very different. The example I gave you with the fertilizer does not apply to developing countries for instance, because their farms are too small. They do need to use fertilizer in that space.
There are still challenges when you talk about the common use cases of remote sensing in agriculture because of the way farming is done. To give you a good example, we have a solution for regular crop monitoring that applied for developed countries but not that much for developing countries because of different agricultural practices. For instance, especially in India, farmers have small spaces and they grow multiple crops in it, so it is good for the soil and the environment but it does require a much higher resolution of insights. But then if the farmer has one or two hectares, they don’t really need precise monitoring, they can go and walk to the farm every day if it is closed enough.
Now, if we contemplate a group of farmers and not individuals, it starts to become more interesting, especially because remote sensing can help them to adapt to weather forecasts and changes resulting from climate change. Another important aspect is the soil type and availability of water. We have very good repositories of soil types in developed countries because they have the money to send individual soil surveys across the countries. But it is not something developing countries can afford so there is a missing piece of information that is very critical to know in order to choose the right type of crop, when to plant, and how you manage given the soil that the farmers are working on, and the nutrients available on that soil. There is a missing piece here and this is when remote sensing is coming into the picture.
Caroline Lair: What is the positioning of SpaceSense and how do you leverage machine learning and transfer learning?
We are btob technology providers. Any time any business can benefit from some information that is derived from satellite data, we should be the ones providing it. We don’t provide raw satellite images, but direct insights that they can integrate into their own digital ecosystem and use either for decision-making, new products, cost reduction or provide them to their customers.
To do so, we are building a machine learning system that is very much specialized for satellite imagery. Satellite imagery can be multispectral, hyperspectral data, or radar among many others. We are in the process of specializing our machine learning system for each of these, meaning we are developing model architectures that can derive different types of information with different types of datasets. It means we are also making it available as a product, in a production environment, which is quite a challenge as 99% of ML projects usually don’t make it to production mainly because it is very complex to do. We are doing this with satellite imagery on top, which makes it even harder because of the size and magnitude of the data itself.
Transfer learning is one of the approaches that enable the development of more use cases very quickly while providing operational products that can be customized by the user to their own specific region or specific use case. To give you an example, let’s say there is a product that makes a prediction on the yield of wine, and it was built based on a huge dataset. I am a user, working in France, and I need this product to work in every field and in every variety in France with a minimum of 90% accuracy. What we are able to provide in production is the capacity for the consumer to automatically adapt the solution to their own field, so they know how accurate and reliable it is for their specific fields and can integrate it into their decision-making.
At the same time, they can create more value with the data they have, like all the historical business data they have just laying around, they can use this information to improve accuracy and bring more value to their own products by using that to improve the accuracy, up to 95%, even 98% and further. That is how transfer learning is used to adapt their solutions at a more local scale. So in a nutshell, we are able to create custom-like solutions but at scale.
Caroline Lair: When it comes to remote sensing, what are the biggest challenges you see for the next 5 years in this industry?
With remote sensing and especially with AI, it did completely open up the scope of use cases we can solve but also created a question mark: what can be solved? It is picking up in terms of research work but the industry is still not mature enough, so we are still looking at which use cases can be possible and it is a question mark.
Also, there is a huge knowledge gap in industries that do not usually use remote sensing and the big challenge will be, for companies like us, that are trying to reach out to these industries to be able to make it mainstream. That is very very hard as they still look at it in a very skeptical way.
There is also a challenge in terms of business models. Today, what we do is technologically feasible but the way the business models are is not feasible, and pricing is often not feasible, which is the scenario for most use cases that require high-resolution solutions. It is changing, hopefully, sooner or later, as more and more use cases become products rather than research papers.
The biggest challenge will still be the availability of data that can be used for this machine-learning system to build in the first place. Because of the way of the industry, a lot of data is proprietary and is not easy to get and if you want to do it yourself, it is going to be very expensive. That is one of the biggest challenges.
Caroline Lair: How do you navigate the lack of data on a daily basis?
We are taking it case by case, it is very hard to generalize this. Depending on these use cases, we find partners from whom we can find a way to benefit from each other resources and who also have the data we can use. For some other use cases, we may need to take the time and effort to actually do it ourselves. What helps also is close collaboration with research institutions where we can collaborate on this kind of effort.
Caroline Lair: What’s in the roadmap for the next 12 months? In which other areas do you see your business expanding in the future?
I believe that Satellite imagery tech with AI will be one of the key technologies to address climate change and the challenges of this generation. And the best way we can contribute to it is by making it truly accessible for everyone.
Until now we have been focussing on providing ready-to-use insights as APIs to enable operations for agriculture businesses. We now want to give the power of satellite imagery technology to organizations & businesses to build solutions that have the most impact on them. Through our toolbox for Satellite imagery-AI, we want to enable non-traditional users to leverage this technology, and combine it with their expertise to build truly impactful solutions without the requirement of any expertise in satellite data or AI scaled up by the cloud.
Our next year is dedicated to bringing this toolbox to the users to speed up the development of solutions for the environment as well as add additional capability to leverage petabytes of satellite data to solutions that will make a difference.
Caroline Lair: Which kind of talent are you looking for?
At the moment we have the team we need. We plan to expand our team with different kinds of profiles, with computer vision engineers, remote sensing experts, backend developers, DevOps engineers, and more later next year. We also definitely need people that can work with the customers for marketing and sales and product managers.
ABOUT THE AUTHOR
Caroline Lair is the CEO and Founder of The Good AI. She is also a co-founder of the Women in AI non-profit. Her academic background is in International Relations, with a degree from Université Jean Moulin (Lyon III). And a business management degree from Emlyon Business School.