Satellites to Monitor Fields

Estimated Reading Time: 4 minutes, 7 seconds.

• NASA able to see increase in Ukranian wheat exports using satellites

• Basic Machine Learning with satellite data.

• McKinsey report on Satellite usage for the future

As the commercial space becomes cheaper, we are seeing the birth of what could be a trillion dollar industry. The cost per kilogram to launch into low earth orbit is plummeting. Satellites are improving internet access, and creating global transparency. They could also bring the next age of exploration

Why satellite data?

Computer vision and satellite data are a match made in heaven. Only looking at satellite data isn't enough. Advances in Computer Vision let us begin to automate parts of this process and get more from satellite images. Computer Vision lets us measure, and detect objects by finding patterns in images.

What changes are pushing this?

Earth observation from satellites has become popular from 2 major advances. The price of satellite images and their computation has fallen.

Satellites stream massive amounts of data from space to processing centers, for analysis. This analysis is possible due to a few changes. Decreases in the cost of compute with cloud services from Amazon, Google, and Microsoft have made this even easier. Research in fields like Computer Vision and Machine Learning have had enormous impact on how well we can utilise data from space.

The cost of a launch to space is plummeting, which has started a boom in the number of satellites in space. Private space companies have found new ways to get the most out of rockets and shuttle items into orbit. This has let private industry launch satellites and capture proprietary data.

National space agencies, like Copernicus from the European Space Agency, have also made data available. It makes Earth Observation Data accessible for users. This data is invaluable to the those that are helping farmers improve their yields with improved agricultural data.

Farmers are receptive to precision agriculture. It helps them observe, measure and respond to the changing conditions on their fields. It also helps them address the dwindling labor supply on farms. Some farmers claim yield gains of up to 30% when adding precision agriculture to their farms.

How is this useful for Agriculture?

Satellites can help predict crop yields, which addresses the systematic threat that climate change is to the food system. If we can predict yields, we can build more resilience into the system.

Much of the data Satellites collect go beyond what the naked eye can see. Satellites offer multi-spectral Earth observation, which uses different waves to get information about the earth. This can be visual light, or radar to provide valuable data about the land.

This creates rich datasets we wouldn't otherwise be able to understand, since it offers insight beyond what we can see. Instead of a farmers observing their fields through observation, they can get a more complete picture. Including things they can't see, like water and carbon cycles. With this information they can make decisions on a site specific basis, and act on a smaller scale. Ultimately, the land ends up getting treated far better.

For conventional, broad based agriculture this is of massive importance. The use of data and precision agriculture allows farmers to adjust to the conditions of their fields beyond what they can see. It allows for more automation across farms, and uses technology to identify solutions to classic farming challenges.

How complex is this stuff?

To test this out, I decided to run an experiment. Years ago, UC Merced open sourced a dataset of tagged satellite images, which is perfect to create a multi label classification model. Within a few hours and some very basic machine learning I was able to build a model that performed... surprisingly well. With 86% accuracy on a test dataset.

With domain knowledge, the right data and a bit of machine learning you can very simply build a model that represents the real world... pretty well. See my code, here.

What challenges exist for the use of satellites in agriculture?

As the famous statistician, George Box, said "every model is wrong but some models are useful." None of this works without an understanding Ground Truth. To work around this issue, some companies are using farm level data like soil tests to train their models.

Farmers have a better understanding for the context surrounding their land, and this can't be understood by computers alone. There are partnerships, decisions and politics that make decisions on farms extremely difficult to reach.

Satellites also face challenges. They cannot be refueled with propellant while they are in orbit. This means that they can reach an end of life and must be replaced. While this functionally shouldn't have much impact, it does increase the cost of working with satellites.

Final thoughts

Overall, satellite data will become a standard piece of broad based agriculture. The data they produce will help farmers make decisions that are in line with their land as well as their bottom line. While satellite data isn't a panacea, it is going to become one of the most important tools in the reinvention of farming.