AI to Use Satellite Imaging Tech to Predict Food Crises Before They Happen
Food, especially food shortages, remains a controversial topic in a lot of economic forums. But it appears one of the most important aspects of human ecology is about to take a technological revolution.
A company aims to use machine learning to analyze satellite images to predict food supply movement months in advance compared with current methods used by the U.S. government.
This could predict and prevent food crises before they even happen.
According to Motherboard, it can be remembered that economist Thomas Malthus argued that the increasing standards of living seen around the globe are unsustainable because the population will inevitably outgrow food production.
Interestingly, others argue that Malthus did not take into account technological advancements that can make production easier.
Sadly, the World Bank is finding out that the world is set to 25-percent of crop yields to climate change despite the need to produce 50-percent more food to keep up with the population.
The burden now is to make technology to forecast food crises in the coming decades, and Descartes Labs' latest machine learning tech can do the trick.
According to Descartes Labs, their machine can collect 5TB worth of data from NASA, ESA, and other commercial satellites. If combined with data such as weather forecasts and prices of agricultural products, their new machine can track and forecast food supply changes with astounding accuracy.
The platform must first process the imagery from the sources to make it uniform. The initial method was to train the machine learning platform on a petabyte's worth of satellite images on a supercomputer with 30,000 processing cores.
It can be noted that the fastest supercomputer has three million cores.After thousands of models, the algorithm can finally differentiate individual corn fields on its own. But that's just the beginning as the platform can now determine whether the plot of land is being used to grow corn or soy depending on how much light is on the surface of its special field.
With this, it can efficiently monitor production levels in a given area, whether it's on a county, state or even country.
Mark Johnson, Descartes Labs CEO, told Motherboard that initial tests had forecasted U.S. corn production on Aug. 6, 2015. They were able to come within 1.9 percent of the final report on corn yield five months before the USDA report was released.
For Descartes Labs, analyzing yields half a year after the fact is too late to stave off a crisis, meaning they need to have accurate analyses in real time.
The platform has been so far applied to corn and soy yields in the U.S., but the algorithm has been able to produce hyper-detailed forecasts on a weekly basis for all 3,114 counties in the state.
For Johnson and his colleagues, the ability to have fine-grained data on agricultural production can help make the food supply chain more efficient and be even more accurate.