Predicting Crop Yields with Satellites

Agriculture is one of the oldest means of producing food – and has always been vital to human civilisation. Poor harvests can be the difference between the rise or fall of an empire. Historically it was impossible to know if it would rain later that day, let alone predict the weather or crops in advance. All humanity could do was make vague assumptions based on animal behaviours and pray to their gods.

Today we forecast the weather to within one degree and can see crops growing around the world, predicting yields and tracking harvests as they happen.


Satellites have revolutionised the farming industry – from weather data enabling predictions to near-perfect mapping via GPS.

Satellite imagery is another way in which technology is transforming farming. This data is now being used to understand, monitor, and predict crops.

By fusing together images from different satellites and using supervised machine learning highly accurate predictions about crops can be made. Meter by meter, field by field, farm by farm data is collected, process and analysed to know exactly what is being grown, where and how much.

Farms in countries such as Brazil are ideal for satellite monitoring, with many of them over 1,000 acres for a single crop. This scale and simplicity enables companies to quickly measure the size of fields and predict yields.

Challenges of Scale

Countries such as Thailand are more challenging. This is for three main reasons: Their farms are small, complex, and they are routinely covered in the cloud.

The average size of a Thai farm is just 10 acres – 50 times smaller than the average in Brazil.

Calculating the area of crops in thousands of small fields is a lot harder than the same calculation across several big fields – whether you are on the ground or working with satellites. The farms in Thailand are also complex – they have mixed crops – growing sugar, rice, and cassava alongside each other. The calculation of area becomes even more challenging as the crops can change every year – with no fixed boundaries. Simply put – small fields of sugar, rice and cassava all look very similar to a spacecraft 700km away travelling at over 25,000 kilometres per hour.

Finally – to make Thailand even more challenging is the high-volume clouds. Optical satellites cannot see through the clouds. These factors combined means that predicting crops in areas such as Thailand and India has, historically, not been very accurate.   All of these issues are now being resolved as technology, computing power and analytics all improve.

The issue of cloud cover can be negated by using more optical satellites – to collect more data. With more frequent images the “gaps” in the cloud can be combined to produce a single image. This means that the cloud is effectively removed.

The cloud can also be “ignored” by using radar satellites –  as radar will see through the them. Radar data alone is not enough, but when fused with optical images then entirely new data sources, of much higher quality, becomes available.

Scale of Data

Using more data –  optical and radar – results in vast quantities of data being collected –  hundreds of terabytes for a single country. Fortunately, due to the increase in computing power, it is now easy to collect and analyse this volume of data.

Once “new data”, without cloud cover, is created then far more advanced analytics can be conducted. Modern satellite analytics will examine every “pixel” of a country, every few days, looking at the “signals” of multiple crops. The crop signals – how they grow, how their colour changes and how their chlorophyll levels vary – combine to create a unique “fingerprint” for each crop, in each country.

This fingerprint can then be used to identify crops across a country  – instantly detecting the type and size of crop– whether it’s detecting sugar in a 5,000-acre mega farm in Brazil or rice in a 5-acre micro-farm in Thailand.

As a “pixel” is 10 meters by 10 meters and there are multiple signals for any crop, and data is collected every few days – this means that a country is examined at extraordinary detail.

The analytics for a single crop, in a single country, will involve trillions of calculations.

This incredibly high number of calculations is then validated against drone data and historical data. The result is the ability to predict crops with over 98 percent accuracy, even in the most complex of regions. As the available data increases and analytics improve so will the ability to understand and predict crops, supporting farmers and governments around the world.