New ways of using remote-sensing satellite data could help the banana industry prepare for and respond to threats from fungal disease and climate change
Can you imagine a world without bananas? Whether used as an on-the-go snack or baked into banana bread, bananas are one of our favourite fruits worldwide, and they are an important staple for many developing countries. But the future of banana production is under threat. Many plantations across South East Asia have already been wiped out by a lethal fungal pathogen: Fusarium wilt TR4 (also known as Panama disease). Commercial bananas have no resistance to Fusarium and there is no effective chemical treatment, hence global supply chains would be heavily affected if this disease spread across the major banana production areas in Latin America and the Caribbean. On top of this, climate change may cause some regions to become unsuitable for banana production, besides allowing new pests and diseases to spread.
There is an urgent need to prepare for these uncertainties and to assess the potential impacts on supply chains. However, bananas are under-researched compared with many other major crops. In particular, there is very little data on where bananas are currently produced. This is being addressed by BananEx, a project led by Professor Daniel Bebber (University of Exeter) through the UK Global Food Security programme. The STFC Food Network+ (SFN) has supported BananEx by providing funding for a project investigating how satellites and remote sensing technologies could be used to map banana plantations at unprecedently high resolution.
“In previous work to assess the likely impacts of climate change on banana production, we didn’t have accurate information about where bananas are grown” says Dr Varun Varma (Rothamsted Research), an ecosystems services modeller and part of BananEx’s research team. “This meant we could only make very rough estimates based on assumptions, for instance, the assumption that most banana plantations are in low-lying, relatively flat areas. We need much better data in order to make more robust and targeted analyses.”
Varun spotted an opportunity when he learnt that the European Space Agency (ESA) were making a wider range of their remote-sensing satellite datasets publicly available. These included earth observation images collected using synthetic aperture radar (SAR). This technology was originally developed for military purposes to produce maps at fine-scale, and has since been applied in geosciences and disaster management. More recently, it has started to be used for biological and agricultural purposes, for instance mapping crop types grown across Belgium.
“We realised that SAR could be perfect for mapping banana plantations” says Varun. SAR doesn’t result in what most of us think of as a ‘satellite image’: a full-colour optical photograph produced using multispectral imaging. As Varun explains, “Instead of taking a traditional photo, SAR takes a radar image. It basically creates an image of the texture of the land surface.”
In SAR, the surface is actively illuminated with radio waves. The amount of radar signal that is scattered back depends on the surface’s texture, with smooth surfaces (such as water) reflecting very little. Different crops have unique structural properties (for instance, the large, upright leaves of banana plants), which allows them to be detected and differentiated.
“A particular benefit with using these datasets from ESA is that they are routinely collected as their satellites orbit the earth, so new images are made available at least every 2 weeks. Consequently, we can be sure they are up-to-date” says Varun. Another key advantage is that, unlike multispectral imaging, SAR is not sensitive to cloud cover, and so provides an uninterrupted signal of the earth’s surface over time.
Despite these benefits, using SAR presented a new challenge for Varun, whose previous work mapping forests and savannahs has been mainly based on multispectral imaging. “Using SAR required a different way of thinking to understand what the data was telling us. Instead of asking ‘What is the colour of the canopy?’, we are now asking ‘What is the structure of the surface?’”
Naturally, working out how to use the data and develop it into an initial mapping model involved a period of trial and error. This in itself could have posed a problem, with each image typically covering 20,000 km2 and using over a gigabyte of data. Combining multiple images to produce a map is simply far beyond the processing power of a standard desktop computer. “This is where it was a real advantage to draw on our SFN collaborators, particularly Professor Seb Oliver (University of Sussex), who helped us access the high-performance computing facilities at the STFC” says Varun. “It meant we could trial ideas quickly, see if they didn’t work and start again if necessary.”
The project also partnered with CORBANA, Costa Rica's National Banana Corporation, who provided ‘on the ground’ information about where their banana plantations were located. This was used to test and refine the initial model until it achieved an accuracy of 98% at a resolution of 10 m. BananEx has since produced a preliminary map of banana plantations covering large parts of Central and South America, and the Caribbean.
Banana plantation distributions in parts of Panama, India and Colombia (credit: BananEx).
Having demonstrated the high accuracy of their model, Varun hopes that in the near future it could be developed into a continuously monitoring system. “Agricultural landscapes change very quickly, so we can’t afford to produce ‘snapshots in time’ every 8-10 years. For instance, extreme weather events can badly affect banana plantations, but we will only see how quickly they recover if we have continuous monitoring.”
“I’m very grateful to the SFN for enabling this project. It has given me the space and opportunity to explore a new kind of data for me and it has been a good example of ‘learning by doing.’ Being part of the network has also introduced me to people with very diverse backgrounds and expertise- including physicists, chemists, biologists and data scientists – who are all working together to improve food systems” Varun concludes.
May 2021 - Caroline Wood, University of Sheffield