Turns out artificial intelligence can get its hands dirty: UK and German researchers have just used machine learning to analyse how soil salinity has changed across the world.
Their paper, published in the journal PNAS, found that an area significantly bigger than the US has been salt-affected over the last 40 years.
The amount of salt and sodium in soil varies over space and time, especially in the top 30 centimetres of the ground, which are more exposed to climatic conditions. These variations are typically due to natural events like droughts, floods or erosion, but are now increasingly driven by human activity like fertilisation, irrigation, and the effects of climate change.
High salinity isn’t great news for the land. It can affect microbial processes, damage plants including crops, and in turn impact human livelihoods.
“Soil salinisation is one of the main land-degrading threats influencing soil fertility, stability, and biodiversity,” the researchers note in their paper.
It is therefore vital to understand how and why these variations occur. But until now, most soil salinity studies have been localised, both in space and time.
This study – led by Amirhossein Hassani from the University of Manchester – looked at the problem on a broader scale than ever before. The team used machine learning techniques to help make sense of climactic, topographic, soil and remote sensing data from across the world, which comprised nearly a quarter of a million soil measurements from 1980 and 2018 at different latitudes, longitudes and soil depths.
The analysis reveals that 11.73 million square kilometres have been salt-affected in the last 40 years – an area 20% bigger than the US. Of this, 0.16 million square kilometres were croplands.
In general, the net salinity changes were geographically highly variable, but the continents with the highest salt-affected areas are Asia (particularly China, Kazakhstan and Iran), Africa and Australia. In these arid and semi-arid climates, evaporation dominates over precipitation and so tends to allow salt to concentrate in the soil.
The research also found that Brazil, Peru and Sudan have the highest annual rate of increase in soil salinity.
But although existing research agreed that salt-affected areas are expanding, these new findings suggest that the total area of salt-affected soils has been highly variable, showing both decreasing and increasing trends on both national and continental scales.
Looking to the future, these results are useful as they now allow researchers to predict the likelihood of soil salinity in particular areas, and therefore help manage land and soil resources.
The authors note that “this information can also be valuable for enhancing our understanding of terrestrial carbon dynamics, food security and agricultural modelling, climate change impacts, water resources and irrigation management, and the efficiency of organic/inorganic reclamation practices.”
Samantha Grover, a soil scientist from RMIT University in Melbourne who was not involved in the study, says the long-term predictions made by this study can help us better manage Australia’s soils.
“Salinity and sodicity are serious ongoing problems in Australia that compromise our soils’ capacity to support food and fibre production, as well as causing onsite and offsite reductions in the ecosystem services that our soils provide,” she explains.
“This is exciting new research that has great potential for Australia as we seek to restore our soils to enhance our food security and contribute to climate change mitigation.”
The method used by Hassani and colleagues – using machine learning to analyse the data and produce models – could also be applied more broadly to study the variability of other dynamic soil properties, such as soil nutrients, organic carbon content, and pH.
“The results have significant implications for agroecological modelling, land assessment, crop growth simulation, and sustainable water management,” the authors conclude.