Stopping soil loss
Erosion, compaction, salt build-up and lost organic matter destroy farmland soil far faster than nature rebuilds it.
Open in the interactive tree →A study from Bonn's Center for Development Research and IFPRI found that the quality of 33% of pastureland, 25% of arable land and 23% of forests deteriorated over 30 years, on land that 3.2 billion people depend on. Estimates of erosion range from about 36 billion tonnes a year from water alone (Nature Communications) to about 75 billion tonnes for all erosion, 13-40 times the natural rate. The UN desertification convention (UNCCD) says soils erode up to 100 times faster than natural processes replenish them.
As of October 2026
UNCCD reports that 131 of the world's 196 countries have pledged or are working to stop land degradation by 2030, and projects that intensive agriculture will cost the land 32 gigatonnes of carbon between 2015 and 2030. It says halting degradation needs three things at once: preventing new degradation, reducing existing degradation through sustainable practices, and restoring degraded land.
What is missing
- Cheap, standard soil-health measurement everywhere
- Financial incentives that reward farmers for soil carbon and structure
- Secure land tenure so farmers invest in long-term soil
- Methods to speed up soil formation or rebuild eroded soil
- Enforced land-use policy
Becomes possible once solved
- Farmland that keeps its yields for centuries
- Soils that store carbon instead of releasing it
- Fewer floods, dust storms and silted-up reservoirs
Open steps
- Cheap soil-health measurement High AI leverageReplace lab soil tests with spectra, satellite and sensor data that give soil carbon, structure and salt at field scale and low cost.
- Verifying soil-carbon change Medium AI leverageMeasure real changes in soil carbon over a few years against large natural variability, so payments can rest on evidence.
- Erosion and degradation mapping High AI leverageDetect gullies, compaction and salinisation early from satellite and drone images, and predict where erosion will hit next.
- Rebuilding eroded soil faster Low AI leverageFind management and amendment recipes that speed up soil formation or restore structure on eroded land, where nature needs centuries.
Where AI could help
Medium AI leverage. AI makes soil measurement and erosion mapping far cheaper; incentives, land tenure and enforcement still decide whether soil is protected.
- Turn cheap infrared and satellite data into soil carbon and structure maps instead of lab tests
- Detect gullies, compaction and degradation from satellite and drone images over whole regions
- Predict erosion risk per field to target terraces, cover crops and buffer strips
- Verify soil-carbon claims so payments to farmers can rest on measured change
Shown so far
- In February 2017, SoilGrids used machine learning on about 150,000 soil profiles and 158 satellite-based covariates to map global soil properties at 250 m resolution. source
- In June 2021, SoilGrids 2.0 added quantified uncertainty to its machine-learning maps and showed that high latitudes especially lack soil observations. source
Prerequisites
- The plough~3500 BC
- Soil science1883
- Green Revolution1960s
- Climate models1967