The moving boundary inside AI’s forecast of saltier soils

Water can leave a field while much of its dissolved salt stays behind. FAO’s irrigation guide describes the balance: irrigation adds salts, crop water use and evaporation concentrate them, and water draining below the roots can carry some away. Salt accumulation depends on all three processes.[2]

The remaining water also becomes harder for a plant to extract. Salt adds an osmotic burden: roots must expend more energy to absorb water from a salty solution. Two soils can therefore contain the same amount of water while offering different amounts that a plant can use.[2]

Let the map’s boundary move

A Nature Communications study published as an accepted manuscript on 19 October estimates how this problem changes across expanding drylands. It combines 51,835 unique soil-salinity observations from 2004–2023 with environmental and human factors, using XGBoost machine-learning models driven by 12 climate models.[1]

Drylands are identified by rainfall relative to potential water loss through evaporation and plants. The researchers train separate models for dryland and wetter samples, updating which model applies as each mapped location’s climate changes. They predict salinity at several depths and combine these into the upper metre of soil.[1]

Under the high-emissions SSP5-8.5 scenario, dryland saline-soil area is projected to be about 8% larger in 2100 than the 2004–2023 baseline. Here, “saline” means a saturated-extract electrical conductivity of at least 2 dS/m.[1]

Historical testing used randomly held-out sample points. Future satellite-derived inputs remain at their historical averages, a strong assumption acknowledged by the authors. The projection describes a conditional pattern; it cannot establish an individual field’s future condition.[1]

Report an error or suggest a correction