Options
Global-Scale Prediction of Soil Salinization under Different Climate Scenarios Using an Advanced Machine-Learning Technique
Publikationstyp
Conference Proceedings
Date Issued
2020-12
Sprache
English
Author(s)
Institut
Citation
American Geophysical Union Fall Meeting (2020)
Contribution to Conference
Publisher Link
Soil salinization, referring to high accumulation of soluble salts in the soil, is one of the serious land degrading threats, which can undesirably affect the soil resilience, fertility, stability, and biodiversity. Developing predictive and quantitative tools to identify the locations of saline soils and their spatio-temporal variability are crucial for devising sustainable land and water management plans. However, previous studies primarily focused on the past or current trends in spatio-temporal variability of the soil salinity. A comprehensive investigation of the future trends of soil salinization, specifically in the face of future climate uncertainty, is rare partly due to the complexity of the processes influencing soil salinization at multiple scales, which motivated the present research. In this work, we used soil profiles data, a set of purely spatial predictors stacked from the soil and land topographic properties, and a set of spatio-temporal predictors derived from the output of available Global Circulation Models (GCMs) in the Fifth and Sixth Phases of the Coupled Model Inter-comparison Projects (CMIP5 and CMIP6; Eyring et al., 2016; Taylor et al., 2012) to develop a predictive model for projecting the response of the primary (naturally occurring) soil salinity to climate change by the end of 21th century. We linked the experimentally measured soil salinity information obtained from the input soil profiles data and predictors' values (auxiliary data used for prediction of the soil salinity) through application of the tree-based Machine Learning algorithms to predict the soil electrical conductivity (ECe) as a representative of the soil salinity at different locations, times and soil depths. Overall, nearly 72% of the variability of measured ECe could be described by the prediction of the trained models (10-fold cross-validation R2 = 72.49%). The developed models were then applied to new projected spatio-temporal predictor data obtained from the output of GCMs to annually predict the average of soil ECe to the depth of one meter from the surface between 1905 and 2100 for the drylands of the world (lands with an aridity index ≤ 0.65). By comparison of the 1971 - 2100 and 1961- 1990 periods and under different greenhouse gas concentration trajectories (RCP 4.5, RCP 8.5, SSP 2-4.5, and SSP 5-8.5), our model predicts that the dryland areas of South America, southern and Western Australia, Mexico, southwest United States, and South Africa and to a lesser extent, drylands of Spain, Morocco, and northern Algeria will be the salinization hotspots in response to the key climatic drivers of primary salinization used here. On the other hand, we predict a reduction in the soil salinity of the drylands spread across the northwest United States, the Horn of Africa, Eastern Europe, Turkmenistan, and west Kazakhstan, relative to the reference period (1961 - 1990). The predictive model developed in the present study can inform land and water resources managers and policy makers to distinguish the salinisation hotspots with the priority of the required human intervention, devising viable plans for mitigation of the negative consequences of salinization, and putting those plans in action toward sustainable use of terrestrial ecosystems.
Subjects
soil salinization