Spatiotemporal Assessment of Soil Organic Carbon Change Using Machine-Learning in Arid Regions

Soil organic carbon (SOC) is an essential property of soil, and understanding its spatial patterns critical to vegetation management, soil degradation, environmental issues. This study applies a framework using remote sensing data digital mapping techniques examine the spatiotemporal dynamics SOC for Yazd-Ardakan Plain, Iran, from 1986 2016. Here, conditioned Latin hypercube sampling method was used select 201 sites. A set 37 predictors were obtained Landsat imagery taken in 1986, 1999, 2010 modeled 2016 Random Forest (RF), support vector regression (SVR), artificial neural networks (ANN) machine-learners by correlating with data. The results showed that RF yielded highest accuracy (R2 = 0.53), compared other two learners. By performing variable importance analysis model, normalized difference index, modified ground-adjusted index determined be most important predictors. applying model calibrated 1999 2010, substantial decrease SOC; these decreases mainly attributed land use changes agricultural activities.