A Data-Driven Hybrid Random Forest and IMDPA Approach for Mapping Soil Degradation in Iranian Arid Lands

In dry regions of Iran, it would be important to quantify the processes involved in desertification in the management of the land. This paper illustrates a combination of hybrid framework of a Random Forest (RF) model and the Iranian Model of Desertification Potential Assessment (IMDPA) to go beyond the assessment of a single point in time and to model the spatiotemporal interactions between soil degradation and other processes in the Yazd-Ardakan plain. RF model accurately forecasted the important indicators of degradation such as EC (R ²  = 0.73), SAR (R ²  = 0.76), and the Soil Quality Index (SQI) (R ²  = 0.77) using 201 soil samples (2016) and multi-temporal Landsat (1986–2016). The variable importance analysis established that two process types (natural hydrological processes) (groundwater salinity) and (anthropogenic pressures) (proximity to mines, roads and urban areas) are the primary causes of degradation, respectively, in soil EC and SQI and heavy metals respectively. We quantitatively measured the acceleration of the process of desertification by rebuilding historical soil maps within 30 years of time. The findings indicate that there is a drastic growth (more than 60-fold, between ~ 30 ha and ~ 1800 ha) of the Very High desertification category, which is spatially clustered around the Yazd-Ardakan industrial belt. The given direct, data-based connection between human activity and the accelerating process of degradation offers an innovative, quantitative instrument of the accurate monitoring and specific management of soil resources in the vulnerable arid environments.