Random forest-based bioavailable strontium isoscape for environmental and archaeological applications in central eastern Argentina and western Uruguay
Fig 5
Predictive variables and model accuracy in Sr Isoscape analysis.
Cross-validation scatterplots of observed vs. predicted values for the test–validation sample splits, Increase in Mean Squared Error (%IncMSE), Increase in Node Purity (IncNodePurity) and most predictive variables from Bataille et al. (2018; 2020) [18,43]: r.elevation is referred to SRTM topography (m), r.bulk is Bulk Density (kg*m−3), r.fert is Global Fertilization, r.dust is Multi-models average (g.m−2.yr−1), r.pet is Global Potential Evapo-Transpiration, r.cec is Cation Exchange Capacity. Other variables in Bataille et al. (2018; 2020) [18,43] and Reich et al. (2024) [71].