Model-level dataset from a quantitative review of XRF-based soil property prediction and data-fusion approaches

The database was compiled from peer-reviewed studies published between 2012 and 2026. Each row represents one predictive model entry rather than one unique article.</p><p dir="ltr">The dataset includes information on the target soil attribute, sensor configuration, data-fusion level, fusion implementation strategy, modeling algorithm, preprocessing method, experimental environment, geographical scope, sample size, country, climate classification, soil texture information, and model performance metrics.

The main performance metrics include R², RMSE, RPD, and RPIQ when reported in the original studies.</p><p dir="ltr">Sensor configurations were classified as XRF-only when the predictive model used only XRF-derived variables and as data fusion when XRF was combined with at least one additional source of information. Data-fusion approaches were harmonized into three main levels according to the stage at which information was integrated: low-level, mid-level, and high-level fusion.