Key points are not available for this paper at this time.
Soil mapping is essential for sustainable land management, precision agriculture, and environmental monitoring. This synoptic review evaluates the evolution of classical and modern geostatistical methods, spanning 2000 to 2024, and their integration with machine learning (ML) and remote sensing (RS) technologies. Key techniques such as spatial autocorrelation, variogram analysis, kriging and its variants, Bayesian models, Markov Chains, geostatistical simulations, and ensemble modeling were assessed in the context of digital soil mapping (DSM). Emphasis was placed on hybrid approaches that fuse geostatistics with ML algorithms including Random Forest (RF), Extreme Gradient Boosting (XGBoost), Support Vector Machines (SVM), and Artificial Neural Networks (ANN), along with the enrichment of spatial models using RS data. Hybrid approaches combining geostatistics with ML algorithms (e.g., RF, Boost, SVM, ANN) demonstrate promise in addressing spatial uncertainty, while RS data enhances covariate enrichment and near-real-time applications. Although advancements in variogram estimation and kriging techniques have optimized sampling strategies, and improved prediction accuracy, challenges persist in computational efficiency and uncertainty quantification. However, emerging trends such as, Bayesian methods, multiscale modeling, and sensor networks show the potential to mitigate these limitations. Future research should focus on improving model accuracy, interpretability, transparency, and explainability. Additionally, addressing data quality issues, scalability, and increasing sampling density to support robust spatial analysis and uncertainty quantification are major research gaps and future directions for soil mapping and information systems. This review is a valuable resource for researchers, practitioners, and policymakers engaged in data-driven land and soil management.
Caires et al. (Fri,) studied this question.