Harnessing expert knowledge and legacy data for digital soil mapping with no new field surveys

Abstract

Legacy soil maps, derived from extensive soil surveys, contain invaluable information crucial for soil management practices. However, these maps risk obsolescence due to outdated technology, changes in classification systems, and evolving soil types. Addressing the need for high-precision and high-resolution soil maps, particularly in regions lacking comprehensive survey data, this study proposes an innovative framework for Expert Knowledge and Diagnostic Information-based Digital Soil Mapping (ED-DSM), enabling digital soil mapping with no new field surveys by integrating expert knowledge with diagnostic information. The framework leverages diagnostic horizons and attributes from Chinese Soil Taxonomy (CST), combined with expert assessments of the probability of certain diagnostic features within legacy map units, to extract pseudo-points and assign diagnostic feature types using expert-guided probability-constrained deterministic annealing clustering. Through repeated random sampling and random forest modeling, probability distributions for all diagnostic features are generated, and retrieval rules are constructed to create probabilistic soil type maps. Application of the framework in a county in China yielded the following key findings: (1) ED-DSM successfully generated probability distributions for 16 diagnostic features and produced maximum and secondary probability distribution maps of soil types at the order, suborder, group, and subgroup levels based on the CST, demonstrating exceptional spatial detail; (2) Validation using 33 soil profiles showed an average mapping accuracy for diagnostic features ranging from 0.62 to 0.99, while the average accuracy for soil types at the order, suborder, group, and subgroup levels under maximum probability were 65.86 %, 65.03 %, 47.85 %, and 44.15 %, respectively; and (3) Considering secondary probabilities improved soil type mapping accuracy by 3.55 %–7.19 %, further confirming the method's efficiency and robustness. The ED-DSM framework enables rapid mapping of soil diagnostic features and types without the need for additional soil surveys, offering a cost-effective and scalable solution for resource-limited regions and providing actionable scientific support for soil management practices.