The importance of zeros in digital soil mapping I: a review

The presence and influence of excess zeros has been understudied and inconsistently applied within the digital soil mapping (DSM) field. Other disciplines have identified the challenges associated with fitting empirical models when the data are highly zero-inflated and have contributed to the development of modelling frameworks that can be employed to overcome them. This paper presents the need to address zero-inflation (ZI) and two of the primary methodologies currently used: zero-inflated models and two-part frameworks. The selection of which modelling framework to use relies primarily upon the researcher understanding the source of ZI within their data set, a frequent challenge when using legacy data in DSM. Several examples of properties where ZI has been recognized or is likely to occur in DSM are detailed, including depth to bedrock, coarse fragment content, horizon thickness, soil inorganic carbon, soil contaminants, and soil organisms. Ultimately, the review of ZI within DSM literature revealed that few papers have employed modelling strategies to handle excess zero values, some apply the frameworks inconsistently, and no papers directly address the source of zero observations. Future areas of research are introduced including the integration of machine learning into ZI frameworks, the suitability of ZI models as a reference to identify possible false zeros, specific use cases in spatial applications, and the use of ZI in non-target observations that may be encountered in bulk soil sampling for several properties.