Quantification of the uncertainty of geoscientific maps relying on human sensory engagement

Abstract

Uncertainty of a geoscientific map derived from data sampled by sensory engagement of human experts cannot be quantified from within the processing chain followed to produce the map due to subjective components in the underlying data. Nevertheless, quantified uncertainty associated with such maps would be essentially required for correct information retrieval from such maps. An approach mimicking an expert elicitation through digital agents in an information fusion procedure is suggested to quantify the trustworthiness of a partially subjective map (such as a soil map or any other geoscientific map inherently linked to partial subjectivity, e.g., a geological map). This procedure links a partially subjective map with collocated maps resulting from technical measurements provided with quantified uncertainty. Variation of methodology in this approach results in an ensemble of maps with variable degree of matching with the partially subjective map which allows for probabilistic statements about the trustworthiness of the partially subjective map. This enables including partially subjective maps in further data integration relying on numerical analysis, e.g., in the context of computational modeling or machine learning. By being able to assign a quantitative trustworthiness to a partially subjective map the risk of over-utilization of this map is reduced.