A probabilistic and model-based approach to the assessment of glacial detritus from ice sheet change

Detrital provenance methods aid to understand ice sheet conditions in the past, often seeking to define the location of the eroding ice sheet margin. To support interpretation, we develop a probabilistic approach to predict detrital production linked to changing ice sheet conditions. The approach is applied in western Wilkes Land, Antarctica. Considering four candidate ice sheet states, spatial variations in erosion likelihood are estimated from ice sheet models. Spatial analysis with a graticule shows that this process is highly effective in eliminating unlikely source regions. Spatial analysis with a subglacial bedrock geology classification, returns a more comprehensive result tied to geology. Analysis results are consistent with observed modern detritus in the region and highlight variations in predicted detritus production with ice sheet retreat, including the changing importance of different source regions and changes in the nature and diversity of detritus from these. We use the relative likelihood and its deviation from base-rate probability to define qualitatively the level of support for different ice sheet states from detrital records. With current detrital records, qualitative interpretations cannot categorically differentiate ice sheet states, except locally. The quantitative differentiation of ice sheet states is demonstrated for detrital occurrences, given a prior probability and a transport effectiveness for each ice sheet state, showing a high power of discrimination. Our approach supports a more robust capacity to use the detrital record to interpret ice sheet change, even where records are sparse, but high-quality and well constrained detrital records are needed to maximise benefit from quantitative analysis.