Testing spatial interpolation methods for deep-time organic carbon burial in epicontinental seas by taking Sunda Shelf as an example

Abstract

Quantifying the burial of organic carbon (OC) in epicontinental seas is crucial for understanding its role in regulating global long-term carbon cycle and climate. Utilizing spatial interpolation methods, prior works have quantified OC burial globally or regionally based on limited, unevenly distributed measurements. However, there remains a notable lack of comparative studies and assessments regarding their applicability and uncertainty in deep-time research. Taking the middle Miocene Sunda Shelf OC burial estimation as an example, four popular spatial interpolation methods are assessed quantitatively and qualitatively, including Thiessen polygons, Inverse Distance Weighting (IDW), Ordinary Kriging (OK) and Random Forests (RF). Based on quantitative and qualitative evaluation, the data-driven RF method demonstrates superior performance due to fewer assumptions, effectively capturing nonlinear relationships and complex spatial patterns in heterogeneous, non-Gaussian deep-time data, and demonstrating strong generalizability and robustness. High-resolution RF-based reassessment reveals significant spatial-temporal heterogeneity of OC burial on the Sunda Shelf between the Miocene Climatic Optimum (MCO) and Middle Miocene Climate Transition (MMCT). Although the overall OC burial and sediment accumulation rates (SAR) increase during the MMCT, notable spatial discrepancies emerge, with OC burial rates elevated near basin margins but decreased in distal regions. These variations primarily reflect the combined influences of eustatic sea-level fall and enhanced terrigenous input, highlighting the complex interplay of factors modulating OC burial efficiency. Machine learning methods such as RF prove highly effective in handling deep-time spatial data, but their application should be adapted to specific objectives, geological conditions, and data characteristics.