Reconstructing the Topographic Evolution of Active Margins Since the Devonian Using Artificial Intelligence

Active margin topography has profoundly shaped Earth’s climate, biodiversity, and natural resource distribution over geological time. However, reconstructing paleotopography in these regions remains challenging due to the sparse and uneven distribution of proxies like stable isotope paleoaltimetry, palynology, paleobotany, and thermochronology. These traditional methods often leave large spatial and temporal gaps, with uncertainties in paleoelevation estimates reaching up to 2,000–3,000 m. To address these challenges, we introduce an innovative workflow utilizing artificial intelligence to reconstruct paleotopography at active margins since the Devonian. Using Explainable Boosting Machines (EBMs), we identify key factors such as plate kinematics, mantle dynamics, and climate that govern active margin topography. Insights from the EBM analysis guided the development of a Random Forest (RF)-based regressor which was then used to predict paleotopography through time. Our RF model achieved a mean error of 554 m when validated against present-day ETOPO elevation data. Our model highlights time-evolving subduction flux, trench migration rates, and upper mantle temperature as the primary controls on active margin topography. To validate our approach, we compare our reconstructions with existing paleotopographic models and geological proxies in two regions: the Cenozoic Andes and Mesozoic-Cenozoic Eastern China. For the Andes, our model closely matches the existing reconstructions, highlighting a ~4,000 km rapid rise of the Altiplano since the late Oligocene, driven by an increase in subduction flux (from 0.03 km³/yr to 0.10 km³/yr) and a transition in trench migration from retreating (2 cm/yr) to stationary, likely due to slab anchoring. In Eastern China, our model predicts sustained high topography (>2,500 m) during much of the Cretaceous, attributed to high subduction flux (>0.12 km³/yr) from the Pacific Plate and an advancing trench. A subsequent shift to trench retreat (-2 cm/yr) in the Late Cretaceous–Early Cenozoic led to back-arc extension and a decline in elevation to ~1000 m. Our study offers a transformative approach to bridging gaps in paleotopographic constraints, improving our understanding of the interplay between surface and interior processes. By providing a robust framework for reconstructing past landscapes, our model has significant implications for studying ecosystems, biodiversity evolution, and the metallogenesis of convergent margins.