Monitoring geomorphological evolution and assessing conservation priorities of geoheritage sites using multi-temporal DEMs and deep learning

Abstract

Geoheritage faces increasingly severe threats from climate change and human activities, with traditional monitoring methods showing limitations in spatiotemporal coverage and quantification. This study takes the karst landform area in southwestern Guizhou as a case study, integrating multi-temporal DEM analysis and deep learning technology to develop a 3D U-Net + LSTM model that achieves automatic identification and trend prediction of geomorphological evolution. The ICP algorithm is employed to improve registration accuracy, and a conservation priority assessment system is constructed based on three dimensions: geomorphological change intensity, scientific value, and vulnerability. The study identifies three evolution patterns: gradual, abrupt, and periodic, establishes a five-level conservation classification system, reveals the dominant role of precipitation erosion in landform degradation, and predicts evolution trends for the next decade. This research provides a transferable technical framework for intelligent geoheritage monitoring, offering a reference for quantitative assessment and conservation management practices.