Projected Permafrost Thermal State and Degradation Maps for the Qinghai-Tibet Plateau (2041–2100)

This Zenodo record provides a comprehensive collection of permafrost thermal state and degradation maps for the Qinghai-Tibet Plateau, together with the core GeoTARTE fusion code used in this study. The data correspond to the findings presented in the manuscript titled "A Hybrid Deep Learning Method for Permafrost Prediction Under Data Scarcity and Spatial Heterogeneity on the Qinghai-Tibet Plateau". The dataset includes permafrost maps in GeoTIFF format at 1 km spatial resolution. These maps were generated using the GeoTARTE framework, a hybrid deep learning method that integrates tabular foundation models with spatial attention mechanisms. The dataset contains the following variables for each scenario: Mean Annual Ground Temperature (MAGT, °C)Active Layer Thickness (ALT, cm)Degraded permafrost mask (binary)The maps cover the following time periods and climate scenarios: Current (baseline period, 2006-2020)SSP1-2.6 (2041-2060, 2061-2080, 2081-2100)SSP3-7.0 (2041-2060, 2061-2080, 2081-2100)SSP5-8.5 (2041-2060, 2061-2080, 2081-2100)In addition to the mapped products, this record also includes the core GeoTARTE fusion code developed for this study. The deposited code contains the key MAGT and ALT fusion workflows, including TARTE feature preparation and GeoTARTE optimization scripts. The GeoAggregator component was adapted from the original GeoAggregator repository (https://github.com/ruid7181/GeoAggregator), and the TARTE component was adapted from the original TARTE repository (https://github.com/soda-inria/tarte-ai).