Code and parameter package for long-term mapping and object-based typing of freeze–thaw colluvial deposits (1990–2024) along the Yunnan–Tibet corridor

This repository accompanies our manuscript submitted to <i>Catena</i> and provides the core code and parameter files used to identify, type, and track the spatiotemporal evolution of freeze–thaw colluvial deposits (FTCD) from 1990 to 2024 across the Yunnan–Tibet transportation corridor. The package includes: (i) Google Earth Engine scripts for pixel-level Random Forest classification across eight epochs (1990, 1995, 2000, 2005, 2010, 2015, 2020, 2024); (ii) ArcPy workflows for object-based post-processing and a tri-axial typing scheme (genesis, spatial belts, material composition); and (iii) configuration files (YAML/JSON) documenting preprocessing options, model hyperparameters, probability thresholds, rule dictionaries, and version hashes. To support reproducibility without redistributing large third-party imagery, we provide a <b>minimal example tile</b>, sample training/validation/testing indices with <b>spatial blocking</b> (6:2:2) and an <b>independent extrapolation mask</b>, plus templates for change-detection and attribute export. The object layer schema (e.g., <code>obj_id</code>, <code>tri_label</code>, <code>conf</code>, <code>dist_rd</code>, <code>dist_rv</code>, <code>area_chg5y</code>, <code>insar_v</code>, <code>elev_mean</code>, <code>slope_mean</code>) matches the fields used in the manuscript, enabling direct operational uptake. Users can re-pull Landsat/Sentinel and ancillary layers via the included GEE queries. Known caveats (e.g., cloud/shadow contamination, cross-sensor harmonization, terrain illumination residues, “representative-year covariate” assumption) are noted in the README. Tested with Python 3.x/ArcGIS Pro (ArcPy) and the GEE Code Editor. Please cite this dataset together with the related article. Correspondence: Hui Ye ([e-mail]).

本开源仓库配套于投稿至《Catena》的研究稿件，提供了用于识别、分类并追踪1990年至2024年滇藏交通走廊沿线冻融崩积物（freeze–thaw colluvial deposits, FTCD）时空演化过程的核心代码与参数文件。本软件包包含以下内容：(i) 谷歌地球引擎（Google Earth Engine, GEE）脚本，用于8个时间节点下的像素级随机森林（Random Forest）分类，覆盖时间节点为1990、1995、2000、2005、2010、2015、2020、2024；(ii) ArcPy工作流，用于面向对象的后处理以及三轴分类体系（涵盖成因、空间分带、物质组成三类维度）；(iii) YAML/JSON格式的配置文件，记录了预处理选项、模型超参数、概率阈值、规则字典与版本哈希值。为保障研究可复现性且无需分发大量第三方影像数据，本仓库提供了<b>最小示例瓦片</b>、采用<b>6:2:2空间分块</b>划分的训练/验证/测试样本索引，以及<b>独立外推掩膜</b>，同时附带变化检测与属性导出模板。对象图层架构（例如<code>obj_id</code>、<code>tri_label</code>、<code>conf</code>、<code>dist_rd</code>、<code>dist_rv</code>、<code>area_chg5y</code>、<code>insar_v</code>、<code>elev_mean</code>、<code>slope_mean</code>）与稿件中使用的字段完全匹配，可直接投入业务应用。用户可通过内置的GEE查询脚本重新拉取Landsat、Sentinel影像及辅助图层。README文件中已注明本数据集的已知局限性，例如云/阴影污染、跨传感器协调问题、地形光照残留效应以及“代表性年份协变量”假设。本仓库已通过Python 3.x、ArcGIS Pro（ArcPy）及GEE代码编辑器完成测试。请将本数据集与相关研究论文一同引用。联系方式：叶辉（[e-mail]）