Survey Data: A Two-Stage Reproducibility Framework for Remote Sensing-Based Landslide Mapping

Landslides are major global hazards that require reliable mapping to help reduce loss of life and economic damage. Rapidly evolving remote sensing methods lack reproducibility assessments, which hinder verification and reuse. Our survey of 134 studies (2020-2025) reveals only 9.0% code and 12.7% data availability, stressing structural barriers necessitating a systematic evaluation approach. To address this, we present a two-stage landslide mapping reproducibility framework: (1) a metadata-based transparency assessment of code, data, and methodology, and (2) an execution-based assessment of data preparation, execution, and output consistency. Applying our framework to selected studies reveals that high transparency does not guarantee successful execution, and the theoretical and practical reproducibility gap must be bridged to improve operational readiness. To establish an empirical baseline, we surveyed a representative set of RS-LSM studies in Scopus to assess code and data availability. We limited the survey to journal articles, and to ensure broad coverage, we used keyword queries targeting a range of methods, from simple change detection to standard machine learning (e.g., RF, SVM) and advanced deep learning (e.g., CNNs, GNNs, Transformers). Each query was sorted by relevance, and up to 20 top-ranked results per query were retained to maintain balance across categories. This strategy provided a representative sample suited to our goal of a preliminary analysis rather than an exhaustive literature review. After merging results, we filtered out duplicates and irrelevant studies lacking RS-LSM methods. Subsequently, we evaluated each study based on the availability of its source code and input data. For both criteria, we marked data or code as unavailable if access required permissions or author contact, as these barriers do not guarantee long-term availability. This dataset includes the queries used for the survey, and resulting list of publications.滑坡是全球性重大灾害，需通过可靠的制图工作以降低人员伤亡与经济损失。快速发展的遥感技术方法尚未开展可复现性评估，这阻碍了研究成果的验证与复用。我们对2020至2025年间的134项相关研究开展调研后发现，仅9.0%的研究公开了代码，12.7%的研究公开了数据集，凸显了当前存在的结构性障碍，亟需系统性的评估方案。为解决这一问题，我们提出了两阶段滑坡制图可复现性框架：（1）基于元数据的代码、数据集与研究方法透明度评估；（2）基于执行流程的数据集制备、模型运行与输出一致性评估。将该框架应用于筛选后的研究后发现，高透明度并不等同于可成功复现的研究结果，需弥合理论可复现性与实际可复现性之间的差距，以提升研究成果的实际应用就绪水平。 为建立实证基准，我们针对斯高帕斯（Scopus）数据库中具有代表性的遥感滑坡制图（Remote Sensing for Landslide Mapping, RS-LSM）研究开展调研，以评估代码与数据集的公开情况。本次调研仅纳入期刊论文；为确保覆盖范围全面，我们采用关键词检索策略，涵盖从简单变化检测到标准机器学习（如随机森林RF、支持向量机SVM）及先进深度学习（如卷积神经网络CNNs、图神经网络GNNs、Transformer）在内的多种方法。每个检索式均按相关性排序，每个检索式保留前20条高排名结果，以保证各方法类别的样本均衡。该采样策略可获得具有代表性的研究样本，契合我们开展初步分析而非全面文献综述的研究目标。合并检索结果后，我们剔除了重复文献与不涉及遥感滑坡制图方法的无关研究。随后，我们基于源代码与输入数据集的公开情况对每篇文献进行评估。针对这两项评估标准，若需获取权限或联系作者才能获取代码或数据集，则判定为未公开——此类获取障碍无法保证资源的长期可及性。 本数据集包含本次调研使用的检索式以及最终纳入的文献列表。