Beyond Outcome Variance: Addressing Ranking Instability via an RSSI-SA-ADF Framework for Adaptive Ecological Restoration Decision-Making
收藏资源简介：
This dataset contains the full Supplementary Materials (S1–S17) for the research article “Beyond Outcome Variance: Addressing Ranking Instability via an RSSI-SA-ADF Framework for Adaptive Ecological Restoration Decision-Making.” Research Hypothesis We hypothesize that parameters driving outcome variance (e.g., costs) are statistically distinct from those causing ranking reversals (e.g., stakeholder preferences). The Ranking Stability Sensitivity Index (RSSI) captures this structural instability directly, including interaction effects (S3.3). Figure S4 empirically demonstrates this decoupling. Data Content and Methodology The document includes: the Multi-dimensional Lifecycle Evaluation (MD-LEA) model (S2); mathematical definition and computational algorithm of RSSI with bootstrap confidence intervals (S3); AHP-assisted threshold calibration protocol (τ=0.15) and stationarity tests for bounded preference uncertainty (S4); exploratory LSTM forecasting analysis (S5); TOPSIS-based Pareto solution selection (S6); probability distributions and data sources for 12 key parameters across Nanyang, Shangqiu, and Jiaozuo projects (S7, Table S4); computational efficiency benchmarks (S8); convergence analysis of RSSI and Sobol’ indices (S9, Figures S3–S4); cross-site comparison of ranking reversal events and critical thresholds (S10, Tables S6–S7); failure mode analysis under policy shocks and ecological time-lags (S11); stakeholder feedback and comparison with traditional decision frameworks (S12, Table S8); reproducibility resources (S13); glossary (S14); detailed ROI calculation for the Nanyang case (S15, Table S15-B); cost assumptions for the adaptive-baseline experiment (S16); and sensitivity of RSSI to baseline specification (S17). Code is available in the linked GitHub repository. Notable Findings 1.Social preference parameters (βESV, βcarbon) dominate ranking reversals, while technical parameters (e.g., unit cost) drive outcome variance (Figure S4, S10). 2.The RSSI alert threshold τ=0.15 minimized total error (FPR+FNR=13.3%) with stable cross-validation (S4.3). 3.In the Shangqiu project, SA-ADF achieved 0% cost overrun, 6.6% replanting rate, and 109.0% ROI (Table S8); the Nanyang case achieved 108.8% conservative and 277.8% full ROI (S15). 4.RSSI estimates are insensitive to the choice between median and stakeholder best-estimate baselines (maximum difference 0.01; S17). Interpretation and Usage Use the parameter distributions in S7 for model testing. Recalibrate the alert threshold τ via the protocol in S4.1–S4.3, considering project-specific cost ratios. Consult S10 for context-dependent risk structures; do not apply a universal threshold. Distinguish the ex-post ROI in S15 from the ex-ante cost ratio (10:1) used in S4.2 for threshold calibration. Framework limitations under abrupt policy shifts and ecological time-lags are documented in S11. Data format: .docx (Microsoft Word) Study sites: Nanyang, Shangqiu, Jiaozuo (Henan Province, China)
本数据集包含研究论文《超越结果方差：基于RSSI-SA-ADF框架解决自适应生态修复决策中的排序不稳定性》的全部补充材料（S1–S17）。 研究假设 我们提出如下假设：驱动结果方差（如成本）的参数，与引发排序反转（如利益相关者偏好）的参数，在统计学上存在显著差异。排序稳定性敏感性指数（Ranking Stability Sensitivity Index，RSSI）可直接捕捉这类结构性不稳定性，包括交互效应（详见S3.3章节）。图S4通过实证验证了这种参数解耦现象。 数据内容与研究方法 本补充材料涵盖以下内容：多维生命周期评价（Multi-dimensional Lifecycle Evaluation，MD-LEA）模型（S2）；排序稳定性敏感性指数（RSSI）的数学定义、计算算法及Bootstrap置信区间（S3）；基于层次分析法（Analytic Hierarchy Process，AHP）的阈值校准流程（τ=0.15），以及针对有界偏好不确定性的平稳性检验（S4）；探索性长短期记忆网络（Long Short-Term Memory，LSTM）预测分析（S5）；基于理想解法（Technique for Order Preference by Similarity to an Ideal Solution，TOPSIS）的帕累托最优解筛选（S6）；南阳、商丘、焦作三地12项关键参数的概率分布与数据来源（S7，表S4）；计算效率基准测试（S8）；RSSI与索博尔指数（Sobol’ indices）的收敛性分析（S9，图S3–S4）；排序反转事件与临界阈值的跨站点对比分析（S10，表S6–S7）；政策冲击与生态时滞场景下的失效模式分析（S11）；利益相关者反馈及与传统决策框架的对比研究（S12，表S8）；可复现性资源（S13）；术语表（S14）；南阳案例的详细投资回报率（Return on Investment，ROI）计算过程（S15，表S15-B）；自适应基线实验的成本假设（S16）；以及RSSI对基线设定的敏感性分析（S17）。相关代码可在关联的GitHub仓库中获取。 核心研究发现 1. 社会偏好参数（βESV、βcarbon）是引发排序反转的主导因素，而技术参数（如单位成本）则是结果方差的主要来源（图S4、S10）。 2. 当RSSI预警阈值τ=0.15时，总误差（假阳性率（False Positive Rate，FPR）+假阴性率（False Negative Rate，FNR）=13.3%）达到最小，且交叉验证结果稳定（S4.3）。 3. 商丘项目中，SA-ADF框架实现了0%的成本超支、6.6%的补植率及109.0%的投资回报率（表S8）；南阳案例则分别达到108.8%的保守型投资回报率与277.8%的全口径投资回报率（S15）。 4. RSSI的估计结果对基线选择（中位数或利益相关者最优估计基线）不敏感，最大差异仅为0.01（S17）。 使用说明与解读 可使用S7章节中的参数分布开展模型测试。需结合项目特定的成本比例，通过S4.1–S4.3中的流程重新校准预警阈值τ。针对特定场景的风险结构可参考S10章节，请勿直接使用通用阈值。请注意区分S15章节中的事后投资回报率与S4.2章节中用于阈值校准的事前成本比例（10:1）。本框架在政策突变与生态时滞场景下的局限性详见S11章节。 数据格式：.docx（Microsoft Word文档） 研究站点：中国河南省南阳、商丘、焦作
