Supporting Dataset — Spatiotemporal Dynamics of Urban Resilience: An Integrated Assessment across Four Case Studies (1996–2024)

This dataset supports the manuscript: Spatiotemporal Dynamics of Urban Resilience: An Integrated Assessment across Four Case Studies (1996–2024), submitted to Humanities and Social Sciences Communications (Springer Nature). Overview: This supporting dataset provides the processed tabular data, model performance metrics, and planning framework thresholds underlying the results reported in the manuscript. All primary satellite inputs were derived from publicly available sources (ESA CCI, ERA5, Sentinel-2). This file contains the intermediate and derived datasets necessary to interpret, replicate, and build upon the study's findings. Study context The manuscript examines urban hydrological resilience across four contrasting case studies spanning 1996–2024: the Greater Bay Area (GBA), China (subtropical megalopolis); the United Arab Emirates (hyper-arid extreme events); southern Madagascar (Global South drought vulnerability); and the Andean páramos, Ecuador/Peru (natural reference system). The core analytical output is the Robustness Index (RI = μ(SM)/σ(SM)), derived from ESA CCI satellite soil moisture data, which underpins a three-tier spatial planning framework (Conservation: RI > 2.5; Mitigation: 1.5 < RI < 2.5; Engineering: RI < 1.5). Contents — six data tables (Excel workbook, 7 sheets) S1 — Annual Soil Moisture and RI (1996–2024): Area-averaged volumetric soil moisture (%) and Robustness Index values for all four case study regions, derived from ESA CCI SM v08.1 (0–10 cm layer, 0.25° resolution). S2 — RI Tier Classification Summary: Mean, minimum, and maximum RI by case study sub-period; tier assignments; and flood frequency validation data (events per decade) drawn from GBA Water Resources Department (2024), Otto et al. (2025), and Mosquera et al. (2015). S3 — Multi-Source Fusion Model Performance: Cross-validated R² and RMSE for SVR and KNN models using DINOv3-derived features from Street View Imagery (SVI, 768-dim) and Remote Sensing Imagery (RSI, 1024-dim), tested across buffer distances of 15, 30, 50, and 100 m and all four case study sites (Leave-One-Site-Out cross-validation). S4 — ETa Estimation Method Comparison: Performance metrics (NSE, Bias, nRMSE, Pearson r) for seven actual evapotranspiration estimation methods — PMCal, Eddy Covariance, HBV-Light, ERA5-Land, MOD16, SiB-Model, and Water Balance — validated against eddy covariance measurements at Andean páramo sites. S5 — UAE April 2024 Extreme Rainfall Event: Monthly NDVI recovery data (Sentinel-2, 10 m) and observed rainfall totals for the Arabian Desert, March–September 2024, documenting the post-ERE greening response. Includes event rainfall total (254.8 mm in 24 hours) and exponential recovery model parameters. S6 — Three-Tier Planning Framework Thresholds: Empirical RI and Compactness Index (CI) thresholds, impervious surface ranges, flood frequency benchmarks, representative case study sites, primary intervention strategies, and GIS geodatabase layer naming conventions for each planning tier. Primary data sources (all publicly available) ESA CCI Soil Moisture v08.1: https://www.esa-soilmoisture-cci.org/ ERA5 reanalysis (ECMWF/Copernicus): https://cds.climate.copernicus.eu/ Sentinel-2 MSI imagery (ESA): https://scihub.copernicus.eu/ Google Street View API: https://developers.google.com/maps/documentation/streetview Key findings supported by this dataset The GBA shows a statistically significant declining SM trend of −0.12%/year (Mann-Kendall Z = −8.74, p < 0.001) over 1996–2024, with RI declining from 2.8 to 1.4 — a shift from the Conservation to Engineering tier. The UAE April 2024 event recorded 254.8 mm in 24 hours (>3× the annual average), with NDVI increasing 275% within 30 days. Multi-source SVI+RSI fusion achieved a mean cross-validated R² of 0.417–0.649 for soil moisture prediction. The Andean páramos maintain RI = 3.8–4.2, serving as the natural reference benchmark. Ethical statement No human participants, personal data, or biological material were involved. All data are derived from publicly available satellite and reanalysis products. No ethical approval was required.