METEOR 2.5D: An Open Geospatial Dataset of the Spatiotemporal Evolution of Physical Vulnerability in UN-recognized Least Developed Countries (as of 2020) at Five-year Intervals, 1975-2030

Tracking spatiotemporal changes in disaster risk is essential for monitoring progress toward the UN Sendai Framework for Disaster Risk Reduction (SFDRR) 2015-2030. Despite rapid advances in information technology and big data, our collective progress remains insufficient. Existing SFDRR indicators offer simple and globally comparable metrics at the national level but rely heavily on short-term trends and inadequately capture the probabilistic nature of disaster risk. While large-scale modeling of hazard and building exposure has already advanced significantly through Earth observation and data-driven methods, progress still lags in modeling another equally important yet challenging element of the risk equation: physical vulnerability. Therefore, we develop a data-driven probabilistic approach to model the regional dynamics of building exposure and physical vulnerability over time. Our work combines recent advances in graph deep learning, state-space modeling, and variational inference, leveraging time-series satellite-derived products with existing expert belief systems. We present METEOR 2.5D, an open geospatial dataset of the spatiotemporal evolution of physical vulnerability in UN-recognized Least Developed Countries (as of 2020) at five-year intervals, 1975-2030. We integrate rasterized temporal exposure datasets, such as DLR World Settlement Footprint Evolution and Global Human Settlement Layer multitemporal products, with the existing static METEOR dataset as prior information to generate dynamic maps with a five-fold improvement in spatial resolution (i.e., from 450-meter to 90-meter scale). By addressing critical gaps in modeling physical vulnerability at large scales, our work enhances the understanding and auditing of our global disaster risk, both now and beyond 2030. The METEOR 2.5D dataset is publicly available in two parts: https://doi.org/pzq4 and https://doi.org/pzrd.