Portfolio-Scale Screening of Conditional Life-Loss Consequences fromTailings-Dam Failure Using Open Global Data

Tailings storage facilities (TSFs) continue to expand globally, and documented failures have produced catastrophicdownstream consequences [1], [3], [4]. This motivates portfolio-scale consequence screening methods that are transparent,reproducible, and fast enough to apply across many sites where detailed breach hydraulics and evacuation modelling are notfeasible as a first pass [2]. This study presents a screening-level framework to estimate conditional Potential Loss of Life(PLL; expected fatalities given failure) using open global datasets and a standardized scenario-set logic implemented inGoogle Earth Engine (GEE).
The workflow integrates (1) DEM-derived dam geometry, (2) a regression-based release proxy, (3) simplifieddownstream routing, and (4) population-based loss estimation. For each site, dam height and stored tailings volume areestimated using DEM cut-and-fill on Copernicus DEM GLO-30, and a cap-aware release set is generated via an externalExcel regression model [3], [4]. Routing follows an analyst-traced centerline guided by MERIT Hydro drainage structure.Released volume is mapped to a longitudinal thickness field that decays exponentially with distance and is trimmed at aminimum threshold. Exposure is evaluated on the native WorldPop 100 m grid by reprojecting depth to the population gridprior to any counts (population is not resampled), applying a depth-fatality step relationship per cell, and summing expectedfatalities across the corridor [5], [6].
We demonstrate the approach for a 17-site TSF portfolio using a deterministic scenario set of 144 runs per site, formedby the cross-product of four release fractions and corridor-parameter grids (decay length, minimum thickness, and half-width), with equal weights as a screening assumption. Site-level outcomes are summarized using exceedance-probabilitybehaviour and upper-tail metrics (VaR95 and ES95). Across the portfolio, upper-tail PLL varies by multiple orders ofmagnitude, yielding clear separation into a small high-consequence subset, an intermediate group, and a majority low-consequence group. Scenario distributions differ substantially in shape. Several sites show a concentrated bulk with apronounced long tail, where typical scenarios produce negligible losses but a small fraction of combinations generate severeoutcomes. Cross-site comparisons indicate that tail severity is more strongly associated with downstream exposed populationthan with stored-volume proxies, consistent with exposure dominating conditional life-loss magnitude under a commonrouting parameterization. The workflow outputs site-ranked consequence metrics and EP-style summaries suitable for rapidprioritization and targeting of detailed site investigations.
The framework is intended to support standardized ranking and prioritization for follow-up studies, while detailed breachhydraulics and evacuation analyses remain necessary for site-specific assessment. Results should be interpreted withinscreening assumptions (regression-based releases, corridor routing, transferable depth-fatality mapping, populationuncertainty, and equal scenario weights), and they represent conditional consequences given failure, not annualized risk.