Drought and flood catastrophes threaten water security in the Yellow River Basin, yet prediction is hampered by sparse extreme event records. This study couples catastrophe series extraction via GM(1,1), Weibull extreme value modeling, and bootstrap uncertainty quantification, applying it to 74 years (1951–2024) of annual precipitation in Zhengzhou, China. Percentile thresholds of 410 mm (drought) and 864.6 mm (extreme precipitation) yield five drought and seven extreme precipitation catastrophe years. Both Weibull fits pass three goodness-of-fit tests, whereas the GM(1,1) benchmark fails for the flood series (mean error 21.5% vs. 3.31% for drought). The 0.9 quantile recurrence times (17.8 and 22.9 years) project the next drought catastrophe around 2031 (90% bootstrap window 2025–2033) and the next extreme precipitation catastrophe around 2044 (2034–2047); conditional residual life forecasts shift these by only 1–2 years (to 2033 and 2045, respectively). Two truncated record hindcasts (2013 drought; 2003 flood) fell within the predicted windows, and threshold perturbation and alternative distribution analyses support their robustness. Divergent shape parameters (k = 2.47 vs. 1.29) reveal accelerating drought but decelerating extreme precipitation hazard, providing probabilistic lead times for disaster risk management in data-scarce regions.

Mid-latitude climate transition zones (30–35° N), where monsoon, westerly, and orographic forcing interact, face reorganizing precipitation regimes; yet, urban flood design standards there remain calibrated to stationary historical statistics. Using Zhengzhou, China, as a representative case, we reconstruct its precipitation regime from a homogenous 40-year monthly record (1985–2024) and diagnose the July 2021 “7·20” event through a thermal saturation–water-vapor surplus–terrain locking (THTL) framework built on ERA5 environmental fields (used only diagnostically, strictly separated from ground-truth rainfall), combined with the official design-benchmark audit and a four-event intensity–impact record (2021–2025) used as initial anchors. Pettitt and Buishand tests consistently locate a regime transition around 2002/2003: the mean annual precipitation rose by 23.7% and the extreme-month frequency nearly doubled, while excluding 2021 shows the variability surge is dominated by that single year. The “7·20” event (624.1 mm in 24 h; record 201.9 mm h⁻¹) overloaded the drainage system by 1.6–2.5 times its design basis, with failures cascading through the metro, tunnel, and reservoir systems; peak accumulations occurred over the foothill–urban transition belt, consistent with terrain locking. The study does not re-compute design return periods; rather, it assembles the evidence base towards re-benchmarking design standards in transition zones—regime-aware, non-stationary design updating coupled with mechanism-based early warning—with quantitative non-stationary return-period reassessment defined as the essential next step.

Timely, differentiated distribution of emergency materials reduces flood losses and supports resilient, sustainable cities, and yet demand urgency is rarely quantified jointly with routing. We propose an integrated framework combining game-theoretic AHP–CRITIC weighting, grey relational TOPSIS, exact Fisher–Jenks partitioning, and demand-driven ant colony routing applied to Jiaozuo’s July 2021 “7·20” flood in China. Demand urgency indices for eleven areas span 0.3064–0.6020 and are divided by exact Fisher–Jenks optimization into four grades mapping onto China’s four-tier emergency response system; the grades are robust across weighting schemes (identical under CRITIC-only and equal weights, with only two one-grade shifts under AHP-only weights) and stable under bootstrap perturbation. The routing model incorporates an edge risk exposure term; the optimal route trades an 8.0% increase in adjusted distance for a 6.1% reduction in urgency-weighted delivery time (F* = 235.8774), meeting the local 6 h first-wave dispatch target. At a matched evaluation budget, the improved and standard ant colony algorithms tie in solution quality, and both significantly outperform the genetic and particle swarm algorithms (p < 0.001). Robustness holds across weighting schemes, detour factors, and bootstrap perturbations, supporting resource-efficient, disaster-resilient emergency logistics (SDG 11.5).