Antecedent storage governs downstream flood outcome at a flood-control reservoir: canyon dam, the 2002 and 2025 Guadalupe River floods, and the probable maximum flood

Canyon Dam on the Guadalupe River in Texas has experienced two extreme inflows whose outcomes were opposite, and the difference between them was not the size of the flood. In July 2002, with the pool already above its conservation level, an inflow peaking near 110,000 ft 3 /s delivered approximately 700,000 acre-ft, nearly twice the 355,000 acre-ft flood-control pool filling the reservoir, overtopping the uncontrolled emergency spillway by 7.32 ft, and discharging 66,800 ft 3 /s downstream, with damage described in the interagency record as catastrophic. In July 2025, the upper Guadalupe produced the largest flood in its recorded history: 315,000 ft 3 /s at Hunt, the maximum of record, and at least 119 fatalities in Kerr County; yet the reservoir released only 389 ft 3 /s, and the corridor below it was untouched. Canyon Dam attenuated the 2025 event by 98.9%. The controlling variable was antecedent storage. Canyon Lake stood at roughly 46% of capacity in July 2025, drawn down by three years of drought, and the inflow consumed 17.4% of the flood-control pool. In 2002, the pool already stood above its conservation level before the rain began. The larger headwater flood produced no downstream damage; the smaller one destroyed the corridor below the dam. This pairing constitutes a natural experiment in the dependence of reservoir performance on conditions that precede the storm, and it establishes that the protection Canyon Dam affords is contingent rather than structural. We quantify that contingency and extend it to its physical limit. Flood-wave attenuation and travel times are measured from 5- and 15-minute instantaneous gage records. Gridded Probable Maximum Precipitation is transformed to runoff using the calibrated interagency (InFRM) HEC-HMS model of the basin, re-calibrated here to include the July 2025 event, and routed through Canyon Lake across a matrix of antecedent pool states and inflow magnitudes spanning frequency storms to the Probable Maximum Flood. Reservoir routing uses an elevation–storage relationship from the Texas Water Development Board volumetric survey and the storage–discharge relationship provided by the U.S. Army Corps of Engineers, verified independently against the 2002 flood of record. The resulting release hydrographs drive two-dimensional HEC-RAS hydrodynamic simulations of the regulated corridor, from which inundation extent and the population at risk follow. These simulations show that the Probable Maximum Flood inundates approximately 790–1,170 mi² of the corridor and places roughly 430,000–620,000 people at risk, both rising as the controlling storm shortens from 72 to 24 hours; the 2002 flood of record inundated about 200 mi² and 128,000 people, whereas the 2025 release remained within the channel. Because the flood exceeds the reservoir’s storage to the spillway crest by a factor of roughly 2.4, no antecedent pool state, including an empty reservoir, averts loss of control. The study contributes the first quantification of the 2025 attenuation from instantaneous records, the first Probable Maximum Flood analysis of the Guadalupe basin, and the first systematic mapping of the antecedent-storage conditions under which Canyon Dam ceases to control its own release.