A benchmark dataset of null-calibrated streamflow spectral-memory diagnostics for five CAMELS-family archives, with open rainfall-runoff model outputs for CAMELS-GB v2

Abstract. This article describes a reusable benchmark dataset, rather than a standalone methods paper, of streamflow spectral-memory diagnostics computed from daily discharge records in five CAMELS-family archives (CAMELS-US, CAMELS-GB v2, CAMELS-BR v1.2, CAMELS-AUS v2 and CAMELS-DK; 3,010 gauges in total). For each gauge the release provides the positive-integral autocorrelation time of discharge anomalies, local spectral-slope curves on the raw frequency axis and on the release-defined memory-normalized coordinate De =τacf f, dispersion statistics with gauge- and region-cluster uncertainty, and matched stochastic reference products. For CAMELS-GB v2 it also provides the outputs of three open rainfall-runoff models and two seasonal references under a common chronological split. The De coordinate and the null-calibration ladder are operational fields and reference products defined for this release from standard signal processing and transparent simulations; they are not presented as previously validated, universal techniques. The archive-level characterization is intentionally bounded: memory scaling reduces cross-catchment spectral-curve dispersion in US, GB and BR (24.0 %, 14.7 % and 11.8 %), is weak and uncertain in AUS (7.7 %, region-cluster 95 % interval -0.7–27.6 %), and shows no reduction in the lowland DK archive (-4.7 %, -20.2–11.1 %). The matched references quantify how much alignment can arise from serial correlation alone. A CAMELS-GB worked reuse case, based on three independent models and 544 gauges (308 disagreement gauges), does not support spectral distance as a universal model-selection rule (median decision delta +0.124, bootstrap 95 % CI +0.085 to +0.163). The release is intended for catchment-scale spectral characterization, frequency-specific model comparison and reproducible benchmark construction; it does not establish a storage mechanism or a cross-archive ranking rule. The curated dataset is available from Zenodo at https://doi.org/10.5281/zenodo.22055294 (Ren, 2026).