Design storm duration from hourly rainfall records in a bimodal Andean climate

Integrating WMO-168 data screening, cascade infilling, and IETD-based frequency analysis for the CVC hydrometeorological network, Valle del Cauca, Colombia

Abstract

Context and motivation. Hydrological design in convective tropical regions requires a reliable estimate of the design storm duration: the continuous rainfall time that feeds rainfall-runoff transformation models for a given return period. In the Valle del Cauca (Colombia), the Corporación Autónoma Regional del Valle del Cauca (CVC) hydrometeorological network provides hourly
pluviographic records, but their direct use for design is hampered by three systematic obstacles: (i) a proprietary block-monthly storage format with mixed missing-data codes; (ii) structural data availability capped at 64–67 % because the dataloggers operate on a variable duty cycle (typically 8–16 active hours per day depending on acquisition mode); and (iii) the absence of a standardised quality-control and frequency-analysis protocol adapted to the bimodal Andean climate.
Objective. To present and validate a complete methodological framework for processing hourly rainfall series of the CVC network, from raw data screening through the probabilistic estimation of design storm durations for return periods of 2–200 years, applied to La Primavera station (Guadalajara de Buga, 1 644 m a.s.l., 1970–2025).
Methods. The framework integrates four complementary stages: (i) quality control and cascade infilling (piecewise linear interpolation for gaps ≤ 6 h; hourly climatology by month for longer gaps, following Paulhus & Kohler [9] for the climatological stage only); (ii) annual screening against WMO-No. 168 Criteria C1–C2 [15], adapted to the structural availability of CVC dataloggers; (iii) identification of independent storm events through the Inter-Event Time Definition (IETD) criterion of Restrepo-Posada & Eagleson [12], with a physical upper bound of 6 h for Andean tropical convection [10]; and (iv) frequency analysis of storm durations fitting four probability distributions (Exponential, Gamma, Log-Normal, Weibull), evaluated by the Kolmogorov-
Smirnov goodness-of-fit test.
Results. Nine years (2016–2025, excluding 2017) pass the WMO-168 screening, yielding a mean availability of 64 %. With IETD = 6 h, 1 027 independent storms are catalogued (rate = 114 storms/yr; mean depth = 19 mm; mean duration = 5.1 h). The inter-event time coefficient of variation (CvTBT = 4.47) reflects bimodal seasonal clustering and confirms a non-Poisson arrival process — relevant only to the IETD-based storm definition; the subsequent frequency analysis is empirical and arrival-process-agnostic. The Log-Normal distribution (μln = 1.352, σln = 0.743; DKS = 0.093, lowest among the four candidates) best describes the duration population. Design durations are 3.86 h (Tr = 2 yr), 10.0 h (Tr = 10 yr), and 21.7 h (Tr = 100 yr), consistent with the regional hydrometeorological literature for the Colombian Andes.
Conclusions. The proposed framework provides a reproducible, normatively grounded procedure for deriving design storm durations directly from the observed hourly record without requiring neighbouring-station data or regional regression models. The methodology is transferable to any CVC station by adjusting a minimal set of site-specific parameters.