Storage Droughts Across Peru's National Reservoir System: Data and Reproducible Methodology

Data and reproducible processing framework supporting the manuscript “Storage Droughts Across Peru’s National Reservoir System: Propagation, Attribution and Monitoring Capability”. The 21 reservoirs of Peru’s national regulated storage system are analysed from their daily operational records, coupled to gridded precipitation over their delineated contributing catchments, in order to characterize how meteorological drought propagates into reservoir storage, how much of that storage variability is attributable to catchment climate, and what the existing national drought warning can actually resolve. Scope of the dataset. 21 reservoirs · 120,943 retained daily records · 3,903 hm3 of combined maximum observed capacity · 34,424 km2 of delineated contributing catchments · monthly precipitation 1981–2025 · 158 storage drought events and 252 meteorological drought events. Contents code/ — fifteen sequential Python scripts covering the full chain from quality control to case verification, plus a shared module of constants and helpers and the table-generation script. Each script reads only the outputs of its predecessors, so any step can be re-run in isolation. All thresholds are declared as named constants at the head of the script that applies them. data/ — every intermediate and final product in CSV and GeoPackage format: the consolidated daily storage database, the quality-control log, the delineated catchments, the catchment precipitation series, the standardized indices, the drought catalogue, the propagation grids, the decomposition, the ENSO response, the net-balance series and the monitoring skill tables. figures/ — the sixteen figures of the manuscript and its supplement, in raster (400 dpi PNG) and vector (PDF) form. The methodological workflow figure is additionally provided as an editable SVG with live text. tables/ — the thirteen tables, in CSV and LaTeX. Method The Standardized Precipitation Index is computed with a gamma distribution fitted per calendar month over accumulation scales of 1 to 48 months, with zero handling by the mixed distribution of Thom (1958), following McKee et al. (1993) and the WMO SPI User Guide (WMO-No. 1090). The standardized storage index follows the construction of the Standardized Reservoir Supply Index (Gusyev et al., 2015) but is computed nonparametrically with the Gringorten (1963) plotting position, after Farahmand and AghaKouchak (2015): monthly storage is bounded above by the capacity of the vessel, so imposing a parametric family would introduce a distributional artefact that cannot be separated from a hydrological signal. Drought events are extracted by run theory (Yevjevich, 1967) at a threshold of −1.0, which is the boundary of the “moderately dry” category applied in Peru’s operational drought bulletin. Propagation is characterized by lagged correlation over the full scale × lag grid, after Barker et al. (2016), with the accumulation window taken as the centroid of the correlation profile rather than its arg-max. Contributing catchments are delineated by upstream traversal over HydroBASINS level 12 and screened with a runoff-coefficient criterion that detects inter-basin transfers. Warning skill is evaluated with a contingency table against the accumulation scale set published monthly by the national meteorological service, reporting the probability of detection, the false-alarm ratio, the probability of false detection and the Hanssen–Kuipers skill score, over a sensitivity grid of 36 configurations. Negative results included This deposit deliberately contains the material behind two results that did not confirm the analysis’s initial expectation. First, an apparent relationship between reservoir memory and reservoir size over the full record (ρ = +0.52, p = 0.039) does not survive a common-period control (ρ = +0.28, p = 0.37); the confound is that record length correlates with reservoir size in any national archive, while the precision of the window estimate depends on record length. Second, we predicted that differencing storage into a monthly net balance would raise the variance explained by climate; it fell instead, from 0.21 to 0.08, because differencing the storage does not remove operation but exposes it. Input data This deposit contains derived products. The primary inputs are publicly distributed by their providers and are not redistributed here: daily reservoir storage from the National Water Information System of the National Water Authority of Peru; gridded precipitation from PISCOp v3.0 (Aybar et al., 2020); sub-basin topology from HydroBASINS v1.c (Lehner and Grill, 2013); surface water from the JRC Global Surface Water dataset (Pekel et al., 2016); and the coastal and oceanic ENSO indices from ENFEN and NOAA respectively. Reproducibility Requires Python 3.11 with numpy, pandas, scipy, geopandas, rasterio and openpyxl. Running the fifteen numbered scripts in order, followed by the table generator, regenerates every data product, table and reported figure of the manuscript from the publicly available inputs. Licence. Data and documentation under CC BY 4.0; code under MIT.