Data and processing framework for: Evaporative water loss from Peru's reservoirs and regulated lakes: a 42-year satellite and climate assessment

Data and processing framework supporting the manuscript “Evaporative water loss from Peru’s reservoirs and regulated lakes: a 42-year satellite and climate assessment”. Evaporation removes water from storage without serving any demand, yet it is rarely quantified in national water accounts. This deposit contains the complete processing framework and every derived data product of a study (submitted for peer review, 2026) that quantifies evaporative loss from the 20 regulated water bodies of Peru’s national storage system between March 1984 and June 2026. No primary dataset is redistributed: the satellite, reanalysis, elevation and storage records remain with their providers, and the scripts retrieve them. Scope of the dataset. 20 water bodies · 16 dams and four regulated lakes · 110 to 4,868 m above sea level · three hydroclimatic domains · 3,889 hm3 of combined storage capacity · 466 km2 of mean water surface · 9,534 body-months of satellite water area · 7,348 body-months of estimated loss, March 1984–June 2026. Approach Evaporative loss is computed as the product of two independently derived quantities that share no input and meet only in their product: a monthly water surface area from the JRC Global Surface Water record at 30 m resolution, and a monthly open-water evaporation depth from climate data. That independence is deliberate — an error in one branch cannot be compensated by a matching error in the other, which is what allows each branch to be validated on its own terms. The satellite branch measures water within a per-body measurement envelope; exports the valid-observation coverage alongside the water area, so that a dry vessel can be distinguished from a clouded one; reconstructs cloud-affected months from operational storage through calibrated area–volume relations; and extends the series from 2022 to 2026 by the same device once the satellite product ends. The climate branch converts a national reference evapotranspiration product to open water using a single Penman–Monteith equation in which only albedo and surface resistance change between surfaces, and adds interannual variability through a delta bias correction in which the reanalysis enters twice, on both occasions as a ratio of itself, so that its systematic bias cancels. Validation The area extension is validated on withheld data: the relation is refitted on the first 60 % of each body’s observed months and used to predict the final 40 %, giving r2 = 0.961 over 898 withheld months, with a median normalised RMSE of 6.2 %. The evaporation depth is contrasted against an independent Priestley–Taylor benchmark computed from net radiation, which requires neither wind speed nor vapour pressure deficit — the two variables a reanalysis represents worst over mountainous terrain. The median ratio across the 16 dams is 0.97, and 14 of 16 fall within 15 %. External comparison against the only long instrumental record of open-water evaporation on the Andean plateau agrees to within 3 % at the body lying within that basin. Main findings The system loses 629 hm3 yr-1. The 16 dams account for 275 hm3 yr-1, equal to 11.0 % of their combined storage capacity, while the four regulated lakes — one fifth of the bodies but 63 % of the water surface — account for 354 hm3 yr-1. An assessment restricted to dams therefore omits more than half of the national total. Expressed at the outlet, 12 % of the water released from the dams during drawdown evaporates instead of reaching a demand, ranging from 3 % to 77 % between reservoirs, and 50 % for the regulated lakes. The loss does not concentrate in the drawdown season, because rising evaporative demand and a contracting water surface offset one another. Contents code/ — seventeen Python scripts covering the full chain, from the definition of the measurement envelopes to the volumetric loss and the generation of the manuscript tables, plus a shared module of paths, presentation names and the LaTeX table exporter. Each script reads only the outputs of its predecessors, so any step can be re-run in isolation, and all thresholds are declared as named constants at the head of the script that applies them. Running them in the order of their numbers regenerates every file in data/ and tables/ from the publicly available inputs. data/ — every intermediate and final product in CSV and JSON format: the measurement envelopes, the monthly water area with its observation-coverage diagnostics, the area–volume calibrations, the reanalysis meteorological series, the bias diagnosis against the national product, the comparison of climate routes, the water-to-grass ratio by body and month, the bias-corrected depth series, the extended area series with its validation, the Priestley–Taylor contrast, the elevation of each body, and the monthly and annual loss series. tables/ — the five tables of the manuscript and the five of its supplement, in eleven files because the annual-loss table is split in two, each in comma-separated and LaTeX format. Primary inputs, not redistributed JRC Global Surface Water, Monthly History v1.4 — European Commission Joint Research Centre ERA5-Land monthly aggregated — Copernicus Climate Change Service PISCOeo_pm reference evapotranspiration — SENAMHI, Peru Copernicus DEM GLO-30 — European Space Agency Daily reservoir storage — National Water Authority of Peru (SNIRH) The first, second and fourth are accessed through Google Earth Engine, which requires a free account. This deposit: https://doi.org/10.5281/zenodo.21745042 License. Code under the MIT License; derived data and tables under Creative Commons Attribution 4.0 International (CC BY 4.0).