Skip to Content
HydrologyHydrology
  • Article
  • Open Access

20 November 2026

33 Pages

Provenance-Aware Evaluation of Reference Evapotranspiration Methods at a Humid Tropical Site Using Random Forest and ERA5-Seamless Reconstruction

,
and
1
Civil and Agricultural Department, Faculty of Engineering, Universidad Nacional de Colombia, Bogotá 111321, Colombia
2
Corporación Colombiana de Investigación Agropecuaria-AGROSAVIA C.I.T la Libertad, Villavicencio 500008, Colombia
*
Author to whom correspondence should be addressed.

Abstract

Assessing reference evapotranspiration (ETo) in humid tropical areas is often difficult because meteorological records are incomplete. This study compared four simplified methods with the FAO-56 Penman–Monteith (PM) method, depending on the source of meteorological data at the La Libertad station in the Colombian Orinoquía between 2018 and 2025. We checked hourly records for quality and filled missing data using a multistage approach that combined Random Forest and ERA5-Seamless reanalysis. Of 2713 complete days, only 527 (19.4%) were fully quality-checked, and none were from 2021 to 2024. In blocked validation, the hybrid reconstruction lowered hourly errors by 5–15% compared to calendar-based estimates and by 10–42% compared to raw reanalysis. The average PM ETo was 3.78 mm/day. The Priestley–Taylor method, when using the same net radiation as PM, matched daily PM results most closely (RMSE = 0.54 mm/day; 95% CI 0.49–0.60; PBIAS = +6.5%). Hargreaves–Samani and a daily Thornthwaite adaptation gave values about 20% higher than PM, and the simplified Blaney–Criddle was 44% higher. Monthly, Priestley–Taylor and conventional Thornthwaite gave nearly the same results (RMSE = 14.8 and 14.9 mm/month). The daily ranking of methods was the same on fully quality-checked days and when ETo was calculated from either observed or reconstructed data, though reconstructed data led to daily PM errors of about 1 mm/day. Priestley–Taylor should be considered only as a preliminary alternative when radiation, temperature, and humidity data are available, and further local validation is needed.

Article Metrics

Citations

Article Access Statistics

Article metric data becomes available approximately 24 hours after publication online.