Assessing reference evapotranspiration estimation from reanalysis weather products. An application to the Iberian Peninsula

Abstract

Computing crop reference evapotranspiration (ETo) with the FAO Penman–Monteith method (PM‐ETo) requires maximum and minimum air temperature, shortwave radiation, relative air humidity and wind speed. These data are often not available, thus requiring alternative computation procedures. Although some proposed approximations may provide ETo values with small estimation errors, the physics of the ET processes may then not be well described. The use of reanalysis data, which is common in climate studies, represents an alternative to observation data for the weather variables referred above, when these are not available. This study focuses on the use of the National Center for Environmental Prediction/National Center for Atmospheric Research (NCEP/NCAR) blended reanalysis products with gridded data sets for the computation of PM‐ETo in the Iberian Peninsula. A monthly time step was adopted. The PM‐ETo time series computed with the blended reanalysis data sets were compared with those obtained using observations for 130 weather stations in the Iberian Peninsula. Results show that the PM‐ETo computed with blended reanalysis compares well with the series computed from observation data (average root mean square error, RMSE = 0.49 mm day−1). The weather variables derived from reanalysis were also compared with observation data. Results supported the quality of ETo computations because, overall, there was a good match between solar radiation (average RMSE = 1.76 MJm−2 day−1) and maximum temperature (average RMSE = 1.48 °C) derived from reanalysis and in situ observations. By contrast, the wind speed from reanalysis highly overestimated observations and this is likely a reason for the slight overestimation of ETo computed from reanalysis (percentage bias, PBIAS>20% in 89% of cases). In addition, the reanalysis products are apparently influenced by modelled warming, which contributes to overestimation of the minimum temperature and, to a lesser extent, of the relative humidity. The spatial pattern of accuracy indicators reveals that poorer results correspond to the southern and south‐eastern coastal areas of Iberia, where climate is semi‐arid. The compatibility of the PM‐ETo computed with monthly inputs and of the daily ETo cumulated to the month using the PM‐ETo equation was confirmed, thus allowing to extend conclusions of this study to daily computations. Alternative reanalysis products were also assessed. Tests for ERA‐Interim reanalysis products revealed overestimation of ETo and those for NCEP/NCAR Reanalysis II have shown large underestimation. Results suggest that the blended reanalysis products are suitable for the estimation of ETo in Iberia since they integrate an appropriate correction of radiation and temperature, which proved essential for the good estimation results obtained.

Asynchronous Hydroclimatic Modeling for the Construction of Physically Based Streamflow Projections in a Context of Observation Scarcity

“…Climate reanalyses are reconstruction of past climate through the blending of observations with numerical models that are spatially and temporally complete, they are thus produced from a consistent physical representation of the climate processes. On the other hand, reanalyses are known to be biased relative to observations in a magnitude that varies locally (Li et al, 2013;Jones et al, 2017;Martins et al, 2017). Recently, reanalyses have been used to provide target distributions for statistical postprocessing of climate simulations (Grenier, 2018) and to force hydrologic models during calibration (Fuka et al, 2014;Lauri et al, 2014).…”

Trend analysis of evapotranspiration over Iran based on NEX‐GDDP high‐resolution dataset

“…The NSE is a normalized statistics that determines the relative magnitude of residual variance (noise) compared to the measured data variance (information) (Nash and Sutcliffe, 1970; Martins et al , 2017). In other words, it indicates the relative magnitude of the mean square error (MSE) and the observed data variance (Legates and McCabe Jr, 1999).…”

“…The Forced To the Origin (FTO) (Eisenhauer, 2003) linear regression is defined by y = b 0 x between predicted ( y ) and observed ( x ) values. This model assumes proportionality between NEX‐GDDP and in‐situ values; b o being the constant of proportionality where b o > 1 denotes overestimation and b o < 1 indicates underestimation (Martins et al , 2017). The b o is determined as follows: …”

“…The analysis, reanalysis, and other gridded datasets are examples of data with adequate spatial and temporal coverage that could potentially fill for missing observational data. Recently, a number of publications have used such datasets for evapotranspiration estimation (e.g., Raziei and Pereira, 2013b; Martins et al , 2017; Rajabi and Babakhani, 2018) or for spatial–temporal analysis of precipitation across Iran (Raziei et al , 2014; Sharifi et al , 2016).…”

Trend and Sensitivity Analysis of Reference Evapotranspiration in the Senegal River Basin Using NASA Meteorological Data

“…Therefore, in this study, data reanalysis and observation of the POWER project of NASA Langley Research Center (LaRC) funded by the NASA Earth Science/Applied Science program (https://power.larc.nasa.gov/data-accessviewer, accessed on 20 December 2018) were used as an alternative to observed data that are inaccessible and generally scattered [44,46]. These data, already used and validated in previous studies [47][48][49], have the advantage of having spatial and temporal coverage on a global scale [46,[50][51][52] and provide the climatic variables necessary for the estimation of evapotranspiration [46,53]. These datasets which result from satellite and model-derived weather data contain uncertainties however, and where possible these should be evaluated and validated locally against available in situ measurements [54].…”