Accuracy evaluation of evapotranspiration estimated by inverse analysis using FLUXNET2015 at forest sites

Abstract

AbstractConventionally, the Bowen ratio method and eddy covariance are used to measure evapotranspiration (ET); however, they have limited accuracy. Inverse analysis (IA), a novel method presented here, can estimate evapotranspiration (ETa) using commonly measured climatic variables. However, because actual ET data are limited, the accuracy of IA remains uncertain. Recently, IA accuracy has been improved, thus, observed latent heat flux (lEobs) in FLUXNET2015 data, a global dataset, and estimated values can be compared. Herein, we aim to evaluate the accuracy of the IA method. Eight forest sites in the dataset during 2000–2014 were selected as test sites, and hourly, daily, monthly, and yearly comparisons of lEobs and ETa were conducted using regression analysis and root mean square error (RMSE) calculations. The results are as follows: ① Comparison of lEobs‐hourly and ETa for all sites showed a similar pattern, with an average root mean square error (RMSE) of 0.0053 mm h−1. ② Comparison of lEobs‐daily and ETa showed high correlation coefficients, with slopes ranging from 1.016–0.717 and an average RMSE of 1.08 mm h−1. ③ Comparison of lEobs‐monthly and ETa showed similar patterns, with slopes ranging from 1.042–0.811 and an average RMSE of 17.9 mm m−1. ④ The slope of lEobs‐yearly and ETa was 0.988 and the average RMSE was 71.0 mm year−1. ⑤ The average of lEobs‐yearly and ETa showed a strong relationship, with a slope of 0.992. ⑥ The differences between the time steps of lEobs and ETa were evaluated qualitatively using RMSE/lEobs‐yearly, with hourly–yearly differences of 8.4%, 70.2%, 40.4%, and 14.2%, respectively. The above results indicate that IA can be used for an accurate estimation of ET. The findings of this study contribute significantly to water management and a deeper understanding of the hydrologic cycle.