Comparison of smart applications for evapotranspiration assessment automation with manual estimation using the Thornthwaite method

Evapotranspiration assessment is critical for effective agricultural water management, as it is the primary factor influencing decisions regarding the amount, location, timing, and method of crop irrigation. Manual evapotranspiration calculations are time-consuming and labor-intensive, particularly for long-term assessments. Consequently, various automated approaches have been developed in recent decades to improve the speed and accuracy of evapotranspiration estimation. However, not all automated methods demonstrate sufficient reliability for practical applications. The primary objective of this study was to evaluate the performance of widely used and innovative evapotranspiration calculation tools, namely EVAPO, AgSAT, and Evapotranspiration Calculator (Ukraine), and compare their results with the reference Thornthwaite method. The study was conducted in Kherson Oblast from September 2022 to June 2025, utilizing the data from the regional hydrometeorological center and the default settings of the applications. The statistical analysis involved generating evapotranspiration distribution plots, box plots, scatter plots, residual plots, and Bland-Altman plots for each application, alongside calculations of mean differences, root mean square error (RMSE), mean absolute error (MAE), R-squared, and Nash-Sutcliffe Efficiency (NSE). The overall comparisons were conducted using Repeated Measures ANOVA or the Friedman Test, with the Bonferroni correction applied for multiple paired comparisons. The results revealed that the EVAPO application provided the closest evapotranspiration estimates to the reference method (RMSE = 0.78 mm, MAE = 0.60 mm, R-squared = 0.74, NSE = 0.72), while the AgSAT application exhibited the poorest performance. Additionally, AgSAT tended to underestimate evapotranspiration, whereas Evapotranspiration Calculator (Ukraine) tended to overestimate it. These tendencies should be considered when using these applications without prior calibration. The results yielded by ANOVA confirmed that all the evaluated approaches significantly differed from the reference manual estimations using the Thornthwaite method, indicating that even the best-performing application, EVAPO, requires refinement before practical implementation in irrigation scheduling.