Information Entropy-Based Hybrid Models Improve the Accuracy of Reference Evapotranspiration Forecast

Table 2

Performance metrics of single and hybrid models from January 2 to February 1, 2022, using the Kruskal–Wallis test.

Models	SVR	Bayesian	Lasso	Ridge	Information entropy	Variance reciprocal
Observation (mm d^−1, median ± MAD)	2.058 ± 1.101
Forecast (mm d^−1, median ± MAD)	2.113 ± 1.092	2.282 ± 0.998	2.136 ± 0.814	2.27 ± 1.043	1.941 ± 1.041	1.787 ± 0.931
Accuracy (%)	93.3	94.2	92.3	91.4	97.5	94.4
Precision	0.9331	0.9421	0.9472	0.9136	0.9618	0.9475
F1 score	0.9413	0.9561	0.9543	0.9273	0.9742	0.9532

Note. Bold values indicate the best performance among SVR, Bayesian, ridge, lasso, information entropy, and variance reciprocal models. MAD is the median absolute deviation around the median.