TABLE 4.

Evaluation of model performance on the test set for single-day models (trained on data at late booting stage) and 3D-CNN.

Model	Fold	RMSE (Mg/ha)	MBE (Mg/ha)	MAE (Mg/ha)	R 2
Null	A	0.93	−0.50	0.79	n.d.
	B	0.75	0.08	0.57	n.d.
	C	0.80	0.30	0.60	n.d.
	D	0.83	0.13	0.62	n.d.
	Mean	0.83	0.00	0.65	n.d.
Linear	A	1.06	−0.78	0.93	0.17
	B	0.70	0.07	0.51	0.18
	C	0.94	0.59	0.70	0.15
	D	0.83	0.10	0.63	0.04
	Mean	0.88	−0.01	0.69	0.14
XGBoost	A	0.88	−0.45	0.72	0.11
	B	0.74	−0.02	0.53	0.19
	C	0.83	0.50	0.61	0.22
	D	0.83	0.11	0.63	0.04
	Mean	0.82	0.04	0.63	0.14
2D-CNN	A	0.73	−0.19	0.57	0.18
	B	0.68	0.13	0.47	0.29
	C	0.63	0.02	0.45	0.30
	D	0.84	0.28	0.62	0.10
	Mean	0.72*	0.06*	0.53	0.22
3D-CNN	A	0.80	−0.37	0.67	0.08
	B	0.67	0.17	0.48	0.27
	C	0.77	−0.24	0.61	0.37
	D	0.90	0.45	0.67	0.06
	Mean	0.79	0.00	0.61	0.20

*Indicates significant difference in mean value between 2D-CNN and linear model only (p ≤ 0.05; one-way ANOVA).

RMSE, root mean squared error; MBE, mean bias error; MAE, mean absolute error;R², coefficient of determination. n.d., not defined (observed ∼ predicted yield is a vertical line for the null model).