Table 4. Hyperparameter optimisation overview.

Tested and selected hyperparameters for each of the sub-models included in the FAPAR EML meta-estimator.

Model	Hyperparameter	Lower limit	Upper limit	Selected
Extremely randomized trees	Number of estimators	10	100	44
	Maximum tree depth	5	100	92
	Maximum number of features	0	1	0.84
	Minimum samples for splitting	2	100	16
	Minimum samples per leaf	1	10	2
Gradient descended trees	Number of estimators	10	100	81
	Maximum tree depth	3	100	50
	Alpha	0	2	1.19
	Reg Alpha	0	0.2	0.007
	Eta	0	2	1.999
	Reg_Lambda	0	0.2	0.12
	Gamma	0	2	0.05
	Learning rate	0	0.2	0.06
	colsample_bytree	0	1	0.88
	colsample_bylevel	0	1	0.66
	colsample_bynode	0	1	0.47
Artificial neural network	Epochs	–	–	10
	Batch size	–	–	256
	Learning rate	–	–	0.0005
	Number of layers	–	–	4
	Number of neurons	–	–	128
	Activation	–	–	ReLu
	Dropout rate	–	–	0.15
	Output activation	–	–	Sigmoid