Table 9. Producer’s and user’s accuracy, Weighted F1-score, and Log loss of the ensemble predictions during spatial cross-validation.

Corine level	Number of classes	Prod acc.	User acc.	Weighted F1	Log loss	Baseline log loss	Log loss ratio
1	5	0.835	0.835	0.834	0.456	2.018	0.774
2	14	0.636	0.639	0.509	1.033	3.596	0.713
3	43	0.494	0.502	0.491	1.544	5.142	0.700