Table 1 Average classification accuracies across three species of arborescent palms for assessing the robustness of the seven different approaches for model training and testing

Model	Training				Testing			Differences over*	Precision	Recall	F1-score
Model	Images from	Year	No. of UAV mosaics	No. of tiles used	Mosaic from	Year	No. of UAV mosaics	Differences over*	Precision	Recall	F1-score
1	Veinte de Enero site	2017	4	764	Veinte de Enero site	2017	1	ic	0,63	0,60	0,61
2	All sites	2017 + 2018	26	10505	Nueva York & 2 de Mayo de Muyuy	2019	2	ic, fc, sr	0,59	0,60	0,59
3	All sites	2018 + 2019	33	10420	Veinte de Enero & Parinari	2017	2	ic, fc, sr	0,86	0,46	0,45
4	All sites	2017 + 2019	34	11651	Jenaro Herrera & Iquitos	2018	2	ic, fc, sr	0,77	0,54	0,56
5	Around the National Reserve Allpahuayo Mishana	All years	13	5157	Around the National Reserve Pacaya - Samiria	2019	2	ic, sp, fl, gl	0,86	0,59	0,67
6	Within the Pastaza Marañon (PM) Foreland Basin	All years	79	27992	Nueva Jerusalen site	2019	2	ic, sp, fl, gl, td	0,72	0,77	0,72
Final	All the training sites from the Loreto region.	All years	81	29902	All the sites in Madre de Dios region	2019-2022	4	ic, sp, fl, gl, td	0,88	0,67	0,74

*ic iIllumination conditions, fc floristic composition, sr spatial resolution, gl geographical location, td amount of training data.
Source data are provided as a Source Data file.

Quick links

Search