Introduction

Drones allow collecting high-spatial resolution images useful for monitoring forest vegetation dynamics in managed forests. There are, however, doubts about the most effective way to use them concerning spatial resolution.

Objective

To identify the optimal spatial resolution of multispectral images captured by drones for mapping land cover types in managed temperate forests in Hidalgo, Mexico.

Materials and methods

Spectral images were preprocessed at spatial resolutions from 0.2 to 2.5 m, at 0.1 m intervals. Pine, oak, other broad-leaved trees, herbs and bare soil cover were classified with the Random Forest algorithm. The effect of spatial resolution on land cover classification was evaluated using the Kruskal-Wallis non-parametric test followed by a Mann-Whitney-Wilcoxon post-hoc comparison (p < 0.05). Classification errors of land cover classes were analyzed graphically.

Results

0.2 m spatial resolution images provided the highest land cover classification accuracy (96 %) but was statistically similar to that of 0.7 m (p = 0.3984). The lowest accuracy (82 %) was obtained with 2.5 m spatial resolution imagery. Omission and commission errors were lower and consistent in classifications with 0.2 to 1.2 m spatial resolution images.

Conclusion

Multispectral images (0.7 m resolution), acquired with a fixed-wing drone, allowed us to classify the land cover/vegetation types and the exact spatial distribution of pine, oak and other hardwood species in a temperate forest under forest management.

        · abstract in Spanish     · text in English | Spanish     · English ( pdf ) | Spanish ( pdf )