Forest restoration monitoring through digital processing of high resolution images

• Main Author: Reis, Bruna Paolinelli
• Publication Date: 2019
• Other Authors: Martins, Sebastião Venâncio, Fernandes Filho, Elpídio Inácio, Sarcinelli, Tathiane Santi, Gleriani, José Marinaldo, Leite, Helio Garcia, Halassy, Melinda
• Format: Article
• Language: eng
• Source: LOCUS Repositório Institucional da UFV
• Download full: https://doi.org/10.1016/j.ecoleng.2018.11.022; http://www.locus.ufv.br/handle/123456789/23963
• topic: Light Detection and Ranging (LIDAR); Unmanned Aerial Vehicle (UAV); Random Forest (RF); Maximum likelihood algorithm; Recovery of degraded areas; Forest restoration

Monitoring and evaluating forest restoration projects is a challenge especially in large-scale, but the remote monitoring of indicators with the use of synoptic, multispectral and multitemporal data allows us to gauge the restoration success with more accurately and in small time. The objective of this study was to elaborate and compare methods of remote monitoring of forest restoration using Light Detection and Ranging (LIDAR) data and multispectral imaging from Unmanned Aerial Vehicle (UAV) camera, in addition to comparing the efficiency of supervised classification algorithms Maximum Likelihood (ML) and Random Forest (RF). The study was carried out in a restoration area with about 74 ha and five years of implementation, owned by Fibria Celulose S.A., in the southern region of Bahia State, Brazil. We used images from Canon S110 NIR (green, red, Near Infrared) on UAV and LIDAR data composition (intensity image, Digital Surface Model, Digital Terrain Model, normalized Digital Surface Model). The monitored restoration indicator was the land cover separated in three classes: canopy cover, bare soil and grass cover. The images were classified using the ML and RF algorithms. To evaluate the accuracy of the classifications, the Overall Accuracy (OA) and the Kappa index were used, and the last was compared by Z test. The area occupied by different land cover classes was calculated using ArcGIS and R. The results of OA, Kappa and visual evaluation of the images were excellent in all combinations of the imaging methods and algorithms analyzed. When Kappa values for the two algorithms were compared, RF presented better performance than ML with significant difference, but when sensors (UAV camera and LIDAR) were compared, there were no significant differences. There was little difference between the area occupied by each land cover classes generated by UAV and LIDAR images. The highest cover was generated for canopy cover followed by grass cover and bare soil in all classified images, indicating the need of adaptive management interventions to correct the area trajectory towards the restoration success. The methods employed in this study are efficient to monitor restoration areas, especially on a large scale, allowing us to save time, fieldwork and invested resources.