Remote sensing of selective logging impact for tropical forest management

Abstract

The study objective is to investigate the feasibility and the technical specifications required to develop an operational system in order to detect and monitor the selective logging activities in tropical forests. The research relies on a detailed understanding of the remotely sensed signal of natural and logged tropical forests. The experimental study is based on a 100,000 ha semi-deciduous forest located in the south-western part of the Central African Republic. This forest includes a protected area and an area allocated to timber production. Two Landsat Thematic Mapper (TM) images, six SPOT multi-spectral (XS) images and field information were used for the analysis. The 5-year interval between the two Landsat-TM images made it possible to study the forest regeneration dynamics. The SPOT XS time series was used to assess the respective influence of the season, the spatial resolution and the geometry of acquisition. While recent impact of tree fall can be detected, there is no way to identify the cause of a single tree fall, i.e. whether it was due to the natural forest dynamics, storm damage or logging. Nevertheless, logging activities were identified due to the local concentration of large gaps and the linear features of the skid trail. The logging trails are clearly detected but with different degrees of success depending on the image geometry. The seasonality and hazy atmospheric conditions did not significantly affect the detectability of the trails. The geometry of observation, i.e. the sensor viewing and sun illumination angles, appeared to be a key factor. More than 40% of the trail network can be missed when the geometry of acquisition is not suitable. Despite rapid canopy closure over the logging roads and skid trail surface, more than 50% of the network was still visible 5 years after the last timber extraction. However, the spatial resolution of the images became critical as soon as the contrast between trails and the forest vanished. On a 120 m regularised image recorded right after the logging, more than 95% of the logging trail network was still visible, but only 34% was detected on a 5-year post-harvest image. These results led to the definition of the main specifications for a remote sensing-based system for monitoring and controlling the logging activities in tropical forests.