Monitoring selective logging intensities in central Africa with sentinel-1: A canopy disturbance experiment

• Published in: Remote sensing of environment Vol. 298; p. 113828
• Main Authors: Dupuis, Chloé, Fayolle, Adeline, Bastin, Jean-François, Latte, Nicolas, Lejeune, Philippe
• Format: Journal Article Web Resource
• Language: English
• Published: Elsevier Inc 01.12.2023; Elsevier
• Subjects: cameras; Cameroon; canopy gaps; Central Africa; certification; data collection; environment; Environmental sciences & ecology; flight; forest canopy; forest management; forests; Life sciences; Logging intensity; remote sensing; Sciences de l’environnement & écologie; Sciences du vivant; Sentinel-1 SAR time series; synthetic aperture radar; time series analysis; Tropical forest degradation; UAV images; Cameroon; Sentinel-1 SAR time series; Logging intensity; Tropical forest degradation; Central Africa; UAV images
• ISSN: 0034-4257; 1879-0704; 1879-0704
• DOI: 10.1016/j.rse.2023.113828

Forest degradation is a major threat to tropical forests, and effective monitoring using remotely sensed data is subject to significant challenges. In particular, consistent methods for detecting subtle changes in the forest canopy structure caused by selective logging are lacking. Here, using a unique dataset collected in southeastern Cameroon, covering over 22,000 ha of monthly harvesting areas, >6000 locations of harvested trees, and an independent canopy gap dataset developed from an uninhabited aerial vehicle flight (UAV; RGB camera) of approximately 1500 ha, a new method was designed to monitor canopy disturbance and logging intensity in Central Africa. Using Sentinel-1 synthetic aperture radar (SAR) data, the method was conceptualised using a two-step, two-scale approach, which better matched logging practices. First, (non-)harvesting activity areas were identified using textural indices at a spatial resolution of 300 m (step 1), and within these harvesting activity areas, canopy gaps were detected at a resolution of 10 m (step 2). Both steps were based on monthly differences in the Sentinel-1 SAR time series computed using the average of the 12 months preceding and the average of the three months following the month of interest. This method identified harvesting activity areas (step 1 at 300 m resolution) of over 12,004 km2 with high accuracy (omission and commission errors for both classes ≤0.05) and, within them, detected canopy gaps (step 2 at 10 m resolution) with a global accuracy of 0.89. Although some canopy gaps were subject to omission and commission errors (0.39 and 0.05, respectively), this method yielded better results than other available approaches. Compared to the UAV canopy gaps, this method detected most of the small gaps (≤ 500 m2), which represent 80% of all disturbed areas, whereas other available approaches missed at least 70% of these and consequently missed most of the disturbance events occurring in a selectively logged forest. Furthermore, the predictions were correlated with logging intensity, i.e., the number of trees and volume cut per hectare, which are two important criteria for assessing the sustainability of logging activities. This two-step two-scale method for short-term, monthly monitoring of logging disturbances and intensity has strong practical implications for forest administration and certification bodies in Central Africa. •A two-step, two-scale approach was adapted to logging practices in Central Africa.•Low intensity logging (16 m3/ha) and subtle changes in canopy were detected.•Logging intensity (trees/volume per ha) using remote sensing data can be predicted.•Independent validation with an accurate dataset from an uninhabited aerial vehicle.•Current systems underestimate forest disturbances by 72%–80%.