Item request has been placed! ×
Item request cannot be made. ×
loading  Processing Request

Assessing forest degradation in Bajo Calima, Colombia, from multi-frequency and multi-temporal Synthetic Aperture Radar (SAR)

Item request has been placed! ×
Item request cannot be made. ×
loading   Processing Request
  • Additional Information
    • Publication Information:
      University of Leicester, 2020.
    • Publication Date:
      2020
    • Collection:
      University of Leicester
    • Abstract:
      The degradation of tropical forests is a significant problem transforming these ecosystems, contributing to greenhouse emissions and biodiversity loss. However, quantifying the impact is difficult, because different factors can indicate that a forest is degraded, and there is a wide variety of definitions and drivers. The motivation of this research is to contribute to the technical challenge of mapping and measuring forest degradation by evaluating the variations in forest structure of a degraded tropical forest and explore the potential of SAR for accurately achieving this goal. The forests of Bajo Calima – Colombia are under pressure from unplanned gold mining and selective logging at different intensities, which has changed their structural properties. In this thesis, I investigated the structural variabilities in the forests of Bajo Calima, and connected them to forest degradation using a combination of multi-temporal in-situ data, light detection and ranging (LiDAR) data, multi-frequency and time series SAR data. Firstly, forest structure data collected from field plots in 2014 and 2017 were classified using K-means cluster analysis. Aboveground biomass (AGB), tree density and basal area and trunk diameter (DBH) were the variables that better explained the variations in forest structure, and with a precision of 0.77% four levels of disturbance were identified in the area: low to severely degraded forest. Secondly, I investigated the potential of integrating multi-frequency SAR data: ALOS PalSAR-2, Sentinel-1, interferometric coherence, and the digital elevation model (DEM) from TanDEM-X, in combination with LiDAR and field data to retrieve the forest structure parameters derived from the field data analysis. A Random Forest (RF) machine-learning approach was used to map AGB, tree density and basal area, and to classify the area into different levels of disturbance. Results showed a very good ability to retrieve information from different metrics of forest structure and categorise them according to different degrees of forest disturbance. Accuracies were higher than 90% for the regressions, while Kappa coefficients were above 0.80 for the classifications. Using the DEM derived from the X-band was key to achieving these higher accuracies. Finally, capabilities of ALOS PalSAR-1, ALOS PalSAR-2 (2007 – 2018) and Sentinel-1 (2014 – 2017) were assessed by integrating time series responses from multi-frequency SAR data in order to evaluate forest degradation as the significant variation in forest structure. Results were promising, demonstrating with high accuracy (≥ 88%) that by using the model derived from this study, it is possible to identify areas that have been considerably impacted by degradation over time.
    • Accession Number:
      10.25392/leicester.data.13322813.v1
    • Accession Number:
      edsble.819485