By Christos Sotiropoulos (Greece)
The objective of the thesis is to develop an approach to combine community-based monitoring (CBM) data with SAR (Synthetic Aperture Radar) and optical time series data streams for near real-time deforestation detection in Kafa, Ethiopia. Combining CBM and SAR-optical data promises to overcome the current limitations of optical-only forest monitoring systems that show limited temporal detection accuracies in tropical regions affected by persistent cloud cover. The integration-method builds upon an existing probabilistic (Bayesian) approach that was initially designed to combine optical and SAR time series (Reiche, de Bruin, Hoekman, Verbesselt, & Herold, 2015). Incorporation of CBM observations to this approach is the main core of the thesis. The proposed method consists of three parts; first we assess the capabilities of optical (Landsat NDVI) and SAR (ALOS PALSAR L-band, Sentinel-1 C-band) to separate forest and non-forest in Kafa. Then, the method is tested in a single pixel to evaluate the integration of the data streams, before applied in a larger area. Finally, we assess the spatial and temporal deforestation detection accuracies using reference data for the method development area. Application of the Bayesian approach in a single pixel showed that CBM observations improved the detection of deforestation 5 days earlier than with optical data streams. However, the overall contribution of CBM observations over the area was minimum (+0.35% overall accuracy, -0.12 days temporal accuracy) compared to optical data streams; as a result of their low temporal density and mostly delayed recording. Overall, CBM observations showed high potential for contribution on early deforestation detection, if only, CBM recordings are frequent.
Keywords: Community-based monitoring; remote sensing; near real-time; deforestation; Bayesian approach; Landsat; ALOS PALSAR; Sentinel-1; Kafa; Ethiopia