Thesis subject

MSc thesis topic: Sentinel-1 radar coherence for mapping of tropical forest loss

Tropical forests are the lungs of the earth and tropical deforestation is one of the major anthropogenic carbon sources. Deforestation occurs at various scales, with commercial loggings often affecting hectares of forest, while selective logging typically consists of the removal of selected trees. In order to effectively detect deforestation and enforce international agreements, a high-resolution monitoring tool is needed.

In the past 10 years, satellite-based alert systems (Reiche et al., 2021, Hansen et al. 2016) have emerged as the primary tool to provide near real-time information on newly disturbed tropical forest areas. However, persistent cloud cover in the tropics limits the capacity of optical-based systems (e.g. GLAD alerts; Hansen et al, 2016). As a consequence, radar-based systems have gained importance. The new RADD (Radar for Detecting Deforestation) alerts (Reiche et al., 2021) uses 20-metre resolution data from the European Sentinel-1 radar satellites to provide weekly disturbance information in the humid tropics.

While the radar-based forest monitoring systems show good performance in the humid tropics, where seasonality and dynamics of forest are moderate, accuracies are poorer in temperate and boreal forest areas, where seasonality is more pronounced. Most radar-based systems use backscatter intensity as the primary observable, which is known to be sensitive to moisture variations.

One radar observable, potentially less sensitive to moisture, is interferometric coherence. Coherence is the degree of similarity between two complex-valued radar images, and it can be estimated from Sentinel-1 single-look complex (SLC) data. Coherence has a well-documented sensitivity to forest change, in particular abrupt change occurring between two images (Akbari and Solberg, 2020, Rizolli et al., 2018). However, reliable coherence estimation typically results in coherence images with coarser resolution than for backscatter intensity.

This thesis begins with selecting and setting up processing chains for coherence and backscatter intensity estimation from Sentinel-1 SLC data. The developed processing chains should ascertain low bias in the estimated quantities, while maintaining a good spatial resolution. ESA Snap software is the recommended tool for this task.

Next, two areas of interest are chosen and studied in detail: one in a forest with relatively low seasonality (e.g., humid tropical forest) and one in a forest with a significant seasonality (e.g., temperate or boreal forest). The student assesses the potential of Sentinel-1 coherence and backscatter for mapping and detection of logging in both test sites, and at different scales. The aspects of detection accuracy and resolution (both spatial and temporal) should be considered.

Finally, the student develops algorithms for detection of selective logging in these two areas, based on coherence, backscatter, and the combination of both. The algorithms are assessed against available reference data, and a conclusion is made about their large-scale applicability and usefulness.

Software: ESA Snap, R/Python/Matlab.

Objectives

  • Generate coherence and backscatter time series for two test sites with different degrees of seasonality.
  • Study and compare the time series of coherence and backscatter with respect to reference data
  • Develop algorithms for detection of change from coherence, backscatter, and the combination of both

Requirements

  • Advanced Earth Observation course
  • Geo-scripting course (Good knowledge in scripting is an asset; e.g. Google Earth Engine, R, python, java script)

Theme(s): Sensing & measuring; Modelling & visualisation