Radar Remote Sensing

Our research utilizes radar remote sensing for unravelling human activities in and dynamics of forest ecosystems, and has a strong focus on near real-time change monitoring. We develop machine learning methods using radar and multi-sensor data to monitor change dynamics and their drivers more accurately and timely.

Satellite radar remote sensing uses long-wavelength energy that penetrates through clouds and is sensitive to changes in vegetation physical structure. These characteristics are major advantages for monitoring forest dynamics, in particular in tropical regions where cloud cover is persistent. New and near-future radar satellites provide spatially detailed information almost daily and cloud computing solutions now enable large-scale applications.

An alarming rate of human-induced forest loss combined with more frequent extreme climate events strongly increases the pressure on the Earth’s forest ecosystems. At the same time, countries and major commodity sectors commit to zero-deforestation pledges. We work on openly accessible and timely information on the dynamics and drivers of changes in forests urgently needed to empower sustainable land management, and law enforcement actions against illegal activities.

Our core research lines are:

  • Fundamental radar methods
  • Near real-time change monitoring
  • Characterising forest changes and dynamics

Current PhD and Postdoc projects

  • Dr. Adugna Mullissa (Postdoc): Radar and deep learning methods for near-real time forest change monitoring
  • Dr. Yaqing Gou (Postdoc): Assessing large-scale tropical forest change dynamics
  • Dr. Milutin Milenkovic (Postdoc): Monitoring tropical forest recovery and resilience using big data approaches for radar and lidar satellite data
  • Dr. Wanda de Keersmaecker (Postdoc): Monitoring tropical forest recovery capacity using Sentinel-1
  • Johannes Balling (PhD student): Characterizing fire-related tropical forest change dynamics utilizing multi-sensor remote sensing data
  • Bart Slagter (PhD Student): Detecting, characterizing and predicting small-scale forest disturbance using satellite data and artificial intelligence

Former colleagues

  • Dr. Christelle Braun (Postdoc; 2018 – 04/2020)

Key publications

Fundamental radar methods

Near real-time change monitoring

Characterising changes and dynamics

Available open-source software

‘Bayts’ R-package: https://github.com/jreiche/bayts.
Set of tools to apply the probabilistic machine learning approach of Reiche et al. (2015, 2018) to combine multiple Radar and/or optical satellite time series and to detect deforestation/forest cover loss in near real-time.

Code for Angular-based radiometric slope correction of Sentinel-1 on Google Earth Engine (Vollrath et al 2020): https://github.com/ESA-PhiLab/radiometric-slope-correction.

Code for deSpeckNet: Generalizing Deep Learning Based SAR Image Despeckling: https://github.com/adugnag/deSpeckNet

Code for CV-deSpeckNet (Mullissa et al, 2021) - Despeckling Polarimetric SAR Data Using a Multistream Complex-Valued Fully Convolutional Network: https://github.com/adugnag/CV-deSpeckNet

Code for Sentinel-1 SAR Backscatter Analysis Ready Data Preparation in Google Earth Engine (Mullissa et al, 2021): https://github.com/adugnag/gee_s1_ard