EGU General Assembly 2023
© Author(s) 2023. This work is distributed under
the Creative Commons Attribution 4.0 License.

Detecting Hydrological Barriers and Fragmentation in Wetlands using Deep Learning and InSAR

Clara Hübinger1, Etienne Fluet-Chouinard2, Gustaf Hugelius1, Francisco J. Peña1,3, and Fernando Jaramillo1,4
Clara Hübinger et al.
  • 1Stockholm University, Physical Geography, Sweden (
  • 2Institute for Atmospheric and Climate Science, Department of Environmental Systems Science, ETH Zurich, Zürich, Switzerland
  • 3KTH Royal Institute of Technology, Stockholm, Sweden
  • 4Baltic Sea Centre, Stockholm University, Stockholm, Sweden

The loss of hydrological connectivity and fragmentation of natural wetlands have driven widespread wetland degradation worldwide. Monitoring techniques are needed to assess the degree of fragmentation and to aid with the restoration of affected wetlands. Hydrogeodetic tools such as wetland Interferometric Synthetic Aperture Radar (InSAR) can be used to monitor wetland hydrology as it provides information on three-dimensional flow dynamics at a high spatial resolution. While this technique has been utilized previously for the manual assessment of hydrological connectivity in wetlands, this study proposes the first deep learning-based approach for the automated detection of barriers to the natural water flow that cannot otherwise be identified by conventional space imagery. To this end, a deep convolutional network is trained by segmenting edge features in ALOS PALSAR-1 L-Band InSAR images captured between 2006 and 2011. The training dataset consists of manually labelled and delineated barriers showing abrupt changes in water surface elevation and 22 wrapped interferograms with high coherence across several sample sites in the Everglades and the wetlands of southern Louisiana, United States. The scenes were processed in the Interferometric synthetic aperture radar Scientific Computing Environment (ISCE). The network is set up using a UNet structure with alternating convolutional and pooling or upsampling layers along a contracting and expanding part. The validation of the resulting pixel-wise segmentation shows that the network can successfully detect hydrological barriers in wetlands. Apart from identifying the location of barriers, the CNN can be applied to identify the type and persistence of the fragmentation over the entire wetland. Utilizing the multitemporal data additionally helps detect seasonal changes in the presence or absence of hydrological barriers in the sample sites. This study demonstrates the potential of deep learning techniques for the automated detection of hydrological parameters in InSAR imagery and sets the groundwork for the automated monitor of wetland fragmentation across the world.

How to cite: Hübinger, C., Fluet-Chouinard, E., Hugelius, G., Peña, F. J., and Jaramillo, F.: Detecting Hydrological Barriers and Fragmentation in Wetlands using Deep Learning and InSAR, EGU General Assembly 2023, Vienna, Austria, 24–28 Apr 2023, EGU23-14324,, 2023.

Supplementary materials

Supplementary material file

Comments on the supplementary material

to access the discussion