Speaking: Ingmar Nitze, Alfred Wegener Institute
The Permafrost Discovery Gateway hosts a monthly webinar series on the second Thursday of each month at 9:00 am Alaska time, raising topics of interest to the permafrost community. The webinar aims to 1) connect the international science community interested in big data remote sensing of permafrost landscapes, and 2) provide the Permafrost Discovery Gateway development team with end-user stories (by the presenter and webinar participants), such as exploring tools the community needs to create and explore big data.
Abstract
In this session of the Permafrost Discovery Gateway (PDG) seminar series, we will provide an insight into ongoing work towards the goal to quantify and monitor permafrost region disturbances, such as lake dynamics or retrogressive thaw slumps. With increasingly available processing capacities, and remote sensing data we are now able to extract information at unprecedented spatial and temporal scales. For mapping lake changes, we are expanding the processing of Nitze et al (2018), who used Landsat trends to map lake changes across 2.3M km² between 1999 and 2014 across four large transects in the arctic and subarctic permafrost region, to the entire arctic permafrost region and the period 2000-2020. For this processing we are leveraging Google EarthEngine and NCSA’s clowder processing pipeline. Furthermore, we are working towards an inventory of retrogressive thaw slumps (RTS) across the arctic, based on 3 meter PlanetScope imagery. Here, we leverage the latest advances in deep-learning object segmentation techniques. First analysis revealed the potential, but also limitations of this method and a limited set of training data. In a community approach (IPA Action Group “RTSInTrain”), we are formalizing training label creation and are constantly collecting training labels to overcome these limitations. The iteratively trained models with more and more new training labels show a constant improvement, giving us a good confidence for a good quality output. The first version will include a large area across the arctic covering ~ 500 km² of RTS rich terrain.