Geo For Good 2020 virtual meetup
|
|
The project team shared lessons from automated land cover mapping for restoration assessment in Indonesia using the Google Earth Engine (GEE) platform.
|
|
Adopting and applying automated mapping approaches in national contexts require, among others, a database of training data, containing land cover information with sufficient and country-relevant thematic details and definitions. Further, remote sensing in the region presents a challenge in creating optimal cloud-free optical images as the basis for both visual interpretation (to leverage time series information in identifying complex land cover classes) and for automated image classification.
During the Workshop, the project team presented automated land cover mapping methodology based on machine learning and cloud computing, developed in the Google Earth Engine (GEE) as part of RESTORE+ activities in Indonesia. Experiences were also shared on implementing the methodology using crowdsourced data collection. It was demonstrated how this approach addresses multiple challenges such as detailed land cover typology requirement for policy-relevant assessment of forest and landscape restoration, lack of reference data for training ML (supervised machine learning, ML) and map accuracy assessment, and persistent cloud cover affecting the optical satellite data (for more details on the tool see RESTORE+ brief “Using machine learning to map Indonesia’s restoration potential”)
With a live demo, using prepared custom GEE apps/scripts, technical solutions to address these challenges in obtaining training data for detailed land cover typology and generating cloud-free composite images, were demonstrated. The virtual meetup ended with discussions on collaboration opportunities, leveraging on the participatory products of the RESTORE+ project which aim to be continuously improved by the GEE community and crowd participation.
More info:
The session was recorded and can be re-watched here: https://www.youtube.com/watch?v=JhtLs2sPPTk
Link to presentation slides: https://drive.google.com/drive/u/0/folders/1DX8qiJ2eLIygKp0CxZj827CrSBBot-_w
Link to apps demoed:
During the Workshop, the project team presented automated land cover mapping methodology based on machine learning and cloud computing, developed in the Google Earth Engine (GEE) as part of RESTORE+ activities in Indonesia. Experiences were also shared on implementing the methodology using crowdsourced data collection. It was demonstrated how this approach addresses multiple challenges such as detailed land cover typology requirement for policy-relevant assessment of forest and landscape restoration, lack of reference data for training ML (supervised machine learning, ML) and map accuracy assessment, and persistent cloud cover affecting the optical satellite data (for more details on the tool see RESTORE+ brief “Using machine learning to map Indonesia’s restoration potential”)
With a live demo, using prepared custom GEE apps/scripts, technical solutions to address these challenges in obtaining training data for detailed land cover typology and generating cloud-free composite images, were demonstrated. The virtual meetup ended with discussions on collaboration opportunities, leveraging on the participatory products of the RESTORE+ project which aim to be continuously improved by the GEE community and crowd participation.
More info:
The session was recorded and can be re-watched here: https://www.youtube.com/watch?v=JhtLs2sPPTk
Link to presentation slides: https://drive.google.com/drive/u/0/folders/1DX8qiJ2eLIygKp0CxZj827CrSBBot-_w
Link to apps demoed:
- Interactive training data collection: bit.ly/training-data-collection . Note that we added a feature allowing users to upload previously downloaded geojson (by copy pasting the geojson string).
- Landsat composites explorer: bit.ly/landsat-composites-explorer
- Interactive map feedback: bit.ly/collect-polygons (Script: bit.ly/collect-polygons-script)
