Skip to content

supervised continuous change detection in a shrubland ecosystem

License

Notifications You must be signed in to change notification settings

GMoncrieff/renosterveld-monitor

Repository files navigation

Shrubland change detection using neural networks

Code for the paper 'Continuous Land Cover Change Detection in a Critically Endangered Shrubland Ecosystem Using Neural Networks' by Glenn Moncrieff

https://www.mdpi.com/2072-4292/14/12/2766/htm

Google Earth Engine is required to run notebooks. Code was run on a GCP Vertex AI Workbench VM.

01_data_export.ipynb exports train, test and valid data to google cloud storage

02_model_fit.ipynb fits models using tf2 with preselected parameters values

03_predict.ipynb uses saved model to predict for a specific date over a region and upload results to earth engine for visualization

04_salient.ipynb calculates saliency using grad-CAM++ on temp-CNN model

Global variables defining region, dates, parameters, filenames, credentials etc are defined in utils/globals.py

Code for the operational prediction pipeline implemented using google cloud functions, cloud run and cloud dataflow can be found at https://github.com/mgietzmann/global_renosterveld_watch.
This prediction pipeline makes predicitons of land cover change and uploads them to Earth Engine every 20 days. Results can be viewed at https://glennwithtwons.users.earthengine.app/view/global-renosterveld-watch

About

supervised continuous change detection in a shrubland ecosystem

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published