Project page for PhD research

The application of machine learning methods to satellite data for the management of invasive water hyacinth.

Monitor Water and Aquatic vegetation

Untitled Document.md

Find the Google Earth Engine Application here. The use of this application requires an internet connection, web browser on a computer or mobile phone.

How to use the application:

  1. Once you go to the link above of the application, it will open in your browser. Once the application is opened, the user needs to apply a spatial (Step 2) and temporal filter (Step 3) to monitor their site of interest for the period of interest.

  2. Spatial filter - Thereafter, the user needs to zoom in to their region of interest and draw a Bounding box arounf their waterbody of interest or a polygon around the water body of interest.

  3. Temporal Filter - This involves setting a start and end date. The start and end date defines the period over which the satellite data is processed for the detection of water and aquatic vegetation. The area of these pixels are then aggregated (sum) to provide a:

    i) Percentage of aquatic vegetation cover for the region of interest
    ii) Provide the Area of water within the defined region of interest
    iii) The Normalised Difference Vegetation Index (NDVI) of the detected aquatic vegetation - A proxy of plant greeness and vegetation health

Untitled Document.md

Change logs

V1 (2020, version released with research paper)

  • Application allows for the monitoring of aquatic vegetation, water and NDVI.

V1.1

– Improved scalability- shorter execution times, reduced number of OOM errors.

  • Instead of using the inctroduced (Singh et al., 2020). The GSW provides more accurate water detection across a wider variety of landscapes. Moreover, since it is readily available on GEE, it reduces computational requirements and results in performance gains. A drawback is that it is not up-to-date and resulted in the use of outdated water extents for monitoring recent conditions.

V1.2

– Reduced the sensitivity to the window parameter by using most recent max water extent.

  • Fixed date bugs. Previously, depending on the dates selected, an empty water image would be returned. By selecting the most recent (in relation to the landsat/sentinel image capture date) maximum extent of water up to 24 months ago (by default) we remove the need (now optional) for users to specify a window and compensate for the lagged update of GSW updated data by selecting a long enough window period.
  • Added a button, to zoom to a user’s location.

V1.3 (10/06/2022)

– Updated app to use the dynamic world dataset with a much more frequent cadence (between 2-6 days)

• Uses the dynamic world dataset instead of the GSW dataset for water detection. The ability to use the dynamic world datasets allows users to get a much more frequent water extent (closer to 2-6 days) and potentially more accurate aquatic vegetation extent. However, there is a tradeoff with accuracy in some cases whereby terrestraial pixels are detected as aquatic vegetation.
• Create This webpage for tracking changes to the app and providing additional information to users.

To do list:

• Combine all Landsat, S1 and S2 data for monitoring.
• Separate algae and macrophytes.
• Download single csv with water extent, NDVI and veg cover data.
• Add a contact page.
• Add a link to the code
• Add a link to explore hyacinth distribution
• Add a link to project website.
• Add option for user defined bounding box/asset. Create and add a project logo.

Find the code used in the research

Code used to create the application and research code.

Learn more