Invasive Species Detection

Invasive weeds spread fast and are expensive to survey on the ground. I wanted a way to flag outbreaks from space before they became unmanageable.

I built a pipeline that stacks Sentinel-1 radar and Sentinel-2 optical time series, then trains a TensorFlow model to separate invasive species from surrounding vegetation using their seasonal signature.

• Reached 90% detection accuracy, ahead of the published state of the art at 82%.
• Fused radar and optical data to stay reliable through cloud cover.
• Ran the whole workflow on Google Earth Engine so it scales to large regions.

Delivered for a large mining company in Australia.