Econumerics recently completed a pilot study to review an existing geospatial dataset, documenting the extent of natural river pools in the Pilbara. Previous GIS analyses had been conducted using various criteria, data sources, and processing workflows.
Due to the two distinct local seasons (dry and wet), the natural pools along the Pilbara rivers include permanent and ephemeral bodies of water, mostly shallow and often highly discoloured, or murky. Without appropriate methods, these characteristics can seriously thwart the classification efforts based on remote sensing. Algal blooms, shadows, outcrops, and riparian canopy cover, for instance, are only some examples of the numerous sources of error that can prevent the correct definition of the features’ edges, size, and overall properties. This makes documenting particularly challenging and subjective, unless reliable and reproducible methods are adopted.
The present project aimed at exploring alternative classification strategies (including machine learning), and detect distinct water signals from remote sensing data products.