Water quality for lakes nobody monitors

Bringing satellite and machine-learning methods to data-deficient lakes at regional and national scale.

The problem

Most of New Zealand’s lakes have little or no routine monitoring. They are invisible to state-of-environment reporting, yet many are under pressure. You cannot manage what you cannot measure.

The approach

Satellite remote sensing extends water-quality estimates to unmonitored lakes; multi-task and transfer-learning models borrow strength across lakes so that data-rich sites help inform data-poor ones.

• Chlorophyll-a and clarity from Sentinel-2/3 and Landsat
• Optical water-type prediction from catchment properties
• Lake-level multi-task learning to handle sparse in-situ labels

The outcome

Consistent water-quality estimates across hundreds of otherwise data-deficient lakes — a foundation for prioritisation and for closing monitoring gaps where they matter most.

Builds on Lake by Lake, Globally: Enhancing Water Quality Remote Sensing with Multi-Task Learning Models and Colour classification of 1486 lakes across a wide range of optical water types.