Sign In


Automated Identification and Enumeration of Midge Populations

Challenge Statement

How might we better monitor the populations of midges at reservoirs?

Background & Current Practice

PUB actively monitors the populations of non-biting Chironomids (midges) emerging from reservoirs to pre-empt any potential further increase in midges. Three methods are used as part of the monitoring: (i) traps are deployed on the reservoir water surface to capture midges that are emerging from the reservoir; (ii) oil-coated boards are set up around the reservoir to capture midges that are attached to them, and (iii) sediment grab samples are taken to measure for midges. Insects collected from the traps, oil-coated boards and sediments will then be manually sorted, identified to species level (if possible) and counted to determine if there is an increasing count in the number of adult midges and whether the main species present is known to cause a nuisance to nearby residential estates and businesses. These findings will inform the appropriate control measures to take.

Opportunities Areas & Key Considerations

We are interested in solutions that allow for an automated process of sorting, identifying, and counting of adult midges collected in traps, oil-coated boards, and sediments. Proposed solutions should help PUB carry out continuous monitoring of midges and scale up the coverage of existing monitoring while reducing the time and manpower spent on monitoring.

Key Challenges
  • There are over 70 species of non-biting midges in Singapore reservoirs, and their behavioural patterns may vary. Some species are known to cause significant discomfort and nuisance to residential estates and businesses that are close to some reservoirs. The response to each outbreak or emergence would vary depending on the nature of the species present.
  • Midges are very small, and some species may look very similar to each other. This makes the identification of species challenging.
  • During an emergence event, the numbers of midges collected can be very large, and this makes accurate sorting and counting a tedious and slow process.

After selection

  • Pictures and videos documenting the different species of midges
  • Captured chironomids for machine learning or artificial intelligence training
Challenge Owners

Catchment & Waterways Department and Water Quality Department

Expected Timeline for Deployment

Total project period - less than 18 months
Development of prototype - 6 - 12 months
Testing and review of the solution - 6 months

Expected Project Outcomes

A site-tested prototype system that is able to sort, identify and count adult midges with an accuracy of at least 90%.

Challenge Winners

Orinno Technology, Singapore

Newcastle Research & Innovation Institute, Singapore