FAQs
Vectrack – Frequently Asked Questions
Introduction
Most questions from Vectrack users relate to the accuracy of the sensor mosquito classification (by genus and sex) compared to the manual classification of the catch bag contents.
A machine learning (ML) model is used to automatically classify the insects recorded by the sensor. The model was developed (trained) using thousands of mosquito flights in the laboratory. The mosquitoes were reared in the laboratory under breeding conditions designed to represent the range of breeding conditions in the field. Their flights were then recorded by a Vectrack sensor in the laboratory under ambient conditions designed to represent the range of ambient conditions in the field.
The current ML model has been trained using two urban mosquito species:
- Aedes albopictus
- Culex pipiens complex
The model covers an ambient temperature range of 18°C to 33°C.
However, we are working to:
- Increase the number of species covered (starting with Aedes aegypti).
- Extend the ambient temperature range down to 15°C to better represent the ambient conditions in which mosquitoes are active in cooler climates.
info
To achieve the best accuracy from the sensor, it is very important to follow the advice given in the Vectrack Sensor User Manual.
| Question | Answer |
|---|---|
| In trials comparing automatic sensor classifications with manual analysis of catch bag contents, why must users keep a strict record of the date and time (to the minute) of the start and end of each catch bag collection cycle? | Accurate timestamps are required to associate each sensor classification with the correct catch bag collection and manual analysis period. Missing or incorrect records can introduce significant uncertainty. When multiple sensors are used, users must also record which sensor each catch bag came from. |
| When comparing sensor classifications with manual analysis, why should the sensor remain running for at least 35 minutes after collecting the final catch bag? | The sensor uploads data every 30 minutes. Turning it off immediately after the final collection may result in the loss of up to 30 minutes of data. Waiting at least 35 minutes ensures that the final data packet is transmitted to the server. |
| Why is the classification accuracy less than 100% for Aedes albopictus and Culex pipiens complex in urban environments? | The model achieves about 94%1 accuracy under laboratory conditions but typically around 88%2 in the field due to greater genetic diversity, environmental variation, and the presence of many non-target insects. Occasional confusion between mosquitoes and non-target insects reduces classification accuracy. |
| Why is mosquito classification accuracy poorer in non-urban environments? | The current model was trained only with the urban species Aedes albopictus and Culex pipiens complex. Accuracy is generally lower for rural, forest, and wetland species. Future versions will include additional species such as Culex theileri, Culex perexiguus, Aedes caspius, Aedes vexans, and Anopheles spp. |
| Why is mosquito classification accuracy poor below 18°C? | The model was trained using temperatures between 18°C and 33°C. Below 18°C, mosquitoes are more likely to be misclassified as non-mosquitoes, and genus and sex classification becomes less reliable. Future updates will extend the model to lower temperatures. |
| Why is classification accuracy low when fewer than 10 mosquitoes are captured per day? | The model does not yet adequately represent low-productivity conditions, such as the beginning of the mosquito season or newly colonized areas. Accuracy improves significantly when captures exceed 10 mosquitoes per day. |
| Why is the sensor count significantly higher than the catch bag count? | This sensor overcount is usually caused by predation (e.g., ants or other insects) or degradation of specimens through dehydration. Using shorter collection cycles (24–48 hours) helps minimize this error. |
| Why is the sensor count particularly higher than the catch bag count for female Culex mosquitoes? | Strong female Culex mosquitoes may crawl back out of the catch bag and escape through the sensor. The sensor records the initial capture but cannot detect the escape because no wingbeat signal is generated while crawling. Overcounts of up to 40% may occur. |
| Why is the sensor count lower than the catch bag count? | Sensor undercounts may result from (1) data loss due to poor mobile network coverage, (2) sensor saturation when captures exceed 50 mosquitoes every 30 minutes, or (3) mosquitoes entering the trap without flying, producing no detectable wingbeat signal. |
| For lab use, why does Irideon recommend not using lab reared mosquitoes beyond the 15th generation? | The ML model was trained using mosquitoes up to the 15th laboratory generation to preserve wild-type characteristics. Later generations may lose these traits, reducing classification accuracy. |
| For lab use, why should mosquitoes not be released closer than 20 cm from the sensor entrance? | Mosquitoes released too close to the sensor may be sucked into the trap before they begin flying. Without a wingbeat signal, the sensor cannot properly characterize or classify them. |