Water quality tool for prawn farming

Managing water quality in prawn ponds using deep learning

In this Digiscape project, we developed a decision support system for managing water quality in prawn ponds. The system uses various sources of data and deep learning models to provide 24-hour forecasting and anomaly detection of important water quality variables, such as dissolved oxygen. It provides prawn farmers with tools to proactively avoid a poor growing environment, thereby optimising growth and reducing the risk of losing stock.

This is a major shift for farmers who are currently only able to correct poor water quality conditions reactively. Farmers could access the system through a browser interface on their computer or smart phone.

The forecasting capability of the system is provided by our very own ForecastNet deep learning model. Anomaly detection is provided by the Transformer deep learning model. Forecasts are presented at 15 minute intervals over 24-hours to help farmers understand the dynamics of ponds and predict their future state. Our forecasts are accurate.

The anomaly detection allows farmers to detect problematic pond conditions or sensors. In our tests, the anomaly detection model was able to detect unusual changes in dissolved oxygen five hours before a dissolved oxygen crash occurred. This is significant as a dissolved oxygen crash can result in the loss of the entire stock in a pond, which is typically 8 to 12 tons of prawns.

Early detection lets farmers take early action to mitigate or avoid such disasters. The anomaly detection model was also able to detect severe biofouling on a neglected sensor. Biofouling reduces the accuracy of a sensor and can damage it in extreme cases.

A water quality forecast

A water quality forecast from our deep learning system


This project is now complete.

Read the publication

Dabrowski J, Rahman A, Hellicar A, Rana M, Arnold S (2022) Deep Learning for Prawn Farming. In: Gama, J., Li, T., Yu, Y., Chen, E., Zheng, Y., Teng, F. (eds) Advances in Knowledge Discovery and Data Mining. PAKDD 2022. Lecture Notes in Computer Science(), vol 13282. Springer, Cham. https://doi.org/10.1007/978-3-031-05981-0_3

Meet the team

  • Ashfaqur sensor data analytics team develop machine learning algorithms for transforming sensor data into decisions. He led Digiscape’s work on smart glass, computer vision and machine learning for efficient prawn farm management.
  • Stuart has been involved in a broad range of research projects related to improving the sustainability and productivity of aquaculture.
  • Joel is part of the CSIRO's sensor data analytics team. His key research areas are machine learning and probabilistic modelling in time series problems.
  • John has a robotics and sensing background which has been applied loosely in the marine space for a decade. His current focus is on bringing IoT and machine learning to the aquaculture and, more broadly, agriculture sectors.