A digital twin for real-time biodiversity forecasting with citizen science data

Area of Science:

• Ecology
• Bioacoustics
• Computational Biology

Background:

• Citizen science generates vast biodiversity data but faces challenges in data quality and participant expertise.
• Bridging the gap between raw data collection and actionable research outputs is crucial for effective biodiversity monitoring.

Purpose of the Study:

• To demonstrate how citizen science, combined with a digital twin and machine learning, can enable accurate real-time avian distribution predictions.
• To overcome limitations in species identification skills among citizen scientists and improve data quality.

Main Methods:

• Utilized a smartphone app to collect raw audio data from citizen scientists across Finland.
• Developed a digital twin integrating machine learning for automated bird sound classification, validation, and reclassification.
• Implemented interval recordings and permanent point count networks to mitigate spatiotemporal sampling biases.

Main Results:

• Generated over 15 million bird detections in two years.
• Digital twin-informed models showed higher accuracy in predicting bird spatiotemporal distributions compared to independent test data.
• The machine learning classifiers continuously improved through ongoing data validation and reclassification.

Conclusions:

• This scalable approach enhances inclusivity in citizen science by enabling contributions from individuals without expert identification skills.
• The digital twin methodology accelerates the generation of reliable biodiversity information, particularly for understudied regions.
• The system has significant potential for real-time biomonitoring and advancing ecological research.