New directions in mapping the Earth's surface with citizen science and generative AI

Area of Science:

• Geosciences
• Artificial Intelligence
• Remote Sensing

Background:

• Satellite imagery availability accelerates Earth surface mapping efforts.
• Map accuracy is limited by insufficient in situ data for training machine learning algorithms.
• Citizen science projects like Geo-Wiki and Picture Pile collect in situ data, but volunteer time optimization is needed.

Purpose of the Study:

• To explore the potential of generative AI in land cover/land use mapping.
• To enhance citizen science approaches for in situ data collection.
• To leverage multi-modal large language models (MLLMs) for improved AI spatial awareness.

Main Methods:

• Discussing the integration of generative AI, specifically MLLMs, into citizen science platforms.
• Focusing on improving the spatial awareness capabilities of AI in mapping.
• Enhancing existing citizen science methodologies for data collection and validation.

Main Results:

• Generative AI, particularly MLLMs, offers a novel approach to overcome data limitations in land cover mapping.
• Potential for significant improvements in the accuracy and efficiency of citizen science data collection.
• Enhanced AI spatial awareness can lead to more precise land cover classifications.

Conclusions:

• Generative AI and MLLMs hold significant promise for advancing land cover/land use mapping through citizen science.
• Optimizing volunteer time and improving data quality are key benefits.
• Future research should focus on practical implementation and validation of these AI-enhanced approaches.