Projects by Alexander van Lanschot
Tracking Climate Change with Satellites and Artificial Intelligence (2023–2024)
As a senior at Duke University majoring in Electrical and Computer Engineering and Computer Science, Alexander van Lanschot played a pivotal role in the 2023–2024 Bass Connections project titled “Tracking Climate Change with Satellites and Artificial Intelligence.” This interdisciplinary initiative aimed to develop scalable, AI-driven models to monitor climate change impacts globally, particularly in regions lacking frequent ground-based surveys. The team focused on creating a “foundation model” capable of identifying vulnerabilities in infrastructure and exposure to climate hazards such as severe weather, fires, and floods.
Van Lanschot contributed significantly to the development of novel approaches utilizing embedding space and zero-shot learning techniques. These methods allowed the models to achieve high accuracy without extensive training on labeled data, demonstrating their potential as cost-effective solutions for analyzing remote sensing imagery. The team’s work showcased the feasibility of applying self-supervised learning to create adaptable models for climate change mitigation and adaptation planning across diverse regions.
AI Meets Satellite Imagery: A New Strategy for Monitoring Climate Change
In the project “AI Meets Satellite Imagery: A New Strategy for Monitoring Climate Change,” Alexander van Lanschot was instrumental in advancing the use of artificial intelligence to automate the tracking of climate change causes and consequences. The team addressed challenges associated with the scarcity of high-resolution training data and the limited applicability of existing algorithms to diverse geographic regions.
By leveraging embedding space representations and zero-shot learning, van Lanschot and his team enabled models to understand intrinsic relationships between textual and visual data, even in the absence of labeled examples. Their experiments with cutting-edge language-image models, not previously trained on the satellite imagery used, yielded impressive results—achieving an average accuracy rate of nearly 80% across multiple datasets. This work underscores the potential of zero-shot learning as a cost-effective and efficient solution for analyzing remote sensing imagery, paving the way for more informed decision-making in climate change monitoring.