AI in Weather and Tokenomics

Level: MA/MAP
Responsible person: Dr Mark C. Ballandies
Keywords:  AI, Machine learning, weather forecasting, Tokenomics, DePIN

Artificial Intelligence (AI) has driven a paradigm shift across numerous industries, including weather forecasting. Recent research has shown that AI-based solutions can outperform conventional numerical weather prediction systems [1]—which typically require expensive supercomputers to run—thus reducing the high computational costs involved. 

Simultaneously, the rise of decentralized physical infrastructure (DePIN) has enabled the collection of large volumes of sensor data through token incentives. This influx of data holds great potential for enhancing AI-driven forecast models. However, data quality remains a serious concern, as much of the community-provided information is often unreliable and thus of limited use [2]. Compounding this challenge is the lack of robust mechanisms for quantifying the value of high-quality data, which can lead to excessive rewards for data providers and undermine the long-term viability of DePIN systems. 

In response, this work proposes leveraging explainable AI methodologies—such as Integrated Gradients [3]—within state-of-the-art weather models to quantify how specific data inputs influence model outputs. By more clearly attributing contributions from each data source to the accuracy and reliability of forecasts, it may become possible to assign value more effectively to useful data. This approach would mark a significant advance for both current research efforts and practical, token-incentivized data platforms. 

Depending on your expertise and the scope of this thesis, you may focus on replicating and benchmarking existing methods or develop novel research directions.  
A strong background in machine learning and the ability to deploy production-ready code are expected, or the drive and aptitude to rapidly acquire these skills through dedicated effort. 
 

References:

[1]: Zhang, Y., Long, M., Chen, K., Xing, L., Jin, R., Jordan, M.I. and Wang, J., 2023. Skilful nowcasting of extreme precipitation with NowcastNet. Nature, 619(7970), pp.526-532. 
[2]: Chiu, M.T., Mahajan, S., Ballandies, M.C. and Kalabić, U.V., 2024. DePIN: A Framework for Token-Incentivized Participatory Sensing. arXiv preprint arXiv:2405.16495. 
[3]: Lundstrom, D.D., Huang, T. and Razaviyayn, M., 2022, June. A rigorous study of integrated gradients method and extensions to internal neuron attributions. In International Conference on Machine Learning (pp. 14485-14508). PMLR.