Published: Jun 16, 2024 by Daning Huang
The resolution-coverage dilemma is a common challenge in climate research, where the coverage is inversely proportional to the resolution at the same cost. Downscaling is a technique to overcome the dilemma by fusing spatiotemporal data of different resolutions. Downscaling has been done for spatially continuous datasets, such as climate simulation data, but it remains a challenge to fuse unstructured distributed data, such as weather station measurements.
In our recent work, we tackle the downscaling involving a mixture of structured and unstructured data via a graph neural network (GNN) framework, and applied the framework to a street-level temperature estimation problem. Furthermore, we conducted in-depth analysis of feature importance to enhance the model interpretability; sensitivity of features are shown in figure below. Our findings highlight GNN’s high potential in capturing the complex dynamics between urban elements and their impacts on microclimate, thus offering valuable insights for comprehensive urban data collection and urban climate modeling in general.
The work is done jointly with Dr. Ashish Sharma and Dr. Peiyuan Li from UIUC/DPI.
