CloudNine: Analyzing Meteorological Observation Impact on Weather Prediction Using Explainable Graph Neural Networks

TL;DR

CloudNine addresses the challenge of quantifying how individual meteorological observations influence weather forecasts across multiple spatial and temporal scales using explainable graph neural networks. It combines an explainable graph neural network (XGNN) based atmospheric state estimator that fuses observations at time $t$ with NWP grid data from time $t-1$ to predict states at time $t$, with attribution provided by layer-wise relevance propagation (LRP). Key contributions include a NWP/DA-system-agnostic framework, multi-resolution observation-impact analysis for single observations, and an interactive web interface for exploration. Experiments on Korea Meteorological Administration data show consistent predictive performance across regions and meaningful, location-specific attributions, with fidelity highest in Asia, demonstrating practical value for observation prioritization and forecast-system design. This work provides a scalable, explainable tool to inform data assimilation and sensor deployment decisions in meteorology.

Abstract

The impact of meteorological observations on weather forecasting varies with sensor type, location, time, and other environmental factors. Thus, quantitative analysis of observation impacts is crucial for effective and efficient development of weather forecasting systems. However, the existing impact analysis methods are difficult to be widely applied due to their high dependencies on specific forecasting systems. Also, they cannot provide observation impacts at multiple spatio-temporal scales, only global impacts of observation types. To address these issues, we present a novel system called ``CloudNine,'' which allows analysis of individual observations' impacts on specific predictions based on explainable graph neural networks (XGNNs). Combining an XGNN-based atmospheric state estimation model with a numerical weather prediction model, we provide a web application to search for observations in the 3D space of the Earth system and to visualize the impact of individual observations on predictions in specific spatial regions and time periods.

Figures (2)

• Figure 1: An overview of the proposed system, CloudNine.
• Figure 2: Examples of visualizing observation impacts by using the interactive web application. Each point indicates nodes in the meteorological graph. The node color represents the types of nodes, and the node size refers to the node's importance.