GeoGNN: Time Series Geo-Localization using Two-Tower Graph Neural Networks

2026-06-06 · Source: Machine Learning · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics · Depth: Expert, quick

Summary

GeoGNN is a novel two-tower Graph Neural Network designed for time series geolocalization, a task focused on inferring the geographic origin of raw time series data. This system aims to provide crucial spatial context, thereby enabling various location-aware applications. GeoGNN's architecture comprises a spatial tower that learns embeddings for geographic cell candidates by employing a geographic adjacency graph, and a temporal tower that extracts informative representations from the time series itself. During inference, the system matches temporal representations against geographic embeddings using dot-product similarity, augmented by an auxiliary classification head, to predict the time series' associated geographic origin. Evaluated on large-scale, countrywide electricity-consumption datasets, GeoGNN demonstrated superior performance, improving both fine- and coarse-grained geolocalization accuracy by approximately 27% on average.

Key takeaway

For data scientists working with time series data requiring spatial context, GeoGNN offers a robust approach to geolocalization. If you are developing location-aware applications or analyzing countrywide consumption patterns, consider implementing a two-tower Graph Neural Network architecture. This method significantly enhances both fine- and coarse-grained accuracy, as demonstrated by its ~27% average improvement on electricity-consumption datasets, providing more precise geographic origins for your time series.

Key insights

GeoGNN's two-tower GNN infers time series geographic origins, boosting accuracy by integrating spatial and temporal embeddings.

Principles

• Utilizing geographic adjacency graphs improves spatial context.
• Two-tower architectures effectively match distinct data types.
• Combining similarity matching with classification boosts prediction.

Method

GeoGNN trains a spatial tower on geographic cell embeddings and a temporal tower on time series representations. Inference matches these via dot-product similarity and an auxiliary classification head.

In practice

• Apply to electricity consumption data for grid management.
• Integrate spatial context into time series analysis.
• Enhance location-aware applications with precise origins.

Topics

• Graph Neural Networks
• Time Series Geolocalization
• Two-Tower Architecture
• Geographic Embeddings
• Electricity Consumption
• Spatial Context

Best for: Research Scientist, AI Scientist, Data Scientist

Related on AIssential

• Semi-Supervised Learning on Graphs using Graph Neural Networks — GNN performance in semi-supervised learning is bounded by approximation, stochastic, and optimization errors, influenced by label scarcity and graph topology.
• DIY #21 - Step-by-Step Guide to Time Series Forecasting with RNN — Vanilla RNNs effectively forecast cyclical time series data by learning temporal patterns through sequential processing.
• STELLAR: Spatio-Temporal Environmental Learning with Latent Alignment and Refinement for Long-Tailed Species Distribution Modeling — STELLAR jointly optimizes dynamic habitat context and community structure in a shared latent space for improved species distribution modeling.
• PLOT: Progressive Localization via Optimal Transport in Neural Causal Abstraction — PLOT localizes neural causal variables efficiently using optimal transport, reducing computational overhead for mechanistic interpretability.
• Storm Surge Modeling, Bias Correction, Graph Neural Networks, Graph Convolution Networks — StormNet, a GNN, significantly reduces storm surge forecast errors by correcting biases using spatio-temporal dependencies.
• Learning Long Range Spatio-Temporal Representations over Continuous Time Dynamic Graphs with State Space Models — CTDG-SSM unifies state-space modeling with graph topology to capture long-range spatio-temporal dependencies in continuous-time dynamic graphs.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Machine Learning.