Beyond Grid Data: Exploring Graph Neural Networks for Earth Observation

TL;DR

A broad spectrum of GNNs’ applications to scientific problems in Earth systems is explored, covering areas such as weather and climate analysis, disaster management, air quality monitoring, agriculture, land cover classification, hydrological process modeling, and urban modeling.

Abstract

Earth Observation (EO) data analysis has been significantly revolutionized by deep learning (DL), with applications typically limited to grid-like data structures. Graph Neural Networks (GNNs) emerge as an important innovation, propelling DL into the non-Euclidean domain. Naturally, GNNs can effectively tackle the challenges posed by diverse modalities, multiple sensors, and the heterogeneous nature of EO data. To introduce GNNs in the related domains, our review begins by offering fundamental knowledge on GNNs. Then, we summarize the generic problems in EO, to which GNNs can offer potential solutions. Following this, we explore a broad spectrum of GNNs' applications to scientific problems in Earth systems, covering areas such as weather and climate analysis, disaster management, air quality monitoring, agriculture, land cover classification, hydrological process modeling, and urban modeling. The rationale behind adopting GNNs in these fields is explained, alongside methodologies for organizing graphs and designing favorable architectures for various tasks. Furthermore, we highlight methodological challenges of implementing GNNs in these domains and possible solutions that could guide future research. While acknowledging that GNNs are not a universal solution, we conclude the paper by comparing them with other popular architectures like transformers and analyzing their potential synergies.

Figures (14)

• Figure 1: Multiple types of data are collected by EO sensors. For example, in Euclidean space, data can be organized as times-series (e.g., teleconnection indices), grided images, and datacubes (e.g., hyperspectral imagery, spatial-temporal images). The non-euclidean data includes points (e.g., monitoring stations, road sensors, meteorological stations), point clouds (e.g., LiDAR), mesh representations, etc.
• Figure 2: Comparison between a CNN layer and a GNN layer. (a) CNNs update inputs through a shift-invariant weighted sum within a regular window. (b) GNNs update inputs by aggregating the information from their neighbors. For clarity, we use a single number to represent the node feature. The example demonstrates how node features are updated using a simple GCN layer.
• Figure 3: The comparison between a traditional pooling layer and a graph pooling layer. (a) Max pooling with 2 strides on a sample Radar map. The pooling on grided data only changes the value/number of elements. (b) The graph pooling requires the update of graph topology due to the change of node numbers.
• Figure 4: Variants of GNNs: Advanced GNNs, their unique advantages to solve EO Challenges (Chapter \ref{['sec:problems']}), and examples of successful applications in the EO Field (Chapter \ref{['sec:applications']}).
• Figure 5: GNNs are widely adopted in various geoscientific research areas, though their applications are unevenly distributed. This survey selects key publications from the available resources, maintaining a balance similar to the original distribution.