Abstract
Earth Observation (EO) data analysis has been significantly revolutionized bydeep learning (DL), with applications typically limited to grid-like datastructures. Graph Neural Networks (GNNs) emerge as an important innovation,propelling DL into the non-Euclidean domain. Naturally, GNNs can effectivelytackle the challenges posed by diverse modalities, multiple sensors, and theheterogeneous nature of EO data. To introduce GNNs in the related domains, ourreview begins by offering fundamental knowledge on GNNs. Then, we summarize thegeneric problems in EO, to which GNNs can offer potential solutions. Followingthis, we explore a broad spectrum of GNNs' applications to scientific problemsin Earth systems, covering areas such as weather and climate analysis, disastermanagement, air quality monitoring, agriculture, land cover classification,hydrological process modeling, and urban modeling. The rationale behindadopting GNNs in these fields is explained, alongside methodologies fororganizing graphs and designing favorable architectures for various tasks.Furthermore, we highlight methodological challenges of implementing GNNs inthese domains and possible solutions that could guide future research. Whileacknowledging that GNNs are not a universal solution, we conclude the paper bycomparing them with other popular architectures like transformers and analyzingtheir potential synergies.