Lukas Arzoumanidis

Traditional 2D spatial pattern matching fails to capture the complexities of real-world entities, but this new 3D approach opens up innovative avenues for applications like urban planning and landmark recognition.

Achieving a basin-mean correlation of 0.94 with half to a tenth of the predictors used by traditional models reveals the power of graph neural networks in reconstructing historical terrestrial water storage data.

Expanding beyond traditional two-modality frameworks, MELT and SALT reveal that more modalities don't always translate to better performance in spatial predictions.