Plenary 1

Dimensionality Reduction and Emulation: Unlocking the Value of Hyperspectral Data

Abstract
Earth observation generates vast, high-dimensional datasets essential for addressing critical societal, environmental, and economic challenges—from climate monitoring to urban planning. Yet, this data richness comes with complexity: redundancy, sparsity, and the curse of dimensionality hinder efficient processing and limit the effectiveness of machine learning models. In this talk, I will explore how dimensionality reduction (DR) and emulation techniques are transforming hyperspectral data analysis. DR methods—ranging from classical PCA to modern manifold learning—preserve essential data structures while simplifying complexity, enabling better compression, fusion, anomaly detection, and visualization. Emulators, built via machine learning, provide fast surrogates to complex physical models, facilitating real-time or large-scale inference. Together, these approaches unlock the full potential of hyperspectral data across the remote sensing value chain, offering practical solutions for more efficient, interpretable, and scalable Earth observation.