How Hyperspectral Satellites Are Revolutionizing Our Cities?

Hyperspectral images capture light across hundreds of bands, like a digital rainbow. This lets scientists tell apart urban materials that look identical to the naked eye, such as asphalt and concrete. Historically, researchers favored airborne platforms because they provide sharper resolution and benchmark datasets. In recent years, satellites have been catching up thanks to more powerful missions and lower costs. The most common techniques include Spectral Unmixing, which separates mixed material signals, and Machine Learning algorithms like Support Vector Machines and Random Forests, perfect for small datasets. Deep neural networks, such as Convolutional Neural Networks, are also entering the field, but they need massive data volumes and heavy computing power. Main applications are land‑cover classification, impervious surface detection, urban vegetation mapping, and Local Climate Zone analysis. Challenges persist: comprehensive training sets are scarce and data‑fusion across different sensors is still underused. With new, more accessible satellites, hyperspectral sensing could soon move from the lab to everyday tools for planning greener, more resilient cities.

Spectral Unmixing relies on linear or non‑linear models to decompose a pixel into pure fractions (endmembers). In urban settings, architectural complexity demands non‑linear approaches, such as kernel‑based unmixing, which handle multiple scattering. Traditional Machine Learning methods, like Support Vector Machines, use RBF kernels to separate classes with optimal margins, but they heavily depend on the choice of C and gamma parameters. Random Forests, on the other hand, combine decision trees to reduce overfitting, offering interpretability through spectral variable importance. Deep Learning, particularly CNNs, treat hyperspectral images as 3‑D volumes, learning spectral‑spatial filters via 3‑D convolutions. However, the scarcity of labeled datasets limits their generalization; data augmentation and transfer learning from networks pre‑trained on multispectral imagery are emerging solutions. Data fusion—e.g., merging hyperspectral with LiDAR or SAR—can enhance discrimination of vertical structures and building height estimation, but it requires precise registration algorithms and multimodal fusion models. Future missions like EnMAP‑2 and the Copernicus Hyperspectral project promise finer spectral resolution (≤ 5 nm) and global coverage, narrowing the gap between airborne and spaceborne data. Open access to these datasets, coupled with cloud‑based processing platforms, should accelerate the shift from experimental studies to operational applications for sustainable urban management.