The proliferation of foundation models, pretrained on large-scale unlabeled datasets, has emerged as an effective approach in creating adaptable and reusable architectures that can be leveraged for various downstream tasks using satellite observations. However, their direct application to hyperspectral remote sensing remains challenging due to inherent spectral disparities and the scarcity of available observations. In this work, we present HyperKD, a novel knowledge distillation framework that enables transferring learned representations from a teacher model into a student model for effective development of a foundation model on hyperspectral images. Unlike typical knowledge distillation frameworks, which use a complex teacher to guide a simpler student, HyperKD enables an inverse form of knowledge transfer across different types of spectral data, guided by a simpler teacher model. Building upon a Masked Autoencoder (MAE) with a Vision Transformer (ViT) backbone, HyperKD distills knowledge from Prithvi (a ViT-based MAE geospatial foundation model trained on lower-dimensional multispectral data) into a student tailored for EnMAP hyperspectral imagery. HyperKD addresses the inverse domain adaptation problem with spectral gaps by introducing a feature-based strategy that includes spectral range-based channel alignment, spatial featureguided masking, and an enhanced loss function tailored for hyperspectral images. HyperKD bridges the substantial spectral domain gap, enabling the effective use of pretrained foundation models for geospatial applications. Extensive experiments show that HyperKD significantly improves representation learning in MAEs, leading to enhanced reconstruction fidelity and more robust performance on downstream tasks such as land cover classification, crop type identification, and soil organic carbon prediction, underpinning the potential of knowledge distillation frameworks in remote sensing analytics with hyperspectral imagery.

Masked Autoencoders struggle with hyperspectral satellite imagery containing 200+ spectral bands, as uniform masking across all channels obscures critical spatial-spectral relationships. We introduce TerraMAE, which employs an adaptive channel grouping strategy to organize bands into statistically coherent groups with independent masking. Together with a customized loss function, this data-driven grouping strategy enables TerraMAE to learn robust spatial-spectral representations from unlabeled HSI. Experiments demonstrate that TerraMAE significantly outperforms baseline Masked Autoencoder and supervised ResNet-50 on soil texture prediction, achieving 15.7% and 6.6% lower error, respectively.

Accurate estimation of soil moisture is crucial for efficient agricultural management and environmental monitoring. However, the task of predicting soil moisture levels becomes challenging in regions with limited data availability. In this study, we propose a knowledge distillation-based deep learning approach to enhance soil moisture prediction with machine learning apporach using the low resolution but wide coverage soil moisture Active Passive (SMAP) satellite data.Our framework leverages the knowledge distillation, where a high-capacity teacehr model (VGG13) which is pre-traineed on a large dataset (SMAP) and a lightweight student model (ResNet8) which is then trained on sensor-based highly accurate but extremely sparse station data. The student model benefits from the distilled knowledge of the teacher model, acquiring a deeper understanding of the underlying patterns and relationships in the data.The space-efficient student model significantly reduces the inference time with high prediction accuracy and demonstrates the potential benefit to agricultural management, water resource planning, and ecological studies by providing accurate and reliable soil moisture predictions in data-scarce regions. Our findings reveal how to identify performant settings for achieving the best trade-off between accuracy and model complexity.