Exploring Generalisability of Self-Distillation with No Labels for SAR-Based Vegetation Prediction

TL;DR

This study investigates how self-supervised models trained on synthetic SAR data generalize across different geographic regions for vegetation prediction, revealing insights into embedding space separability and model transferability.

Contribution

It demonstrates the impact of regional differences on embedding spaces and model generalization in SAR-based vegetation prediction using self-distillation without labels.

Findings

Embedding spaces vary significantly between regions for S1GRD and overlap for GSSIC.

Greater distances in embedding space correlate with higher prediction errors on unseen regions.

Self-supervised models' ability to generalize depends on regional embedding separability.

Abstract

In this work we pre-train a DINO-ViT based model using two Synthetic Aperture Radar datasets (S1GRD or GSSIC) across three regions (China, Conus, Europe). We fine-tune the models on smaller labeled datasets to predict vegetation percentage, and empirically study the connection between the embedding space of the models and their ability to generalize across diverse geographic regions and to unseen data. For S1GRD, embedding spaces of different regions are clearly separated, while GSSIC's overlaps. Positional patterns remain during fine-tuning, and greater distances in embeddings often result in higher errors for unfamiliar regions. With this, our work increases our understanding of generalizability for self-supervised models applied to remote sensing.