Hide and Seek: Investigating Redundancy in Earth Observation Imagery

Tasos Papazafeiropoulos; Nikolaos Ioannis Bountos; Nikolas Papadopoulos; Ioannis Papoutsis

cs.CVMar 13, 2026

Hide and Seek: Investigating Redundancy in Earth Observation Imagery

Authors: Tasos Papazafeiropoulos, Nikolaos Ioannis Bountos, Nikolas Papadopoulos, Ioannis Papoutsis

Organizations: Orion Lab National Observatory of Athens & National Technical University of Athens · Independent Researcher

Abstract

The growing availability of Earth Observation (EO) data and recent advances in Computer Vision have driven rapid progress in machine learning for EO, producing domain-specific models at ever-increasing scales. Yet this progress risks overlooking fundamental properties of EO data that distinguish it from other domains. We argue that EO data exhibit a multidimensional redundancy (spectral, temporal, spatial, and semantic) which has a more pronounced impact on the domain and its applications than what current literature reflects. To validate this hypothesis, we conduct a systematic domain-specific investigation examining the existence, consistency, and practical implications of this phenomenon across key dimensions of EO variability. Our findings confirm that redundancy in EO data is both substantial and pervasive: exploiting it yields comparable performance ( $\approx98.5\%$ of baseline) at a fraction of the computational cost ( $\approx4\times$ fewer GFLOPs), at both training and inference. Crucially, these gains are consistent across tasks, geospatial locations, sensors, ground sampling distances, and architectural designs; suggesting that multi-faceted redundancy is a structural property of EO data rather than an artifact of specific experimental choices. These results lay the groundwork for more efficient, scalable, and accessible large-scale EO models.

Figures & tables

Figure 1: Overview of RViT and RViT-UperNet for classification and semantic segmentation.

Dataset	Input Modality	Sensor	DL Task	# Classes	EO Task	Spatial Res.	Image Size	Coverage
BigEarthNet	MS/SAR	S1, S2	Multi-label classification	19	LULC classification	10m	$120\times 120$	Europe
BigEarthNet-5	MS/SAR	S1, S2	Multi-label classification	5	LULC classification	10m	$120\times 120$	Europe
MLRSNet	RGB	Multi-sensor	Multi-label classification	60	Semantic Scene Understanding	$\approx$ 10 - 0.1m	$256\times 256$	Global
Woody	RGB	Aerial	Image Segmentation	4	Tree-species detection	50cm	$224\times 224$	Chile
Waititu	RGB	Aerial	Image Segmentation	3	Invasion tree-species detection	50cm	$224\times 224$	New Zealand
Flair	RGB/NIR/DEM	Aerial	Image Segmentation	19	LULC semantic segmentation	20cm	$512\times 512$	France

Table 1: Taxonomy of datasets examined in this study. S1/S2 stand for Sentinel-1/2, MS for Multispectral, SAR for Synthetic Aperture Radar) and NIR for Near Infrared. DEM stands for Digital Elevation Model and LULC for Land Use Land Cover.

Figure 2: Examination of redundancy reduction during training across datasets. The x-axis indicates the retention ratios. For Thresholded-diversity based masking strategy we show the corresponding similarity thresholds resulting, on average, in similar retention ratios.

Figure 3: Examination of redundancy reduction directly on inference, for models trained with masking and without. We group models trained with the same retention ratio on the x-axis. We distinguish the inference retention ratio with color.

Figure 4: Examination of GSD impact on the robustness of redundancy reduction for MLRSNet and Flair.

Figure 7: Impact of redundancy reduction on inference for BigEarthNet-5.

Figure 8: Performance-Efficiency trade-off in training.

Figure 10: Impact of Transformer’s patch size on redundancy phenomenon.

Figure 11: Visual examples of the masking strategies at varying retention ratios on the MLRSNet dataset

	Retention ratio%
Dataset	75	50	25	15	5
BigEarthNet	0.95	0.85	0.75	0.65	0.50
MLRSNet	0.98	0.92	0.75	0.65	0.50
FLAIR	0.98	0.95	0.87	0.80	0.67
Woody	0.99	0.97	0.87	0.80	0.60
Waititu	-	0.99	0.94	0.90	0.80

Table 2: Correspondence of similarity thresholds to retention ratios per dataset

Remote Sensing

Remote sensing uses satellite, aerial, and ground-based observations to measure and map Earth and other planetary surfaces.

See all 528 papers

A Machine Learning API for Earth Observation Data Cubes Based on openEO
Sep 11, 2026Brian Pondi, Jonas Hurst, Rolf Simoes +3ML Reproducibility Remote Sensing
Institute for Geoinformatics, University of Münster, Heisenbergstr. 2, Münster, 48149, North-Rhine Westphalia, Germany. · FGV Agro - Center for Agribusiness Studies, Fundação Getulio Vargas, Av. Paulista, 542, São Paulo, 01310-000, São Paulo, Brazil. · Bochum University of Applied Sciences, Am Hochschulcampus 1, Bochum, 44801, North-Rhine Westphalia, Germany.
From Machine Learning to Large-Scale EO Products: Best Practices for Making Maps
Jul 27, 2026Ghjulia Sialelli, Robin Young, Yuchang Jiang +9Remote Sensing
Photogrammetry and Remote Sensing, ETH Zurich, Switzerland · ETH AI Center, Zurich, Switzerland · Dept. of Computer Science and Technology, University of Cambridge, UK +6
Reuse or Relearn? A Spectral View of Earth Observation Foundation Models
Sep 26, 2026Mehmet Ozgur Turkoglu, Valerio Marsocci, Dominik J. Mühlematter +3Fine-Tuning Geospatial Foundation Models
Agroscope, EO of Agroecosystems · ESA, Φ-lab · ETH Zurich

Sep 11, 2026J/K move · Enter open · S save

A Machine Learning API for Earth Observation Data Cubes Based on openEO

Brian Pondi, Jonas Hurst, Rolf Simoes, Jonas Starke+2

Institute for Geoinformatics, University of Münster, Heisenbergstr. 2, Münster, 48149, North-Rhine Westphalia, Germany. · FGV Agro - Center for Agribusiness Studies, Fundação Getulio Vargas, Av. Paulista, 542, São Paulo, 01310-000, São Paulo, Brazil. · Bochum University of Applied Sciences, Am Hochschulcampus 1, Bochum, 44801, North-Rhine Westphalia, Germany.

ML Reproducibility Remote Sensing

Open paper AskPDF arXiv

Earth Observation (EO) data are increasingly organized as spatio-temporal data cubes, while machine learning (ML) methods operate on tabular feature matrices or structured tensor inputs. This mismatch forces platform-specific transformations that are difficult to reproduce or transfer across cloud infrastructures. The openEO specification provides a unified interface for EO data access and processing across heterogeneous backends, but lacks a standardized approach for ML integration. We propose a process-level ML specification for openEO structured into three stages: model initialization, model actions (training, tuning, inference, validation), and model management. It supports classical algorithms such as Random Forest and SVM, as well as deep learning architectures for time-series and spatial patch-based modeling, including TempCNN, Temporal Attention Encoders, and foundation model inference. Three prototype implementations in R and Python demonstrate feasibility across diverse technology stacks. A crop type mapping use case demonstrates cross-backend interoperability by submitting an identical process graph to independent R and Python backends and comparing predictions and evaluation metrics. Two further use cases demonstrate deep learning on time series and foundation model inference, each executed on a dedicated backend. The prototypes reveal, however, that full cross-backend portability requires deeper harmonization of serialization formats and execution semantics than the process level alone can enforce; backend library versions and preprocessing conventions outside the specification's boundary also affect reproducibility. Addressing both through explicit backend conformance profiles represents the most important near-term direction. The specification advances the reproducibility, portability, and accessibility of ML workflows on EO data cubes across cloud platforms.

Abstract

Remote Sensing

Explore similar work

A Machine Learning API for Earth Observation Data Cubes Based on openEO

From Machine Learning to Large-Scale EO Products: Best Practices for Making Maps

Reuse or Relearn? A Spectral View of Earth Observation Foundation Models