Machine Learning · arXiv:2611.11398 v1 · submitted 8 Nov 2026 · 76 references
Decomposing Earth Embeddings with Sparse Autoencoders
The paper says
“Earth observation foundation models produce global metre-resolution embeddings that underpin work across agriculture, ecology, hydrology, and beyond. Yet these embeddings are opaque: each dimension entangles unrelated physical concepts, which makes them difficult to interpret and their robustness hard to assess.”
more of the abstract
Here, we propose sparse autoencoders (SAEs) as a principled way to decompose Earth embeddings into sparse, approximately monosemantic features over an overcomplete dictionary. Because LLM judges of interpretations do not transfer to satellite pixels, and spatial autocorrelation matters, we introduce a tailored evaluation protocol. It scores any SAE on reconstruction at matched sparsity, alignment with reference concepts, and spatial scale, each against baselines that include an SAE whose decoder is a frozen random dictionary. On this protocol, SAEs reconstruct AlphaEarth and Tessera embeddings better than every baseline at matched sparsity. Their features align with far fewer reference concepts than raw embedding dimensions (on AlphaEarth, 88% align with at most one of 20 concept groups, most with none, against a median of 12 per raw dimension). They also carry more of their variance at fine spatial scales (23 to 32% within a 640 m patch for BatchTopK and Matryoshka, against 9%). An SAE with a frozen random decoder shares both properties, which therefore follow from sparsity, whereas only a learned dictionary reconstructs the embedding accurately: on AlphaEarth, trained SAEs leave eight times less variance unexplained. We test the sparse code on two applications, discovering land-cover sub-concepts and detecting out-of-distribution pixels by area-of-applicability estimation. Both indicate how sparse features could be used, but neither is yet reliable enough for practical use. Together, these results are a step toward Earth observation models whose features can be inspected.
from the abstract
Pith says
Machine review · deepseek-v4-flash · 10 Nov 2026 · what was read was not recorded · not checked by a person
It can be wrong. Challenge this review or reply to it.
Holds up. A careful, honest empirical study of SAEs on Earth embeddings; the main reconstruction claim holds, and the paper's main contribution is the evaluation protocol and the finding that sparsity alone explains interpretability gains. This line is wrong
4 minor
How often objections from reviews like this one are right: not yet measured
Not checked: figures, code, what the 76 references say.
4 minor objections
The paper says, at Abstract
Pith says: minor objection 1 of 4
The phrase 'LLM judges of interpretations' is slightly informal; consider rephrasing to 'he standard practice of using LLMs to judge interpretations' as in the introduction.
The paper says, at §2.3, concept alignment
Pith says: minor objection 2 of 4
The threshold of folded AUROC >= 0.65 is stated but not justified. A brief note on why this value was chosen, or a reference to a sensitivity analysis (e.g., Appendix D.4), would be helpful.
show 2 more
The paper says, at Figure 2 caption
Pith says: minor objection 3 of 4
The markers for different dictionary sizes at k=16 are offset horizontally for legibility. Consider clarifying this offset explicitly in the caption so that readers do not misinterpret the x-coordinates as different L0 values.
The paper says, at §3.3
Pith says: minor objection 4 of 4
The term 'frozen-decoder baseline' is sometimes abbrevaited as 'frozen decoder' (e.g., Figure 3). Consistently use the full term for clarity.
What would settle it
Reproduce Figure 2 using a different random subsample of test patches (e.g., 10,000 patches instead of 4,000) and confirm that the ranking of methods and the magnitude of the gap between trained SAEs and the frozen-decoder baseline remain unchanged.
from the skeptic's reading
How the review went
4 readings by deepseek-v4-flash · 10 Nov 2026
- Reader holds up
- Skeptic agrees
- Editor wrote the verdict above
- Referee minor revision: 4 minor
Not run on this paper: the simulated authors' defence.
Pith did not check this review's objections against the paper.
The review in full
The review in brief, machine-written
This is a solid, transparent empirical paper. The central claim—that SAEs reconstruct AlphaEarth and Tessera embeddings better than every baseline at matched sparsity—is well supported by three SAE variants, multiple sparsity levels, dictionary sizes, and seeds, with a frozen-decoder sanity check that separates sparsity from learned dictionary. The reconstruction advantage is real and clear.
What’s genuinely new is the evaluation protocol tailored to geospatial data (reconstruction at matched sparsity, concept alignment against a curated reference registry, spatial scale via variograms) and the systematic comparison with a frozen random decoder. The key finding—that sparsity alone, not dictionary learning, drives the interpretability benefits (alignment with few concepts, fine spatial scale)—is important and reframes expectations. The authors report this honestly, including where the frozen decoder matches or exceeds trained SAEs.
The applications (sub-concept discovery, OOD detection) are preliminary and the authors say so. The sub-concept analysis correctly notes that random spatial partitions match the biome agreement; the OOD detection help is target-dependent and may come from an untrained random code. These are not weaknesses—they are honest limitations that the paper states clearly.
Soft spots: the concept alignment relies on imperfect, temporally mismatched reference maps (2021 vs 2024), which the authors acknowledge. The spatial scale analysis uses a nested sample that leaves an unsampled band between 640 m and 13 km, acknowledged. Neither flaw undermines the main reconstruction claim or the protocol’s utility.
Overall, the paper is a thorough, repeatable piece of work: code is released, experimental design is rigorous, and the authors treat their own limitations with unusual candor. It deserves a serious referee. A reader interested in geospatial interpretability or SAE evaluation will get clear value from it.
The strongest objection
The central claim—that SAEs reconstruct AlphaEarth and Tessera embeddings better than every baseline at matched sparsity—is robustly supported by the experimental design. The comparison with a frozen-decoder baseline controls for the overcomplete dictionary size and shows a clear gap in unexplained variance (2.7% vs 22% on AlphaEarth at L0=16). The result is consistent across three SAE variants, two embeddings, multiple sparsity levels and dictionary sizes, and three seeds with negligible variance. The paper also acknowledges the main confound (decoder expressivity) and addresses it with the frozen-decoder sanity check. No hidden assumption or internal inconsistency undermines this result. The weakest assumptions flagged by the reader (imperfect reference products for concept alignment) do not affect the reconstruction claim. Therefore, no load-bearing concern is identified for the paper's strongest claim.
The reconstruction claim is well-supported by multiple controls: the frozen-decoder baseline, three seeds, three SAE variants, two embeddings, and multiple dictionary sizes. The reader's weakest assumption pertained to concept alignment, which is not part of the central reconstruction claim. I therefore see no reason to adjust the reader's ACCEPT verdict. The proposed concrete test would further reassure that the result is not an artifact of the specific evaluation sample.
The referee's report
If the results hold generally (the paper demonstrates them on five embeddings), this work represents a significant step toward interpretable Earth observation. The evaluation protocol is thoughtfully designed, including a frozen-decoder sanity check borrowed from mechanistic interpretability, and the experimental design is rigorous (multiple seeds, dictionary sizes, sparsity levels). The paper honestly reports limitations and negative results (e.g., frozen decoder shares many properties, OOD detection is target-dependent). The release of code and dataset recipe supports reproducibility. These contributions could enable more transparent and trustworthy global mapping in downstream applications across ecology, agriculture, and hydrology.
Pith's reading of the paper (a machine's summary, not the authors')
This paper applies sparse autoencoders (SAEs) to decompose opaque Earth observation embeddings into sparse, approximately monosemantic features. The authors introduce an evaluation protocol measuring reconstruction at matched sparsity, concept alignment, and spatial scale, comparing trained SAEs with a frozen-random-decoder baseline. They find that all trained SAEs reconstruct embeddings better than every baseline (PCA, ICA, NMF) at the same sparsity. However, on concept alignment and spatial scale, the frozen-decoder baseline—which never updates its decoder—performs nearly as well as trained SAEs. This shows that interpretability gains (features aligned to few concepts, finer spatial scales) come from sparsity itself, not from learning a meaningful dictionary. Accurate reconstruction alone requires the learned dictionary. The paper tests the sparse code on two applications (sub-concept discovery and out-of-distribution detection) but notes both are preliminary. Together, the results are a step toward more interpretable Earth models while clarifying what sparsity alone provides.
What carries the argument. The sparse autoencoder (SAE), specifically the BatchTopK and Matryoshka variants, with an overcomplete dictionary (M > D) and a sparsity constraint (e.g., keeping the Top-K activations per pixel). The frozen-decoder baseline—where decoder columns are random unit vectors that are never updated—isolates the effect of sparsity from the effect of learned representations.
Core claim. Sparse autoencoders can decompose pixelwise Earth embeddings (AlphaEarth, Tessera) into sparse features that align to significantly fewer geographic concept groups and vary on finer spatial scales than raw dimensions or standard decompositions. Critically, a SAE whose decoder is a frozen random dictionary—so that only the encoder is trained—achieves the same concept-alignment and spatial-scale properties as a fully trained model. This indicates that these properties follow from the sparsity constraint, not from learning a meaningful dictionary. However, the fully trained SAE reconstructs the embedding substantially better (eight times less unexplained variance on AlphaEarth), so the learned dictionary is essential for accurate reconstruction. The paper thereby establishes that sparsity alone creates monosemanticity-like features, while the learned dictionary is necessary for preserving the original information.
The premise it rests on. The concept-alignment evaluation assumes that the reference maps (ESA WorldCover, ESRI Land Cover, etc.) and their grouping into 20 concept groups provide accurate ground truth for what concepts the embedding should encode, but these maps have errors and are from different years (e.g., 2021 vs. 2024), so the AUROC-based alignment scores may be noisy.
If the paper is right (Pith's inferences, not the authors')
SAEs make Earth embeddings inspectable, enabling scientists and practitioners to discover what concepts a foundation model has encoded, improving trust and scientific discovery.
The result that sparsity alone confers concept alignment and fine spatial scales means that simpler sparsity-based methods (e.g., random projections with sparsity) could be used for interpretability checks without training a full decoder.
The reconstruction gap shows that learned dictionaries are critical for preserving embedding quality for downstream tasks, so both sparsity and learning are needed for a useful representation.
The evaluation protocol (reconstruction at matched sparsity, concept alignment, spatial scale, with frozen-decoder baseline) can serve as a standard benchmark for future work on Earth embedding interpretability.
The two applications (sub-concept discovery and OOD detection) are not yet reliable but suggest pathways toward label-free land-cover subtyping and more robust transfer mapping.
The finding that a frozen decoder matches a trained SAE on interpretability metrics questions current evaluation practices in mechanistic interpretability for language models: if random dictionaries already yield monosemanticity, then existing metrics may overstate the quality of learned features.
The paper's protocol can be adapted to other geospatial modalities (e.g., location encoders, time-series embeddings) or to foundation models in other domains, by including a random-dictionary baseline to separate sparsity effects from learning.
The target-dependent OOD benefit (canopy height gains, elevation does not) implies that SAE-based distance improvement should be tested on multiple targets before generalization is claimed, and that sparsity helps only when the target's variation aligns with sparse features.
The qualitative sub-concept discovery (tropical vs. boreal forest features) suggests that SAEs can reveal ecologically meaningful regional structure without labels, but formal validation against null models of spatial autocorrelation is needed to confirm that the patterns are not just geographic splits.
Check this review
Pith stored no record of this review's passes and took no fingerprint of it. Reviews made by code from 13 Nov 2026 on carry both.
How often reviews like this one are right
Reviews like this one: written by deepseek-v4-flash, where what was read was not recorded.
How often objections from reviews like this one are right: not yet measured
Reviews by people
This review was written by a machine. No person has reviewed this paper on Pith yet. A scientist listed on Pith can review it for a fee, and their review is published here whatever it concludes.
The paper
Abstract
Earth observation foundation models produce global metre-resolution embeddings that underpin work across agriculture, ecology, hydrology, and beyond. Yet these embeddings are opaque: each dimension entangles unrelated physical concepts, which makes them difficult to interpret and their robustness hard to assess. Here, we propose sparse autoencoders (SAEs) as a principled way to decompose Earth embeddings into sparse, approximately monosemantic features over an overcomplete dictionary. Because LLM judges of interpretations do not transfer to satellite pixels, and spatial autocorrelation matters, we introduce a tailored evaluation protocol. It scores any SAE on reconstruction at matched sparsity, alignment with reference concepts, and spatial scale, each against baselines that include an SAE whose decoder is a frozen random dictionary. On this protocol, SAEs reconstruct AlphaEarth and Tessera embeddings better than every baseline at matched sparsity. Their features align with far fewer reference concepts than raw embedding dimensions (on AlphaEarth, 88% align with at most one of 20 concept groups, most with none, against a median of 12 per raw dimension). They also carry more of their variance at fine spatial scales (23 to 32% within a 640 m patch for BatchTopK and Matryoshka, against 9%). An SAE with a frozen random decoder shares both properties, which therefore follow from sparsity, whereas only a learned dictionary reconstructs the embedding accurately: on AlphaEarth, trained SAEs leave eight times less variance unexplained. We test the sparse code on two applications, discovering land-cover sub-concepts and detecting out-of-distribution pixels by area-of-applicability estimation. Both indicate how sparse features could be used, but neither is yet reliable enough for practical use. Together, these results are a step toward Earth observation models whose features can be inspected.
The paper's own sentences, arranged, kept beside the review and separate from it. Each label is drawn from a closed vocabulary; the number beside it is the probability the classifier gave that label. No sentence in this record was written by a model.
abstract compiled · 2026-11-10
The authors' main claim, as the classifier picked it out 0.96
“Together, these results are a step toward Earth observation models whose features can be inspected.”
How the main claim is stated: asserted 0.67
What the abstract itself states, as probabilities: claims first 0.07 · states limitation 0.97 · human participants 0.03 · controlled comparison 0.98 · randomized assignment 0.02 · states quantitative result 0.99 · compares with prior results 0.97 · mentions code or data availability 0.03
The abstract, sentence by sentence
This table labels the abstract only (arXiv listing abstract v1). The number is the probability the classifier gave the label.
| Sentence | Role | The authors' own work | States the main conclusion | Strength |
|---|---|---|---|---|
| Earth observation foundation models produce global metre-resolution embeddings that underpin work across agriculture, ecology, hydrology, and beyond. | background 1.00 | 0.05 | 0.02 | asserted 1.00 |
| Yet these embeddings are opaque: each dimension entangles unrelated physical concepts, which makes them difficult to interpret and their robustness hard to assess. | gap 0.63 | 0.10 | 0.02 | asserted 1.00 |
| Here, we propose sparse autoencoders (SAEs) as a principled way to decompose Earth embeddings into sparse, approximately monosemantic features over an overcomplete dictionary. | contribution 0.74 | 0.40 | 0.08 | asserted 0.75 |
| Because LLM judges of interpretations do not transfer to satellite pixels, and spatial autocorrelation matters, we introduce a tailored evaluation protocol. | method 0.58 | 0.72 | 0.03 | asserted 0.99 |
| It scores any SAE on reconstruction at matched sparsity, alignment with reference concepts, and spatial scale, each against baselines that include an SAE whose decoder is a frozen random dictionary. | method 0.96 | 0.89 | 0.05 | asserted 0.97 |
| On this protocol, SAEs reconstruct AlphaEarth and Tessera embeddings better than every baseline at matched sparsity. | result 1.00 | 0.96 | 0.34 | asserted 1.00 |
| Their features align with far fewer reference concepts than raw embedding dimensions (on AlphaEarth, 88% align with at most one of 20 concept groups, most with none, against a median of 12 per raw dimension). | result 1.00 | 0.95 | 0.13 | asserted 1.00 |
| They also carry more of their variance at fine spatial scales (23 to 32% within a 640 m patch for BatchTopK and Matryoshka, against 9%). | result 1.00 | 0.94 | 0.06 | asserted 1.00 |
| An SAE with a frozen random decoder shares both properties, which therefore follow from sparsity, whereas only a learned dictionary reconstructs the embedding accurately: on AlphaEarth, trained SAEs leave eight times less variance unexplained. | interpretation 0.57 | 0.94 | 0.16 | asserted 1.00 |
| We test the sparse code on two applications, discovering land-cover sub-concepts and detecting out-of-distribution pixels by area-of-applicability estimation. | contribution 0.81 | 0.97 | 0.05 | asserted 1.00 |
| Both indicate how sparse features could be used, but neither is yet reliable enough for practical use. | limitation 1.00 | 0.90 | 0.11 | asserted 0.82 |
| Together, these results are a step toward Earth observation models whose features can be inspected. | interpretation 0.96 | 0.45 | 0.64 | asserted 0.67 |
What kind of paper the classifier reports this is: method or tool 0.96 · evidence from simulation or computation 0.52
What was examined, and what was not
Sources read
- arXiv listing abstract v1
Not examined
- full text
- figures and tables
- references and citing sentences
- relations to other papers
Levels run
- P abstract
Not run
- S sections
- T body sentences
- C citations
- R cross-paper relations
How the labels were made
Compiled by classifier-1.13.0 via TypeSafe API, question bank abstract-p1-20260923, 2026-11-10: 1 request, 9,008 input tokens.
Every quoted sentence on this record is the authors'. Every label is from Pith's closed vocabulary, with the probability shown beside it. This record as JSON
Figures (6)
Figures are the authors', reproduced from the paper.
References (76)
- [1] K. Klemmer, E. Rolf, M. Russwurm, G. Camps-Valls, M. Czerkawski, S. Ermon, A. Francis, N. Jacobs, H. R. Kerner, L. Mackey, G. Mai, O. M. Aodha, M. Reichstein, C. Robinson, D. Rolnick, E. Shelhamer, V . Sitzmann, D. Tuia, and X. Zhu. “Earth Embeddings: Toward Artificial Intelli... on Pith
- [2] C. F. Brown, M. R. Kazmierski, V . J. Pasquarella, W. J. Rucklidge, M. Samsikova, C. Zhang, E. Shelhamer, E. Lahera, O. Wiles, S. Ilyushchenko, N. Gorelick, L. L. Zhang, S. Alj, E. Schechter, S. Askay, O. Guinan, R. Moore, A. Boukouvalas, and P. Kohli.AlphaEarth Foundations: A... on PithPith's reviewarXiv
- [3] Z. Feng, C. Atzberger, S. Jaffer, J. Knezevic, S. Sormunen, R. Young, M. C. Lisaius, M. Immitzer, T. Jackson, J. Ball, D. A. Coomes, A. Madhavapeddy, A. Blake, and S. Keshav. “TESSERA: Temporal Embeddings of Surface Spectra for Earth Representation and Analysis”. In:Proceeding... on Pith
- [4] H. Herzog, F. Bastani, Y . Zhang, G. Tseng, J. Redmon, H. Sablon, R. Park, J. Morrison, A. Buraczynski, K. Farley, J. Hansen, A. Howe, P. A. Johnson, M. Otterlee, T. Schmitt, H. Pitelka, S. Daspit, R. Ratner, C. Wilhelm, S. Wood, M. Jacobi, H. Kerner, E. Shelhamer, A. Farhadi,... on Pith
- [5] I. Zvonkov, G. Tseng, I. Becker-Reshef, and H. Kerner.Cropland Mapping Using Geospatial Embeddings. arXiv preprint arXiv:2511.02923. 2025 on PitharXiv
- [6] J. Fang, M. Wu, Z. Zhang, and W. Luo. “Application of AlphaEarth GeoFoundation Model for High-Accuracy Crop Yield Estimation”. In:Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision Workshops (WACVW). 2026 on Pith
- [7] Y . Ma, Y . Shen, A. Swatantran, and D. B. Lobell. “Harvesting AlphaEarth: Benchmarking the Geospatial Foundation Model for Agricultural Downstream Tasks”. In:International Journal of Applied Earth Observation and Geoinformation149 (2026), p. 105258. 11 on Pith
- [8] A. Tsao and D. B. Lobell. “PlantationBench: A Multiscale, Multimodal Remote Sensing Bench- mark for Plantation Mapping Under Distribution Shift”. In:Proceedings of the IEEE/CVF International Conference on Computer Vision Workshops (ICCVW). 2025 on Pith
- [9] J. Giezendanner, Q. Yang, R. Huang, E. Schmitt, A. Chandra, Y . Zhang, J. Vila, D. Hohl, C. Watson, and S. Wang.Partial Recovery of Meter-Scale Surface Weather. arXiv preprint arXiv:2602.23146. 2026 on PithPith's reviewarXiv
- [10] P. Qu, L. Ye, W. Ouyang, C. Zhang, Y . Chai, S. Xu, Y . Piao, M. Zhang, and H. Lu.Utilizing Earth Foundation Models to Enhance the Simulation Performance of Hydrological Models with AlphaEarth Embeddings. arXiv preprint arXiv:2601.01558. 2026 on PithPith's reviewarXiv
- [11] M. B. Pettersson and A. Daoud.Leveraging Compact Satellite Embeddings and Graph Neural Networks for Large-Scale Poverty Mapping. arXiv preprint arXiv:2511.01408. 2025 on PitharXiv
- [12] Z. Feng, C. Atzberger, S. Jaffer, J. Knezevic, S. Sormunun, R. Young, M. Lisaius, M. Immitzer, T. Jackson, J. Ball, D. Coomes, A. Madhavapeddy, A. Blake, and S. Keshav.Applications of the TESSERA Geospatial Foundation Model to Diverse Environmental Mapping Tasks. SSRN preprint. 2026 on Pith
- [13] L. Houriez, S. Pilarski, B. Vahedi, A. Ahmadalipour, T. H. Scully, N. Aflitto, D. Andre, C. Jaffe, M. Wedner, R. Mazzola, J. Jeffery, B. Messinger, S. McGinley-Smith, and S. Russell. “Scalable Geospatial Data Generation Using AlphaEarth Foundations Model”. In:NeurIPS 2025 Work... on Pith
- [14] A. Rao, M. Rußwurm, K. Klemmer, and E. Rolf. “Measuring the Intrinsic Dimension of Earth Representations”. In:Proceedings of the International Conference on Learning Representations (ICLR). 2026 on Pith
- [15] I. F. Benavides-Martinez, J. Guthrie, J. E. Arias, Y . A. Garces-Gomez, A. I. Guzman-Alvis, C. V . Portilla-Cabrera, S. Mondal, A. J. Allyn, and A. R. Ganguly. “What on Earth Is AlphaEarth? Exploring Functional Specialization and Shared Associations in Geospatial Embeddings fo... on Pith
- [16] M. Rahman. “Physically Interpretable AlphaEarth Foundation Model Embeddings Enable LLM-Based Land Surface Intelligence”. In:Remote Sensing Applications: Society and Envi- ronment42 (2026), p. 102045 on Pith
- [17] H. Meyer and E. Pebesma. “Machine Learning-Based Global Maps of Ecological Variables and the Challenge of Assessing Them”. In:Nature Communications13.1 (2022), p. 2208 on Pith
- [18] I. Corley, C. Robinson, I. Becker-Reshef, and J. M. L. Ferres.From Pixels to Patches: Pooling Strategies for Earth Embeddings. arXiv preprint arXiv:2603.02080. 2026 on PitharXiv
- [19] B. A. Olshausen and D. J. Field. “Emergence of Simple-Cell Receptive Field Properties by Learning a Sparse Code for Natural Images”. In:Nature381.6583 (1996), pp. 607–609 on Pith
- [20] T. Bricken, A. Templeton, J. Batson, B. Chen, A. Jermyn, T. Conerly, N. Turner, C. Anil, C. Denison, A. Askell, R. Lasenby, Y . Wu, S. Kravec, N. Schiefer, T. Maxwell, N. Joseph, Z. Hatfield-Dodds, A. Tamkin, K. Nguyen, B. McLean, J. E. Burke, T. Hume, S. Carter, T. Henighan, ... on Pith
show 56 more
- [21] R. Huben, H. Cunningham, L. R. Smith, A. Ewart, and L. Sharkey. “Sparse Autoencoders Find Highly Interpretable Features in Language Models”. In:Proceedings of the International Conference on Learning Representations (ICLR). 2024 on Pith
- [22] L. Gao, T. D. la Tour, H. Tillman, G. Goh, R. Troll, A. Radford, I. Sutskever, J. Leike, and J. Wu. “Scaling and Evaluating Sparse Autoencoders”. In:Proceedings of the International Conference on Learning Representations (ICLR). 2025 on Pith
- [23] B. Bussmann, P. Leask, and N. Nanda. “BatchTopK Sparse Autoencoders”. In:NeurIPS 2024 Workshop on Scientific Methods for Understanding Deep Learning. 2024 on Pith
- [24] B. Bussmann, N. Nabeshima, A. Karvonen, and N. Nanda. “Learning Multi-Level Features with Matryoshka Sparse Autoencoders”. In:Proceedings of the International Conference on Machine Learning (ICML). 2025 on Pith
- [25] A. Karvonen, C. Rager, J. Lin, C. Tigges, J. I. Bloom, D. Chanin, Y . -T. Lau, E. Farrell, C. S. McDougall, K. Ayonrinde, D. Till, M. Wearden, A. Conmy, S. Marks, and N. Nanda. “SAEBench: A Comprehensive Benchmark for Sparse Autoencoders in Language Model Interpretability”. In... on Pith
- [26] A. Korznikov, A. Galichin, A. Dontsov, O. Rogov, E. Tutubalina, and I. Oseledets.OrtSAE: Orthogonal Sparse Autoencoders Uncover Atomic Features. arXiv preprint arXiv:2509.22033. 2025 on PitharXiv
- [27] O. Gujral, M. Bafna, E. Alm, and B. Berger. “Sparse Autoencoders Uncover Biologically Interpretable Features in Protein Language Model Representations”. In:Proceedings of the National Academy of Sciences122.34 (2025), e2506316122 on Pith
- [28] C. Claye, P. Marschall, W. Ouerdane, C. Hudelot, and J. Duquesne.Discovering Inter- pretable Biological Concepts in Single-Cell RNA-Seq Foundation Models. arXiv preprint arXiv:2510.25807. 2025 on PitharXiv
- [29] A. V . Orlov, Y . V . Makus, G. A. Ashniev, N. N. Orlova, and P. I. Nikitin.What Do Biological Foundation Models Compute? Sparse Autoencoders from Feature Recovery to Mechanistic Interpretability. bioRxiv preprint. 2026 on Pith
- [30] J. Beattie, S. Stevens, N. Rosser, Y . Su, and T. Berger-Wolf.Towards Open-Ended Visual Scientific Discovery with Sparse Autoencoders. arXiv preprint arXiv:2511.17735. 2025 on PithPith's reviewarXiv
- [31] S. Stevens, W.-L. Chao, T. Berger-Wolf, and Y . Su.Interpretable and Testable Vision Features via Sparse Autoencoders. arXiv preprint arXiv:2502.06755. 2025 on PitharXiv
- [32] T. MacMillan and N. T. Ouellette.Towards Mechanistic Understanding in a Data-Driven Weather Model: Internal Activations Reveal Interpretable Physical Features. arXiv preprint arXiv:2512.24440. 2025 on PitharXiv
- [33] F. King, C. Pettersen, D. Posselt, S. Ringerud, and Y . Xie. “Leveraging Sparse Autoencoders to Reveal Interpretable Features in Geophysical Models”. In:Journal of Geophysical Research: Machine Learning and Computation2.4 (2025), e2025JH000769 on Pith
- [34] Y . Hu, R. Glatt, and S. Liu. “Sparse Autoencoders as a Steering Basis for Phase Synchronization in Graph-Based CFD Surrogates”. In:Machine Learning: Science and Technology7.5 (2026), p. 055020 on Pith
- [35] W. R. Tobler. “A Computer Movie Simulating Urban Growth in the Detroit Region”. In: Economic Geography46 (1970), pp. 234–240 on Pith
- [36] E. Rolf, K. Klemmer, C. Robinson, and H. Kerner. “Position: Mission Critical – Satellite Data is a Distinct Modality in Machine Learning”. In:Proceedings of the International Conference on Machine Learning (ICML). 2024 on Pith
- [37] A. Korznikov, A. Galichin, A. Dontsov, O. Rogov, I. Oseledets, and E. Tutubalina.San- ity Checks for Sparse Autoencoders: Do SAEs Beat Random Baselines?arXiv preprint arXiv:2602.14111. 2026 on PithPith's reviewarXiv
- [38] T. Heap, T. Lawson, L. Farnik, and L. Aitchison. “Automated Interpretability Metrics Do Not Distinguish Trained and Random Transformers”. In:Proceedings of the International Conference on Learning Representations (ICLR). 2026 on Pith
- [39] L. Betti, S. Ricke, I. Obadic, A. J. Stewart, and E. Rolf.What’s in an Earth Embedding? An Explainability Analysis of Location Encoders. arXiv preprint arXiv:2606.24997. 2026 on PithPith's reviewarXiv
- [40] R. Young and S. Keshav.Earth Embeddings are Decomposable Records of the Land Surface. arXiv preprint arXiv:2610.32617. 2026 on PithPith's reviewarXiv
- [41] H. Fang, A. J. Stewart, I. Corley, X. X. Zhu, and H. Azizpour. “Earth Embeddings as Products: Taxonomy, Ecosystem, and Standardized Access”. In:Proceedings of the IEEE International Geoscience and Remote Sensing Symposium (IGARSS). 2026 on Pith
- [42] J. Zbontar, L. Jing, I. Misra, Y . LeCun, and S. Deny. “Barlow Twins: Self-Supervised Learning via Redundancy Reduction”. In:Proceedings of the International Conference on Machine Learning (ICML). 2021 on Pith
- [43] A. Coates and A. Y . Ng. “The Importance of Encoding Versus Training with Sparse Coding and Vector Quantization”. In:Proceedings of the International Conference on Machine Learning (ICML). 2011 on Pith
- [44] J. Adebayo, J. Gilmer, M. Muelly, I. Goodfellow, M. Hardt, and B. Kim. “Sanity Checks for Saliency Maps”. In:Advances in Neural Information Processing Systems (NeurIPS). 2018 on Pith
- [45] D. Chanin, J. Wilken-Smith, T. Dulka, H. Bhatnagar, S. Golechha, and J. Bloom. “A is for Absorption: Studying Feature Splitting and Absorption in Sparse Autoencoders”. In:Advances in Neural Information Processing Systems (NeurIPS). 2025. 13 on Pith
- [46] D. N. Karger, O. Conrad, J. Böhner, T. Kawohl, H. Kreft, R. W. Soria-Auza, N. E. Zimmermann, H. P. Linder, and M. Kessler. “Climatologies at High Resolution for the Earth’s Land Surface Areas”. In:Scientific Data4.1 (2017), p. 170122 on Pith
- [47] M. C. Peel, B. L. Finlayson, and T. A. McMahon. “Updated World Map of the Köppen-Geiger Climate Classification”. In:Hydrology and Earth System Sciences11.5 (2007), pp. 1633–1644 on Pith
- [48] E. Dinerstein, D. Olson, A. Joshi, C. Vynne, N. D. Burgess, E. Wikramanayake, N. Hahn, S. Palminteri, P. Hedao, R. Noss, et al. “An Ecoregion-Based Approach to Protecting Half the Terrestrial Realm”. In:BioScience67.6 (2017), pp. 534–545 on Pith
- [49] T. Hengl, M. G. Walsh, J. Sanderman, I. Wheeler, S. P. Harrison, and I. C. Prentice. “Global Mapping of Potential Natural Vegetation: An Assessment of Machine Learning Algorithms for Estimating Land Potential”. In:PeerJ6 (2018), e5457 on Pith
- [50] H. Meyer and E. Pebesma. “Predicting into Unknown Space? Estimating the Area of Appli- cability of Spatial Prediction Models”. In:Methods in Ecology and Evolution12.9 (2021), pp. 1620–1633 on Pith
- [51] A. Oh, W. Shin, and S. Kim. “Sparsity as a Key: Unlocking New Insights from Latent Structures for Out-of-Distribution Detection”. In:Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). 2026 on Pith
- [52] J. Brandt, S. Yi, J. Tolan, X. Li, P. Potapov, J. Ertel, J. Spore, H. V . V o, M. Ramamonjisoa, P. Labatut, P. Bojanowski, and C. Couprie. “CHMv2: Improvements in Global Canopy Height Mapping using DINOv3”. In:Scientific Data(2026) on Pith
- [53] European Space Agency.Copernicus DEM: Global and European Digital Elevation Model. Dataset, European Space Agency. 2022 on Pith
- [54] A. Francis and M. Czerkawski. “Major TOM: Expandable Datasets for Earth Observation”. In: Proceedings of the IEEE International Geoscience and Remote Sensing Symposium (IGARSS). 2024 on Pith
- [55] M. Reichstein, M. Bahn, M. D. Mahecha, J. Kattge, and D. D. Baldocchi. “Linking Plant and Ecosystem Functional Biogeography”. In:Proceedings of the National Academy of Sciences 111.38 (2014), pp. 13697–13702 on Pith
- [56] J. Engels, L. Riggs, and M. Tegmark. “Decomposing the Dark Matter of Sparse Autoencoders”. In:Transactions on Machine Learning Research (TMLR)(2025) on Pith
- [57] K. Kuckreja, M. S. Danish, M. Naseer, A. Das, S. Khan, and F. S. Khan. “GeoChat: Grounded Large Vision-Language Model for Remote Sensing”. In:Proceedings of the IEEE/CVF Con- ference on Computer Vision and Pattern Recognition (CVPR). 2024 on Pith
- [58] W. Zhang, M. Cai, T. Zhang, Y . Zhuang, and X. Mao. “EarthGPT: A Universal Multimodal Large Language Model for Multisensor Image Comprehension in Remote Sensing Domain”. In:IEEE Transactions on Geoscience and Remote Sensing62 (2024) on Pith
- [59] K. Klemmer, E. Rolf, C. Robinson, L. Mackey, and M. Rußwurm. “SatCLIP: Global, General- Purpose Location Embeddings with Satellite Imagery”. In:Proceedings of the AAAI Conference on Artificial Intelligence (AAAI). 2025 on Pith
- [60] Y . Wang, Z. Xiong, C. Liu, A. J. Stewart, T. Dujardin, N. I. Bountos, A. Zavras, F. Gerken, I. Papoutsis, L. Leal-Taixé, and X. X. Zhu. “Towards a Unified Copernicus Foundation Model for Earth Vision”. In:Proceedings of the IEEE/CVF International Conference on Computer Vision... on Pith
- [61] V . Nedungadi, A. Kariryaa, S. Oehmcke, S. Belongie, C. Igel, and N. Lang. “MMEarth: Exploring Multi-Modal Pretext Tasks for Geospatial Representation Learning”. In:Proceedings of the European Conference on Computer Vision (ECCV). 2024 on Pith
- [62] D. Zanaga, R. Van De Kerchove, D. Daems, W. De Keersmaecker, C. Brockmann, G. Kirches, J. Wevers, O. Cartus, M. Santoro, S. Fritz, M. Lesiv, M. Herold, N. -E. Tsendbazar, P. Xu, F. Ramoino, and O. Arino.ESA WorldCover 10 m 2021 v200. Dataset, Zenodo. 2022 on Pith
- [63] Copernicus Climate Change Service.Land Cover Classification Gridded Maps from 1992 to Present Derived from Satellite Observations, Version 2.1.1. Dataset, Copernicus Climate Change Service (C3S) Climate Data Store. 2019 on Pith
- [64] K. Karra, C. Kontgis, Z. Statman-Weil, J. C. Mazzariello, M. Mathis, and S. P. Brumby. “Global Land Use / Land Cover with Sentinel-2 and Deep Learning”. In:Proceedings of the IEEE International Geoscience and Remote Sensing Symposium (IGARSS). 2021. 14 on Pith
- [65] X. Zhang, T. Zhao, H. Xu, W. Liu, J. Wang, X. Chen, and L. Liu. “GLC_FCS30D: The First Global 30 m Land-Cover Dynamics Monitoring Product with a Fine Classification System for the Period from 1985 to 2022 Generated Using Dense-Time-Series Landsat Imagery and the Continuous Cha... on Pith
- [66] R. Stanimirova, K. Tarrio, K. Turlej, K. McAvoy, S. Stonebrook, K.-T. Hu, P. Arévalo, E. L. Bullock, Y . Zhang, C. E. Woodcock, P. Olofsson, Z. Zhu, C. P. Barber, C. M. Souza, S. Chen, J. A. Wang, F. Mensah, M. Calderón-Loor, M. Hadjikakou, B. A. Bryan, J. Graesser, D. L. Beye... on Pith
- [67] D. Sulla-Menashe, J. M. Gray, S. P. Abercrombie, and M. A. Friedl. “Hierarchical Mapping of Annual Global Land Cover 2001 to Present: The MODIS Collection 6 Land Cover Product”. In:Remote Sensing of Environment222 (2019), pp. 183–194 on Pith
- [68] K. Van Tricht, J. Degerickx, S. Gilliams, D. Zanaga, M. Battude, A. Grosu, J. Brombacher, M. Lesiv, J. C. L. Bayas, S. Karanam, S. Fritz, I. Becker-Reshef, B. Franch, B. Mollà-Bononad, H. Boogaard, A. K. Pratihast, B. Koetz, and Z. Szantoi. “WorldCereal: A Dynamic Open-Source ... on Pith
- [69] Y .-F. Ho, C. H. Grohmann, J. Lindsay, H. I. Reuter, L. Parente, M. Witjes, and T. Hengl. “GEDTM30: Global Ensemble Digital Terrain Model at 30 m and Derived Multiscale Terrain Variables”. In:PeerJ13 (2025), e19673 on Pith
- [70] L. Poggio, L. M. de Sousa, N. H. Batjes, G. B. Heuvelink, B. Kempen, E. Ribeiro, and D. Rossiter. “SoilGrids 2.0: Producing Soil Information for the Globe with Quantified Spatial Uncertainty”. In:SOIL7.1 (2021), pp. 217–240 on Pith
- [71] M. Demuzere, J. Kittner, A. Martilli, G. Mills, C. Moede, I. D. Stewart, J. van Vliet, and B. Bechtel. “A Global Map of Local Climate Zones to Support Earth System Modelling and Urban-Scale Environmental Science”. In:Earth System Science Data14.8 (2022), pp. 3835– 3873 on Pith
- [72] Overture Maps Foundation.Overture Maps Foundation Open Map Datasets. General Avail- ability release. 2024 on Pith
- [73] M. S. Isik, L. Parente, D. Consoli, L. Sloat, V . V . Mesquita, L. G. Ferreira, S. Sabbatini, R. Stanimirova, N. M. Teles, N. Robinson, C. Costa Junior, and T. Hengl. “Light Use Efficiency (LUE) Based Bimonthly Gross Primary Productivity (GPP) for Global Grasslands at 30 m Spa... on Pith
- [74] J. Jasiewicz and T. F. Stepinski. “Geomorphons — a Pattern Recognition Approach to Classifi- cation and Mapping of Landforms”. In:Geomorphology182 (2013), pp. 147–156 on Pith
- [75] S. Chen, J. Wang, S. Yuan, J. Li, Y . Xia, Y . Liao, J. Wei, J. Yuan, X. Xu, X. Zhu, P. Zhu, H. Zhang, Y . Zhou, H. Fu, H. Huang, B. Chen, F. Dai, and P. Gong.Democratizing Planetary- Scale Analysis: An Ultra-Lightweight Earth Embedding Database for Accurate and Flexible Globa... on PitharXiv
- [76] M. Pach, S. Karthik, Q. Bouniot, S. Belongie, and Z. Akata. “Sparse Autoencoders Learn Monosemantic Features in Vision-Language Models”. In:Advances in Neural Information Processing Systems (NeurIPS). 2025. 15 Supplementary Material Appendix A gives implementation details, inc... on Pith
Challenges and discussion
Disagree with the review, or know something it missed? Leave a comment, or challenge this review formally: a challenge becomes a permanent entry in this paper's record, and its decision is published here.
Come back by email
A short list of the week's newly reviewed papers. Nothing is sent until you confirm from your inbox, and every email has a link that stops it. No account.
No comments yet.
Sign in to comment