A Novel Graph-Regulated Disentangling Mamba Model with Sparse Tokens for Enhanced Tree Species Classification from MODIS Time Series
A Novel Graph-Regulated Disentangling Mamba Model with Sparse Tokens for Enhanced Tree Species Classification from MODIS Time Series
Authors: Motasem Alkayid, Zhengsen Xu, Saeid Taleghanidoozdoozan, Yimin Zhu, Megan Greenwood, Quinn Ledingham, Zack Dewis, Mabel Heffring, Naser El-Sheimy, Lincoln Linlin Xu Date: 2026-05-07 Paper ID: openalex:2605.05549
Summary
This paper introduces GDS-Mamba, a novel architecture for tree species classification that improves upon standard Mamba state space models by incorporating graph-based topological modeling, feature disentanglement, and adaptive sparse token selection. By explicitly separating spatial, spectral, and temporal phenology components, the model better handles the high-dimensional coupling inherent in MODIS satellite time series. Experimental results on MOD13Q1 data across Canadian regions demonstrate superior classification accuracy and efficiency compared to twelve state-of-the-art benchmarks.
Key Contributions
- Introduced the Graph-regulated Disentangled Sparse Mamba (GDS-Mamba) architecture to address high-dimensional spatial-spectral-temporal coupling in MODIS data.
- Developed a mini-batch graph-regulated approach to explicitly capture topological correlations across large-scale remote sensing imagery.
- Designed an adaptive sparse token mechanism to mitigate correlation decay in Mamba, achieving 93.94% overall accuracy on MOD13Q1 datasets.
Open Questions & Future Work
Key Concepts
- gds-mamba-model: A hybrid Mamba-based architecture designed to disentangle spatial-spectral-temporal features and model topological context via graph regulation.
Archivist Review
Approved the GDS-Mamba architecture and the open question regarding sparse token optimization in Mamba models. Rejected the dataset candidate as it is a routine remote sensing product, not a specialized benchmark dataset. Applied a strict interpretation of architectural novelty and foundational open problems in state space modeling.
Approved Concepts
- Graph-regulated Disentangled Sparse Mamba (GDS-Mamba): Combines graph-based regularization with state space models to address high-dimensional spatial-spectral-temporal coupling in time series data.
Approved Open Questions
- Optimizing Sparse Mamba Tokens: Sparse tokens are essential for scaling state space models to long sequences while maintaining performance, and resolving the bottleneck of correlation decay is a primary challenge for this architecture’s adoption.
Rejected Candidates
- [dataset] MOD13Q1 (
mod13q1) - not_novel: Routine satellite imagery product that does not qualify as a curated or novel dataset benchmark worthy of a permanent note.
Links
Metadata & Links
- url
- https://arxiv.org/abs/2605.05549
- paper_id
- 2605.05549
- paper_source
- openalex
- domain
- computer-vision
- tags
- time-series-forecastinggraph-neural-networksremote-sensingstate-space-models
- architectures
-
- datasets
-
- concept_slugs
- gds-mamba-model
- dataset_slugs
-
- skill
- TimeSeriesSkill
- created_at
- 2026-05-10T06:04:10Z