Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

FAConvLSTM: Factorized-Attention ConvLSTM for Efficient Feature Extraction in Multivariate Climate Data

Francis Ndikum Nji, Jianwu Wang

TL;DR

FAConvLSTM tackles the computational and representational limitations of ConvLSTM2D for multivariate climate data by factorizing gating, employing shared depthwise multiscale spatial mixing with SE recalibration, and introducing sparse axial attention and a subspace temporal embedding. The approach yields more stable and interpretable latent representations while substantially reducing FLOPs. Empirical results on ERA5 demonstrate superior clustering and reconstruction performance compared to CNN-based and standard ConvLSTM baselines, with improved capacity to capture teleconnection-scale patterns. This architecture offers a scalable, physically interpretable tool for spatiotemporal climate analysis and forecasting.

Abstract

Learning physically meaningful spatiotemporal representations from high-resolution multivariate Earth observation data is challenging due to strong local dynamics, long-range teleconnections, multi-scale interactions, and nonstationarity. While ConvLSTM2D is a commonly used baseline, its dense convolutional gating incurs high computational cost and its strictly local receptive fields limit the modeling of long-range spatial structure and disentangled climate dynamics. To address these limitations, we propose FAConvLSTM, a Factorized-Attention ConvLSTM layer designed as a drop-in replacement for ConvLSTM2D that simultaneously improves efficiency, spatial expressiveness, and physical interpretability. FAConvLSTM factorizes recurrent gate computations using lightweight [1 times 1] bottlenecks and shared depthwise spatial mixing, substantially reducing channel complexity while preserving recurrent dynamics. Multi-scale dilated depthwise branches and squeeze-and-excitation recalibration enable efficient modeling of interacting physical processes across spatial scales, while peephole connections enhance temporal precision. To capture teleconnection-scale dependencies without incurring global attention cost, FAConvLSTM incorporates a lightweight axial spatial attention mechanism applied sparsely in time. A dedicated subspace head further produces compact per timestep embeddings refined through temporal self-attention with fixed seasonal positional encoding. Experiments on multivariate spatiotemporal climate data shows superiority demonstrating that FAConvLSTM yields more stable, interpretable, and robust latent representations than standard ConvLSTM, while significantly reducing computational overhead.

FAConvLSTM: Factorized-Attention ConvLSTM for Efficient Feature Extraction in Multivariate Climate Data

TL;DR

FAConvLSTM tackles the computational and representational limitations of ConvLSTM2D for multivariate climate data by factorizing gating, employing shared depthwise multiscale spatial mixing with SE recalibration, and introducing sparse axial attention and a subspace temporal embedding. The approach yields more stable and interpretable latent representations while substantially reducing FLOPs. Empirical results on ERA5 demonstrate superior clustering and reconstruction performance compared to CNN-based and standard ConvLSTM baselines, with improved capacity to capture teleconnection-scale patterns. This architecture offers a scalable, physically interpretable tool for spatiotemporal climate analysis and forecasting.

Abstract

Learning physically meaningful spatiotemporal representations from high-resolution multivariate Earth observation data is challenging due to strong local dynamics, long-range teleconnections, multi-scale interactions, and nonstationarity. While ConvLSTM2D is a commonly used baseline, its dense convolutional gating incurs high computational cost and its strictly local receptive fields limit the modeling of long-range spatial structure and disentangled climate dynamics. To address these limitations, we propose FAConvLSTM, a Factorized-Attention ConvLSTM layer designed as a drop-in replacement for ConvLSTM2D that simultaneously improves efficiency, spatial expressiveness, and physical interpretability. FAConvLSTM factorizes recurrent gate computations using lightweight [1 times 1] bottlenecks and shared depthwise spatial mixing, substantially reducing channel complexity while preserving recurrent dynamics. Multi-scale dilated depthwise branches and squeeze-and-excitation recalibration enable efficient modeling of interacting physical processes across spatial scales, while peephole connections enhance temporal precision. To capture teleconnection-scale dependencies without incurring global attention cost, FAConvLSTM incorporates a lightweight axial spatial attention mechanism applied sparsely in time. A dedicated subspace head further produces compact per timestep embeddings refined through temporal self-attention with fixed seasonal positional encoding. Experiments on multivariate spatiotemporal climate data shows superiority demonstrating that FAConvLSTM yields more stable, interpretable, and robust latent representations than standard ConvLSTM, while significantly reducing computational overhead.
Paper Structure (6 sections, 1 figure, 1 table)

This paper contains 6 sections, 1 figure, 1 table.

Table of Contents

  1. Introduction
  2. Related Work
  3. Problem Formulation
  4. Methodology
  5. Results and Discussion
  6. Conclusion

Figures (1)

  • Figure 1: Data flow and inner workings of the FAConvLSTM layer at timestep $t$. The layer uses bottlenecked projections, multiscale depthwise spatial mixing with SE and LN, gated ConvLSTM state updates, periodic axial attention refinement of $\mathbf{h}_t$ every $K$ steps, and a subspace head with temporal self-attention to produce compact spatiotemporal representations.