SpecLST: Spectral-Enhanced Long-Short Term Spatio-Temporal Multimodal Modeling for Traffic Forecasting

Kenan Kang; Matthew M. Y. Kuo; Weihua Li

cs.LGOct 26, 2026

SpecLST: Spectral-Enhanced Long-Short Term Spatio-Temporal Multimodal Modeling for Traffic Forecasting

Authors: Kenan Kang, Matthew M. Y. Kuo, Weihua Li

Organizations: Department of Computer Science and Software Engineering, Auckland University of Technology, Auckland 1010, New Zealand · Department of Data Science and Artificial Intelligence, Auckland University of Technology, Auckland 1010, New Zealand

Abstract

Modeling multi-scale dynamics in urban road networks is essential for intelligent transportation systems but is often hindered by the computational cost of long-range historical context and the challenge of fusing heterogeneous modalities. Existing spatio-temporal frameworks face a trade-off between sequence length and memory overhead, while frequently neglecting the transient effects of exogenous factors such as road incidents and weather. To address these limitations, we propose SpecLST, a frequency-enhanced long-short-term spatio-temporal multimodal framework. SpecLST leverages a fast Fourier transform-based frequency-domain representation module to efficiently encode multi-week historical dependencies, bypassing the memory bottlenecks of standard attention mechanisms over raw long sequences. Heterogeneous context is integrated through attention-based fusion, followed by a sparse Mixture-of-Experts backbone with graph augmentation for spatio-temporal representation learning. Additionally, a frequency-enhanced residual head isolates high-frequency transients, improving sensitivity to abrupt traffic anomalies. At the 60-minute horizon, SpecLST achieves mean relative error reductions of 14.10% in MAE, 10.35% in RMSE, and 15.78% in MAPE across eight neural forecasting baselines and four datasets, with the dataset- and metric-specific average reductions ranging from 5.63% to 22.21%.

Figures & tables

Fig. 1: An overview of the SpecLST framework.

Datasets	Methods	Horizon 3			Horizon 6			Horizon 12
		MAE	RMSE	MAPE	MAE	RMSE	MAPE	MAE	RMSE	MAPE
PEMS-BAY	VAR	1.90	4.13	3.92%	2.47	5.57	5.36%	3.42	7.60	7.75%
	SVR	1.52	3.55	3.48%	1.85	4.41	4.43%	2.34	5.61	6.00%
	DCRNN	1.36	2.90	2.90%	1.70	3.90	3.90%	2.07	4.82	4.90%
	MTGNN	1.33	2.79	2.81%	1.65	3.74	3.75%	1.94	4.49	4.62%
	D 2 STGNN	1.24	2.60	2.58%	1.55	3.52	3.49%	1.85	4.30	4.37%

TABLE I: Prediction performance of traffic forecasting models on four datasets. Best results are in bold . Second-best results are underlined .

Reference	Setting	H3			H6			H12
Reference	Setting	$\Delta$ MAE	$\Delta$ RMSE	$\Delta$ MAPE	$\Delta$ MAE	$\Delta$ RMSE	$\Delta$ MAPE	$\Delta$ MAE	$\Delta$ RMSE	$\Delta$ MAPE
$\Delta$ to $BS=4$	8	$-0.054$	$-0.126$	$-0.23$	$-0.063$	$-0.130$	$-0.18$	$-0.090$	$-0.117$	$-0.35$
	16	$+0.016$	$-0.074$	$-0.08$	$+0.012$	$-0.076$	$+0.02$	$+0.029$	$-0.060$	$-0.06$
	32	$+0.063$	$-0.045$	$+0.04$	$+0.044$	$-0.066$	$+0.04$	$+0.056$	$0.000$	$-0.13$
$\Delta$ to $D=64$	128	$-0.145$	$-0.083$	$-0.31$	$-0.210$	$-0.142$	$-0.50$	$-0.259$	$-0.187$	$-0.56$
	256	$-0.154$	$-0.096$	$-0.40$	$-0.241$	$-0.197$	$-0.56$	$-0.307$	$-0.235$	$-0.64$

TABLE II: Sensitivity results on PEMS-BAY22 reported as absolute differences from the corresponding reference settings. Negative values indicate improved performance, and

\Delta

MAPE is reported in percentage points. Best results within each group and column are in bold .

	Horizon 3			Horizon 6			Horizon 12
Method	MAE	RMSE	MAPE	MAE	RMSE	MAPE	MAE	RMSE	MAPE
w/o LAM	13.42	23.41	9.66%	14.42	25.71	10.35%	16.01	28.96	11.64%
w/o AM	13.46	23.51	9.57%	14.46	25.84	10.23%	16.04	29.06	11.52%
Original	13.19	23.17	9.41%	14.13	25.39	10.06%	15.61	28.54	11.26%

TABLE III: Adjacency matrix ablation experiment results. Best results per column are in bold .

	Horizon 3			Horizon 6			Horizon 12
Method	MAE	RMSE	MAPE	MAE	RMSE	MAPE	MAE	RMSE	MAPE
w/o CA	13.23	23.21	9.45%	14.19	25.39	10.13%	15.69	28.62	11.34%
w/o WE	13.22	23.23	9.53%	14.18	25.47	10.18%	15.73	28.62	11.49%
w/o EV	13.26	23.24	9.43%	14.23	25.52	10.12%	15.79	28.77	11.40%
traffic only	13.35	23.27	9.51%	14.32	25.55	10.21%	15.88	28.77	11.46%
Original	13.19	23.17	9.41%	14.13	25.39	10.06%	15.61	28.54	11.26%

TABLE IV: Exogenous modality ablation experiment results. Best results per column are in bold .

	Horizon 3			Horizon 6			Horizon 12
Method	MAE	RMSE	MAPE	MAE	RMSE	MAPE	MAE	RMSE	MAPE
w/o FRH	13.23	23.16	9.43%	14.15	25.41	10.03%	15.66	28.63	11.27%
w/o FFT	13.48	23.47	9.61%	14.55	25.86	10.34%	16.33	29.34	11.71%
w/o Spectral	13.61	23.60	9.77%	14.68	26.03	10.47%	16.55	29.67	11.94%
Original	13.19	23.17	9.41%	14.13	25.39	10.06%	15.61	28.54	11.26%

TABLE V: Spectral component ablation experiment results. Best results per column are in bold .

Panel of Figure 7 — Fig. 2: Efficiency comparison using Pareto frontier analysis on PEMS-BAY22.

Spatiotemporal Forecasting

Spatiotemporal forecasting predicts how quantities change across locations and time, from traffic and weather to disease and environmental systems.

See all 427 papers

DSFNet: Learning Dual-Domain Spectral Operators for Multi-Modality Spatio-Temporal Forecasting in Urban Transportation Systems
Jun 5, 2026Yongchao Li, Yang Li, Zhuoxuan Li +4Intelligent Transportation Systems Time Series Forecasting
School of Transportation, Southeast University, Nanjing 211189, China · Jiangsu Province Collaborative Innovation Center of Modern Urban Traffic Technologies, Nanjing 211189, China · Department of Data Science, City University of Hong Kong, Hong Kong 999077, China +6
FAST: A Synergistic Framework of Attention and State-space Models for Spatiotemporal Traffic Prediction
Apr 15, 2026Xinjin Li, Jinghan Cao, Mengyue Wang +5Spatiotemporal Forecasting Traffic Forecasting
Columbia University New York, NY, USA · San Francisco State University San Francisco, CA, USA · University of California, Berkeley Berkeley, CA, USA +5
UniST-Pred: A Robust Unified Framework for Spatio-Temporal Traffic Forecasting in Transportation Networks Under Disruptions
Feb 15, 2026Yue Wang, Djellel Difallah, Areg Karapetyan +1Intelligent Transportation Systems Spatiotemporal Forecasting
New York University Abu Dhabi

Jun 5, 2026J/K move · Enter open · S save

DSFNet: Learning Dual-Domain Spectral Operators for Multi-Modality Spatio-Temporal Forecasting in Urban Transportation Systems

Yongchao Li, Yang Li, Zhuoxuan Li, Jun Chen+3

School of Transportation, Southeast University, Nanjing 211189, China · Jiangsu Province Collaborative Innovation Center of Modern Urban Traffic Technologies, Nanjing 211189, China · Department of Data Science, City University of Hong Kong, Hong Kong 999077, China+6

Intelligent Transportation Systems Time Series Forecasting Spatiotemporal Forecasting Multimodal Fusion

Open paper AskPDF arXiv

Multi-Modality Spatio-Temporal Forecasting (MoSTF) extends traditional spatio-temporal forecasting by incorporating diverse traffic modalities. Despite significant recent strides in spatio-temporal modeling, existing approaches often fail to explicitly model the coupling relationships between different modality variables. Accurate MoSTF is challenging, as it requires modeling (1) temporal dynamic heterogeneity under exogenous influences and (2) heterogeneous spatial dependencies alongside complex cross-variable couplings. To address these challenges, we propose the Dual-Domain Spectral Filtering Network (DSFNet). Our framework employs dual-domain spectral filtering to capture heterogeneous spatial patterns and explicitly model the relationships between variables. Unlike graph-based message passing or dense attention over node-modality pairs, DSFNet factorizes space-modality interactions into feature-domain and spatial-domain spectral operators, enabling scalable modeling of nonlocal dependencies and cross-modality couplings. Furthermore, we introduce an external gating mechanism to adaptively regulate temporal dynamics under external influences. We validate our method through extensive experiments on five representative real-world traffic datasets. Compared with the second-best baselines, DSFNet reduces MAE by 3.21%-10.16% across these datasets. The results demonstrate that DSFNet significantly outperforms existing state-of-the-art baselines in accuracy while exhibiting efficiency and robustness.

Abstract

Spatiotemporal Forecasting

Explore similar work

DSFNet: Learning Dual-Domain Spectral Operators for Multi-Modality Spatio-Temporal Forecasting in Urban Transportation Systems

FAST: A Synergistic Framework of Attention and State-space Models for Spatiotemporal Traffic Prediction

UniST-Pred: A Robust Unified Framework for Spatio-Temporal Traffic Forecasting in Transportation Networks Under Disruptions