Hydrological Time Series Modeling and Prediction via Time–Frequency Image Representation

Read full text Summarise with VIRAT
Authors: Hua Fan, Dongfang Shen
Cited by: 0
References: 23

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

Hydrological time series are significantly nonlinear and nonstationary and their accurate prediction is a fundamental challenge in hydrology and water resources.It is difficult for existing deep learning-based time series models to effectively capture dynamic timefrequency characteristics, while the integration of image processing techniques into hydrological forecasting has received insufficient attention, constraining the completeness of temporal feature representation and predictive accuracy.To address these limitations, based on time-frequency image representation, this study proposes an end-to-end deep learning prediction framework by establishing a complete modeling pipeline from temporal signals to high-precision regression prediction through the coordination of three modules: adaptive time-frequency image generation, a deep convolutional backbone network coupling spatial-frequency-temporal features (CFT-Net), and dual-stream fusion prediction.A learnable adaptive time-frequency transform dynamically generates multiscale time-frequency images according to local statistical characteristics of hydrological sequences, encoding one-dimensional temporal dynamics into twodimensional spatial patterns.Using a parallel spatial-frequency dual-path architecture with a dual attention mechanism, the core backbone network jointly extracts spatial patterns, frequency distributions, and temporal evolution features from time-frequency images.Furthermore, an adaptive gated fusion strategy achieves complementary integration between the time-frequency image stream and the raw time series stream.Experimental results on representative hydrological datasets demonstrate that the proposed framework consistently outperforms conventional time series models and image-based prediction approaches relying on fixed time-frequency transforms in terms of prediction accuracy and robustness.These findings confirm the effectiveness and superiority of image processing techniques for hydrological time series modeling and provide a promising new pathway for high-precision hydrological forecasting.

Fields of Study & Subjects

Hydrological Forecasting Using AI Time Series Analysis and Forecasting Traffic Prediction and Management Techniques

Publication details

Journal / source
Traitement du signal
Volume
43
Issue
1
Publication date
28 February 2026
Language
EN