Liu et al. (2026) Analyzing the Impact of High-Frequency Noise on Hydrological Runoff Modeling: A Frequency-Based Framework for Data Uncertainty Assessment

⚠️ Warning: This summary was generated from the abstract only, as the full text was not available.

Identification

• Journal: Water
• Year: 2026
• Date: 2026-01-12
• Authors: Tianxu Liu, Wenyu Ouyang, Muhammad Adnan, Chi Zhang
• DOI: 10.3390/w18020195

Research Groups

Not explicitly mentioned in the provided text.

Short Summary

This study proposes a frequency-based framework to systematically evaluate the impact of high-frequency noise on LSTM runoff prediction models and introduces an adaptive exponentially weighted moving average (AEWMA) algorithm to mitigate performance degradation while preserving hydrological signals.

Objective

• To systematically evaluate the impact of high-frequency noise, categorized by hydrological characteristics, on the performance of deep learning-based hydrological forecasting models (specifically LSTM runoff prediction).

Study Configuration

• Spatial Scale: Catchment scale (implied by CAMELS dataset, which includes multiple catchments).
• Temporal Scale: Relevant to hydrological processes, including long-term trends, short-term events, and transient interference, suggesting daily or sub-daily resolutions for runoff.

Methodology and Data

• Models used: Long Short-Term Memory (LSTM) for runoff prediction; Adaptive Exponentially Weighted Moving Average (AEWMA) algorithm for dynamic smoothing.
• Data sources: CAMELS dataset; Synthetic noise injection strategy.

Main Results

• Model accuracy significantly deteriorates when high-frequency noise exceeds 30% of the total signal energy.
• Moderate adaptive smoothing (e.g., α=0.9 & 0.6) effectively preserves hydrological signals while mitigating performance loss.
• Aggressive smoothing suppresses meaningful hydrological variations, leading to suboptimal performance.

Contributions

• Proposes a novel frequency-based framework to assess the robustness of LSTM runoff prediction models to various types of high-frequency noise.
• Defines three hydrologically meaningful noise types: long-term trend, short-term event, and transient interference.
• Introduces the Adaptive Exponentially Weighted Moving Average (AEWMA) algorithm for dynamic, adaptive smoothing based on local signal variability.
• Provides quantitative thresholds (spectral energy ratios) for adaptive data quality control in hydrological modeling workflows.
• Underscores the necessity of noise-type-specific preprocessing for deep learning-based hydrological forecasting.

Funding

Not explicitly mentioned in the provided text.

Citation

@article{Liu2026Analyzing,
  author = {Liu, Tianxu and Ouyang, Wenyu and Adnan, Muhammad and Zhang, Chi},
  title = {Analyzing the Impact of High-Frequency Noise on Hydrological Runoff Modeling: A Frequency-Based Framework for Data Uncertainty Assessment},
  journal = {Water},
  year = {2026},
  doi = {10.3390/w18020195},
  url = {https://doi.org/10.3390/w18020195}
}