Naik (2025) Comparative Study of Wavelet–ANN and Wavelet– ARIMA Models for Groundwater Level Forecasting

Identification

• Journal: International Journal on Science and Technology
• Year: 2025
• Date: 2025-12-25
• Authors: Rashmi Naik
• DOI: 10.71097/ijsat.v16.i4.9944

Research Groups

• Department of Mathematics, DCT’s Dhempe College of Arts and Science, Miramar, Goa, India.

Short Summary

This study evaluates hybrid forecasting models for groundwater levels in Britona, Goa, comparing Wavelet Transform integrated with Artificial Neural Networks (WT+ANN) against Wavelet Transform with ARIMA (WT+ARIMA). The results indicate that WT+ANN is superior for capturing nonlinear fluctuations and flood forecasting, while WT+ARIMA is better suited for long-term baseline trend analysis.

Objective

• To assess the comparative performance of WT+ANN and WT+ARIMA hybrid models in forecasting groundwater levels (GWL) to support flood risk mitigation and sustainable water resource management.

Study Configuration

• Spatial Scale: Localized study focused on Britona, North Goa, India, situated along the Mandovi River estuary.
• Temporal Scale: January 2013 to October 2023, utilizing quarterly observations (frequency of 4 per year).

Methodology and Data

• Models used: Maximal Overlap Discrete Wavelet Transform (MODWT) using the Haar wavelet for decomposition; Feedforward Artificial Neural Network (ANN) with a 3-5-1 architecture; AutoRegressive Integrated Moving Average (ARIMA) model, specifically ARIMA(4,1,0).
• Data sources: Groundwater level data (measured in meters below ground level) sourced from the India Water Resources Information System (India-WRIS). Missing values were handled using seasonal decomposition and interpolation (na_seadec).

Main Results

• WT+ANN Performance: Demonstrated high predictive accuracy with a coefficient of determination ($R^2$) of 0.9148 and a correlation coefficient ($r$) of 0.9565. It effectively captured nonlinear dynamics and sharp fluctuations (RMSE = 0.2873 m; MAE = 0.2285 m).
• WT+ARIMA Performance: While showing extremely low error metrics during training (RMSE = 0.0075 m; MAE = 0.0059 m), it failed in out-of-sample forecasting ($R^2 = -1.4494$) and showed a lower correlation ($r = 0.4569$), indicating an inability to replicate extreme variability.
• Application Alignment: WT+ANN is identified as the optimal choice for real-time flood forecasting and early warning systems, whereas WT+ARIMA is more appropriate for long-term strategic policy planning and baseline trend analysis.

Contributions

• Provides a comparative analysis of hybrid decomposition-based models specifically for monsoon-dominated coastal aquifers.
• Establishes a selection criterion for hydrological models based on the intended application—distinguishing between immediate risk response (ANN-based) and long-term resource management (ARIMA-based).
• Validates the use of MODWT to preserve time series length and improve feature extraction for neural network training in groundwater studies.

Funding

• No specific funding projects, programs, or reference codes were disclosed in the provided text.

Citation

@article{Naik2025Comparative,
  author = {Naik, Rashmi},
  title = {Comparative Study of Wavelet–ANN and Wavelet– ARIMA Models for Groundwater Level Forecasting},
  journal = {International Journal on Science and Technology},
  year = {2025},
  doi = {10.71097/ijsat.v16.i4.9944},
  url = {https://doi.org/10.71097/ijsat.v16.i4.9944}
}

Generated by BiblioAssistant using gemini-3-flash-preview (Google API)