Exploring Machine Learning and Deep Learning Techniques for Univariate and Multivariate Time Series Forecasting in the Area of Barisal in Bangladesh

Abstract

Bangladesh, a densely populated country in South Asia, faces substantial challenges in water resources management due to its geographic location, hydrological features, and socio-economic aspects. Groundwater plays a dynamic role in meeting the water needs of both rural and urban population, accounting for approximately 85% of the total drinking water supply and a significant portion of irrigation water. With rapid population growing, urbanization, and industrialization, the demand for groundwater is constantly growing, increasing pressure on existing water resources and necessitating effective managing approaches. The consequence of this study lies in its potential to improve the correctness and trustworthiness of groundwater level forecasts, thereby supporting more conversant decision-making in water resources management, urban planning, and structure development. By using advanced machine learning and deep learning performances, this research challenges to fill the gaps that exist in traditional predicting methods and contribute to the justifiable use of groundwater resources in Bangladesh.

Introduction

Objective:

To develop accurate, data-driven forecasting models for groundwater level (GWL) in Barisal by leveraging advanced ML and DL techniques, in response to the area's growing agricultural demands and environmental challenges.

Key Elements of the Study:

Study Context:

• Barisal, a southern region of Bangladesh, is highly reliant on groundwater.

• GWL is influenced by climate factors (rainfall, temperature, humidity) and soil moisture (surface, root zone, profile).

Problem:

Traditional numerical models struggle with the nonlinear and dynamic nature of GWL data, prompting exploration of ML/DL approaches.

Methodology:

1. Data Collection:

• Sources: Bangladesh Water Development Board (BWDB) and NASA

• Parameters: Groundwater level, rainfall, temperature, humidity, and soil moisture (collected from 2010 to 2021)

2. Data Preprocessing:

• Cleaning: Handle missing values

• Normalization: Scale variables for consistency

• Feature Engineering: Create lagged variables and derived features like cumulative rainfall

3. Models Applied:

• Deep Learning: LSTM, GRU, LSTM+GRU (hybrid)

• Machine Learning: Random Forest (RF), K-Nearest Neighbors (KNN), Support Vector Regression (SVR)

4. Model Training:

• DL models: Trained using backpropagation and Adam optimizer

• ML models: Trained using grid search and cross-validation

• Evaluation Metrics: MAE, RMSE, R², MAPE

5. Performance Evaluation:

• RF showed the highest accuracy (R² ≈ 0.7).

• LSTM and GRU captured temporal dependencies well.

• KNN had the lowest accuracy (R² ≈ 0.4).

• LSTM+GRU performed better than individual RNN models.

6. Data Splitting Strategy:

• 65% for training, 35% for testing/validation

• Sequential splitting for time-series models (LSTM, GRU)

• Random splitting and k-fold cross-validation for ML models (RF, SVR, KNN)

Execution Tools:

• Languages: Python, R

• Libraries: TensorFlow, Keras, Pandas, Scikit-learn

• Hardware: High-performance systems with GPU support

Key Insights:

• DL models (especially hybrid LSTM+GRU) effectively captured temporal groundwater patterns.

• RF provided strong performance for modeling nonlinear relationships and feature importance.

• Combining models offers improved forecast reliability and management insights.

• The methodology can be replicated in other regions with similar environmental profiles.

Impact:

Supports sustainable groundwater management, improves forecasting accuracy, and assists agricultural planning and climate resilience in Barisal and potentially similar geographic areas.

Conclusion

This research efficaciously explored and realized various machine learning (ML) and deep learning (DL) techniques, including LSTM, GRU, Random Forest (RF), KNN, SVR, and a Hybrid LSTM+GRU model, to forecast critical ecological and groundwater-related parameters, such as groundwater level, rainfall, humidity, temperature, and soil moisture as like as surface, root, and profile, in the Barisal Division of Bangladesh. The study as long as valuable insights into the complex relationships between these variables, leveraging historical and temporal data for precise estimating. Deep learning models, particularly LSTM and Hybrid LSTM+GRU, consistently outperformed traditional ML models due to their ability to capture temporal dependances in the time-series data. The Hybrid LSTM+GRU model verified the best overall performance with high exactness and strength across all strictures. Rainfall, temperature, and humidity were found to be the most significant features for predicting groundwater levels and soil wetness. Integrating lagged features and rolling means pointedly enhanced model correctness. The strong seasonality and variability of ecological data in Barisal Division required erudite time-series models to efficiently capture long-term and short-term shapes. The prognostic models can be useful to design effective groundwater management plans, optimize irrigation timetables, and prepare for drought surroundings. This research contributes to the rising body of information on using ML and DL techniques for bearable resource managing in water-scarce areas. It makes available a groundwork for future research into groundwater and soil dynamics by integrating advanced analytical modeling techniques tailored to local circumstances in Barisal Division. The findings can aid legislators, water resource managers, and agricultural planners in emerging data-driven strategies for resource optimization and climate version. performance. Range the research to integrate spatial variability within Barisal Division using GIS-based modeling to capture regional alterations more successfully. Examine the role of climate change in inducing long-term trends in groundwater levels, soil moisture, and rainfall patterns in Barisal Division. Develop real-time projecting systems integrated with local decision-making contexts to provide actionable insights for water and agricultural management. The research verified the potential of advanced ML and DL techniques in addressing critical encounters in water resource managing. By leveraging state-of-the-art analytical models, participants in Barisal Division can adopt sustainable practices to safeguard effectual use of groundwater, enhance agricultural activities, and moderate the risks associated with climate changeability.

References

[1] Adadi, A., & Berrada, M. (2018). Peeking inside the black-box: A survey on explainable artificial intelligence (XAI). IEEE Access, 6, 52138-52160. [2] Anderson, M.P., Woessner W.W., and Hunt R.J. (2015). Applied Groundwater Modeling: Simulation of Flow and Advective Transport. London: Academic Press. [3] Collenteur, R., Bakker M., Caljé R., Schaars F., and Klop S. (2019). Pastas: open-source software for the analysis of groundwater time series. Groundwater, 10.1111/gwat.12925. [4] Gelaro, R., McCarty, W., Suárez, M. J., et al. (2017). Modern-era retrospective analysis for research & applications, version 2 (MERRA-2). Journal of Climate, 30(14), 5419-5454. [5] Gharbi, S., & Bouaziz, M. (2023). Real-time data assimilation for GWL prediction using machine learning. Water Resources Management, 37(5), 1461-1478. DOI. [6] Haurie, A., & Ghaffari, A. (2022). Explainable AI in hydrology: Challenges and opportunities. Environmental Modelling & Software, 155, 105454. DOI. [7] Karthikeyan, L., Khosa, R., & Singh, V. P. (2020). Deep learning models for soil moisture retrieval from remote sensing data: A review. Environmental Earth Sciences, 79(7), 193. [8] Kratzert, F., Klotz, D., Shalev, G., et al. (2019). Towards improved predictions in ungauged basins: Exploiting the power of ML. Water Resources, 55(12), 11344-11354. [9] Lundberg, S. M., & Lee, S. I. (2017). A unified approach to interpreting model predictions. In Advances in Neural Information Processing Systems (pp. 4765-4774). [10] Sutskever, I., Vinyals, O., & Le, Q. V. (2014). LSTM in Natural Language Processing: Sequence to Sequence Learning with NN. Advances in Neural Information Processing [11] Mojid, M.A., Parvez, M.F., Mainuddin, M. and Hodgson, G., (2019), Water Table Trend- A Sustainability of GWL Development in North-West Bangladesh, Water, Vol 11, pp-1182. [12] Obergfell, C., Bakker M., and Maas K. (2019). Identification and explanation of a change in the groundwater regime using time series analysis. Groundwater. 10.1111/gwat.12891 [13] Peterson, T.J., and Western A.W. (2018). Statistical interpolation of groundwater hydrographs. Water Resources Research 54, no. 7: 4663–4680. [14] Saltelli, A., & Annoni, P. (2010). How to avoid a perfunctory sensitivity analysis. Environmental Modelling & Software, 25(12), 1508-1517. DOI. [15] Shi, X., Chen, Z., Wang, H., et al. (2015). Convolutional LSTM network: ML approach for precipitation nowcasting. Advances Neural Information Processing Systems, 28, 802-810. [16] Simoni, S., & Manoli, G. (2021). Monte Carlo-based sensitivity analysis in groundwater simulation models. Environmental Modelling & Software, 136, 104979. DOI. [17] Sun, W., & Wang, S. (2021). Gradient-based sensitivity analysis in neural networks for groundwater modeling. Water, 13(3), 756. DOI. [18] Tang, J., Zhang, X., & Li, Q. (2021). Groundwater level prediction using a Transformer-based model. Journal of Hydrology, 594, 125707. DOI. [19] Wani, S. P., & Wani, M. H. (2021). Application of Gated Recurrent Units for groundwater level forecasting. Hydrology, 8(2), 61. DOI. [20] Wunsch, A., Liesch T., and Broda S. (2018). Forecasting GWL nonlinear autoregressive networks with exogenous input (NARX). Journal of Hydrology 567: 743–758. [21] Vaswani, A., et al. (2017). Attention is all you need. In Proceedings of the 31st International Conference on Neural Information Processing Systems (pp. 6000-6010). [22] Zarei, A. R., & Sepaskhah, A. R. (2022). Prediction of GWL in an arid region using ML and ensemble methods. Hydrological Sciences Journal, 67(7), 1150-1165. DOI.

Copyright

Copyright © 2025 Rakib Uddin. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.