What does this research mean for the field?

An integrated modeling framework combining ARIMA and multivariate artificial neural networks incorporating rainfall data accurately predicts groundwater level fluctuations, with ARIMA slightly outperforming ANN for linear patterns while ANN better captures nonlinear dynamics. Novelty: ClaimNovelty.METHODOLOGICAL. Consensus alignment: ConsensusAlignment.NEUTRAL.

What question did this study set out to answer?

This research aims to accurately assess groundwater level fluctuations using statistical and machine learning methods in north-western Bangladesh.

May 19, 2026Open Access

Assessment of Groundwater Level Variation by ANN and ARIMA Modeling Coupled with Mann-Kendall Trend Analysis and GIS Spatial Interpolation

Key Points

This research aims to accurately assess groundwater level fluctuations using statistical and machine learning methods in north-western Bangladesh.
Developed 1008 artificial neural network (ANN) and 768 autoregressive integrated moving average (ARIMA) models under univariate and multivariate settings.
Utilized ArcGIS for spatial groundwater level and trend analysis.
Evaluated prediction accuracy using metrics such as MSE and correlation coefficient.
ANN models improved predictive accuracy with a MSE of 0.01 and R² of 0.97, influenced by rainfall data.
ARIMA models showed linear pattern superiority over ANN, yet ANN captured nonlinear dynamics effectively.
Significant increases in groundwater levels were identified at Nawabganj.

Abstract

Groundwater is the largest liquid freshwater reservoir and a critical resource for drinking water, agriculture, and ecosystem sustainability. In the Rajshahi Division of north-western Bangladesh, intensive groundwater use and recurring droughts necessitate an accurate assessment of groundwater level (GWL). This study presents an integrated framework combining statistical, machine learning, and spatial tools for a comprehensive assessment of GWL fluctuations. A total of 1008 artificial neural network (ANN) models and 768 autoregressive integrated moving average (ARIMA) models were developed and evaluated under univariate (UV) and multivariate (MV) settings to determine optimal model functions and lagged effects. Spatial GWL and trend variations were generated using ArcGIS. Results show that ANN models incorporating rainfall (RF) data improved predictive accuracy (e.g., Bogura, MSE 0.01, Correlation Coefficient R = 0.98, NSE = 0.97, KGE = 0.96, R 2 = 0.97) compared to UV models. ARIMA models slightly outperformed ANN due to linear patterns, while ANN captured nonlinear dynamics. Trend analysis indicated significant GWL increases at Nawabganj. This framework provides a robust, transferable approach for evaluating GWL, enabling sustainable irrigation planning, drought risk management, identification of vulnerable zones, and evidence-based groundwater management in other hydrologically stressed regions.

Read Full Paperexternally

Bookmark

View Full Paper

Cite This Study

Hoque et al. (Fri,) studied this question.

synapsesocial.com/papers/6a0bfe2d166b51b53d3796cf https://doi.org/https://doi.org/10.1016/j.htopen.2026.100034

Bookmark

View Full Paper