Aim of the study:
Accurate forecasting of groundwater levels is essential for sustainable water-resource management in data-limited settings. This study develops an operational, reproducible workflow for forecasting a regional shallow-groundwater index based only on historical monitoring records. Monthly groundwater-level observations from six automatic monitoring wells (P1‒P6) in the southeastern coastal plain of Nghe An province (Vietnam) were obtained from the Nghe An Environmental Monitoring Center. For each month, the regional series was calculated as the arithmetic mean of the six well levels, providing a single representative indicator for the study area.

Material and methods:
Monthly groundwater-level data from May 2013 to April 2025 were analyzed using three forecasting approaches: seasonal naïve (SNaïve), seasonal autoregressive integrated moving average (SARIMA), and exponential smoothing state space (ETS). The dataset was divided into a training period (May 2013–April 2024) and a testing period (May 2024–April 2025). Model performance was assessed using RMSE, MAE, and MAPE, supported by residual diagnostics and the corrected Akaike information criterion (AICc), to ensure model adequacy and parsimony.

Results and conclusions:
The ETS model produced the lowest forecast errors and generated residuals closest to white noise, outperforming both SNaïve and SARIMA. These results demonstrate that the ETS offers a robust and reliable framework for forecasting groundwater levels one year ahead . The model’s performance provides valuable support for irrigation planning, drought preparedness, and sustainable aquifer management in regions characterized by strong seasonal dynamics.