Long-term monitoring of water quality in large irrigation systems is often limited by incomplete biochemical oxygen demand (BOD \(_{5}\) ) records due to the time-consuming and resource-intensive nature of laboratory analysis. This study develops a COD-based linear regression approach to reconstruct missing BOD \(_{5}\) data and evaluates its impact on the evaluation of the Water Quality Index (WQI) in the Bac Hung Hai (BHH) Irrigation System, Vietnam. A 20-year dataset (2004–2024) comprising 3014 observations was analyzed, with approximately 42.9% of BOD \(_{5}\) values missing. The missing-data mechanism was assessed and can be reasonably approximated as Missing At Random (MAR) conditional on COD. Model performance was evaluated using a combination of train–test split and tenfold cross-validation. Results indicate a strong and stable linear relationship between BOD \(_{5}\) and COD ( \(R^{2} = 0.953\) ), with low prediction errors (RMSE = 2.94 mg L \(^{-1}\) , MAE = 1.82 mg L \(^{-1}\) ). The model successfully reconstructed missing BOD \(_{5}\) values while preserving the original statistical distribution and the intrinsic BOD \(_{5}\) –COD relationship. Uncertainty analysis showed that 95% prediction intervals achieved a coverage of 94.7%. Comparative analysis of WQI before and after reconstruction revealed statistically significant but practically negligible differences, with no change in water quality classification outcomes. These results demonstrate that a simple and interpretable regression approach can effectively restore missing data without compromising long-term water quality assessment, providing a practical solution for monitoring programs in data-limited environments.
