Considerable uncertainty occurs in the parameter estimates of traditional rainfall–water level transfer function noise (TFN) models, especially with the models built using monthly time step datasets. This is due to the equal weights assigned for rainfall occurring during both water level rise and water level drop events while estimating the TFN model parameters using the least square technique. As an alternative to this approach, a threshold rainfall-based binary-weighted least square method was adopted to estimate the TFN model parameters. The efficacy of this binary-weighted approach in estimating the TFN model parameters was tested on 26 observation wells distributed across the Adyar River basin in Southern India. Model performance indices such as mean absolute error and coefficient of determination values showed that the proposed binary-weighted approach of fitting independent threshold-based TFN models for water level rise and water level drop scenarios considerably improves the model accuracy over other traditional TFN models.

Groundwater level prediction and forecasting using univariate time series models are useful for effective groundwater management under data limiting conditions. The seasonal autoregressive integrated moving average (SARIMA) models are widely used for modeling groundwater level data as the groundwater level signals possess the seasonality pattern. Alternatively, deseasonalized autoregressive and moving average models (Ds-ARMA) can be modeled with deseasonalized groundwater level signals in which the seasonal component is estimated and removed from the raw groundwater level signals. The seasonal component is traditionally estimated by calculating long-term averaging values of the corresponding months in the year. This traditional way of estimating seasonal component may not be appropriate for non-stationary groundwater level signals. Thus, in this study, an improved way of estimating the seasonal component by adopting a 13-month moving average trend and corresponding confidence interval approach has been attempted. To test the proposed approach, two representative observation wells from Adyar basin, India were modeled by both traditional and proposed methods. It was observed from this study that the proposed model prediction performance was better than the traditional model’s performance with R2 values of 0.82 and 0.93 for the corresponding wells’ groundwater level data.

Discrete groundwater level datasets are interpolated often using kriging group of models to produce a spatially continuous groundwater level map. There is always some level of uncertainty associated with different interpolation methods. Therefore, we developed a new trend function with the mean groundwater level as a drift variable in the regression kriging approach to predict the groundwater levels at the unvisited locations. Groundwater level data for 29 observation wells in Adyar River Basin were used to assess the performance of the developed regression kriging models. The cross-validation results show that the proposed regression kriging method in the spatial domain outperforms other physical and kriging-based methods with R2 values of 0.96 and 0.98 during pre-monsoon and post-monsoon seasons, respectively.