Gap-filling missing data in time series using the correlation matrix method of multiple time series in Asadabad Plain, Iran

Groundwater resources are crucial sources of water supply, and preserving the quality of these resources is an undeniable necessity. On the other hand, the lack of complete time series data in the observation wells is one of the major limitations in the studies of water resources. In the present study, 17 observation wells were selected at the Asadabad plain, and sampled Electrical Conductivity (EC) statistics and information were prepared. Multivariate regression models, ARIMAX and SVM were considered to simulate and fill the missing EC data of Dehnoush and Biaj wells. Six input structures for the models were determined based on the highest correlation coefficient value, which was defined using the multiple time series method at different delays between the EC of Dehnoush and Biaj stations and that of other stations. The results indicated that increasing the number of inputs to the model reduced the error rate and increased the correlation rate in the simulation. These coefficients reached their minimum value in Structure 6 for the RMSE and NRMSE criteria and their maximum value for the Pearson coefficient. Therefore, to fill in the missing data, other stations' data with the highest correlation at different delays were used instead of using the information of adjacent stations without delay. The results revealed that while the SVM and regression models had relatively similar simulation accuracy, the SVM model exhibited higher accuracy compared to the regression and ARIMAX models.