Frequent saltwater intrusions in the Chao Phraya River have had an impact on water supply to the residents of Bangkok and nearby areas. Although relocation of the raw water station is a long-term solution, it requires a large amount of time and investment. At present, knowing in advance when an intrusion occurs will support the waterworks authority in their operations. Here, we propose a method to forecast the salinity at the raw water pumping station from 24 h up to 120 h in advance. Each of the predictor variables has a physical impact on salinity. We explore a number of model candidates based on two common fitting methods: multiple linear regression and the artificial neural network. During model development, we found that the model behaved differently when the water level was high than when the water level was low (water level is measured at a point 164km upstream of the raw water pumping station); therefore, we propose a novel multilevel model approach that combines different sub-models, each of which is suitable for a particular water level. The models have been trained and selected through cross-validation, and tested on real data. According to the test results, the salinity can be forecasted with an RMSE of 0.054g/lat a forecast period of 24 h and up to 0.107g/lat a forecast period of 120 h.

Sediment rating curves (SRCs) have been applied to estimate daily-suspended sediment load (Q s ) worldwide because of its simplicity. In this method, current Q s was estimated by a power function of a sole variable, a current daily water discharge at the same measurement station. However, many studies found that its accuracy is not very high. In this study, we developed a new approach to estimate Q s using multivariate hydrological data at the same station and other upstream stations. Using correlation analysis, the additional variables were selected such as upstream water discharges, rainfall at the current or antecedent day. Therefore, spatial and temporal variability was simply considered in our new approach. Then, five methods, a multiple linear regression (MLR), a multiple nonlinear regression (SLR, QLR, and PLR) and an artificial neural network model (ANNs), were applied. The comparison between the SRC method and our new five methods were done using the Q s data at three measurement stations in three basins of Thailand. The results showed that our new approach for all three-study areas (PLR, and ANNs) gave better results with the observed data than the traditional SRC method except MLR, SLR, and QLR. ANNs estimated Q s with the highest accuracy at P1 (EI = 0.96) while PLR gave results similar ANNs at W4A. For Y14 the result of QLR (EI=0.94) better than ANNs Thus, the more complexity of the model structure and the consideration of the spatial and temporal variability can provide a higher accurate estimation of Q s .