Accidental spills and illegal chemical discharges have occurred worldwide, causing adverse effects on human health and the ecosystem. Inverse models play an essential role in source identification based on limited concentration data observed at monitoring sites. However, most studies have been conducted on the inverse inference of pollution sources in atmospheric and groundwater environments. Few studies have focused on identifying pollution sources in rivers. In addition, multiple source identification, uncertainty quantification and sensitivity analysis of inverse models are rarely considered. An integrated inverse modeling system is developed in this thesis for source identification in river pollution incidents, including three independent inverse modeling approaches for an instantaneous point source (IPS), a continuous point source (CPS), and multi-point sources (MPS), respectively. The inverse modeling approach estimates source parameters with uncertainty concerns by combining a water quality model and a probabilistic inverse method based on observed pollutant concentrations.
First, the IPS inverse modeling approach is developed based on the Metropolis-Hastings (MH) method. Then the developed MH-based IPS approach is tested and compared with a genetic algorithm (GA)-based IPS approach for a hypothetical case and a real case study. Results confirm that the MH-based IPS approach performs better than the GA-based IPS approach in terms of accuracy and stability for IPS source identification. According to the sensitivity analysis, the emission mass of the pollution source positively correlates with the dispersion coefficient and the river cross-sectional area, whereas the flow velocity significantly affects the identified values of release location and release time. Second, the CPS inverse modeling approach employed the up-to-date DiffeRential Evolution Adaptive Metropolis (DREAM) algorithm to estimate source parameters. The DREAM-based CPS inverse modeling approach accurately performs in a hypothetical case and a field tracer case. Moreover, compared to the MH-based and GA-based CPS approaches for CPS identification, the DREAM-based CPS approach has an advantage in accuracy, computation time, and reconstructing the time series of pollutant concentrations. The accuracy of the approach can be improved by decreasing observation errors, increasing the monitoring number, and deciding monitoring locations closer to the spill site.  Third, the MPS inverse modeling approach is developed based on the DREAM algorithm. The one-point, two-point, and three-point source problems are considered in case studies to validate the feasibility and accuracy of the DREAM-based MPS approach. Among the three identified source parameters, the identification error of the release time tends to rise obviously in response to the increase in pollution sources. Subsequently, an integrated inverse modeling system with a graphical user interface is established based on the three developed inverse modeling approaches, and its efficiency is tested through case studies. In conclusion, the integrated system can serve as a helpful tool for source identification, model validation, and pollution prediction in the assessment and management of emergency pollution incidents.