The work presented in this dissertation informs on river network modeling at large scales using geographic information systems, parallel computing and the latest advancements of atmospheric and land surface modeling. This work is motivated by the availability of a vector-based Geographic Information System dataset that describes the networks of streams and rivers in the United States, and how they are connected. A land surface model called Noah-distributed is used to provide lateral inflow to an NHDPlus river network in the Guadalupe River Basin in Texas. Challenges related to the projection of gridded hydrographic data from a coordinate system to another are investigated. The different representations of the shape of the Earth used in atmospheric science (spherical) and hydrology (spheroidal) can lead to a significant North-South shift on the order of 20 km at mid latitudes. A river network model called RAPID is developed and applied in a four-year study of the Guadalupe and San Antonio River Basins in Texas using the river network of NHDPlus. Gage measurements are used to estimate flow wave celerities in a river network and to assess the quality of RAPID flow computations. The performance of RAPID in a massively-parallel computing environment is tested and further investigation of its scalability is needed before using RAPID at the state or federal level. The replacement by RAPID of the river routing scheme used in SIM-France -- a hydro-meteorological model -- is investigated in a ten-year study of river flow in France. While the formulation of RAPID improves the functionality of SIM-France, the flow simulations are comparable in accuracy to those previously obtained by SIM-France. Sub-basin parameterization was found to improve model results. A single criterion for quantifying the quality of river flow simulations using several river gages globally in a river network is developed that normalizes the square error of modeled flow to allow equal treatment of all gaging stations regardless of the magnitude of flow. The use of this criterion as the cost function for parameter estimation in RAPID allows better results than by increasing the degree of spatial variability in model parameters. / text