The stress on water resources is increasing as a result of population growth, global warming, urbanization and many other interrelated factors. Because of this, efforts to continuously model the key states in the hydrologic system, specifically streamflow, have resulted. Unfortunately, large scale streamflow simulation in hydrologic sciences has not kept up with the development of large scale numerical weather models in atmospheric sciences. This in part can be attributed to the complexity of the modeling domain where numerical weather models can predominantly run on a continuous gridded domain whereas streamflow simulation models require the integration of both continuous gridded domains (e.g. land surface, subsurface) and continuous discrete feature domains (e.g. river channel network, lakes and reservoirs, point observations). As such, this dissertation presents a fundamental geospatial framework for supporting continuous large scale streamflow simulations through both regional and continental scale applications. As this effort requires myriad water data (much of which is becoming increasingly available via Web services), this dissertation begins with a discussion of atmospheric and hydrologic data in continuous and discrete space within the context of data dissemination. At the regional scale it is demonstrated how data from disparate data sources can be integrated to simulate real-time streamflow for over 5,000 NHDPlus river reaches while at the continental scale, a streamflow simulation system operating on 2.7 million NHDPlus river reaches is presented.

This study develops a general discrete flow model which is discrete both in time and space. A stirred tank network model (continuous time compartment model) is summarized and then compared with the discrete flow model. Both models make use of a fractional input matrix with elements representing the fraction of flow into the jth region which originates in the ith region. It is shown that the discrete flow model is more general and computatiomally much simpler than the stirred tank network model. The fractional input matrix is used to find the sizes of the regions, to predict the residence time distribution (RTD) of the fluid in flow systems, and to compute the conversion for chemical reaction occurring in a flow system. Direct methods for finding the fractional input matrix from either steady state or transient data are developed, and methods and problems of fitting the model to data are presented. Two simple numerical examples are given to demonstrate the use of the model. Lumping is important for modeling flow systems with very many regions, thus a lumping analysis for discrete flow systems is presented. The methods for obtaining the sizes of the regions and the fractional input matrix provide an important tool to analyze the lumping problems in flow systems.

This textbook is a practical guide to real-time streamflow forecasting that provides a rigorous description of a coupled stochastic and physically based flow routing method and its practical applications. This method is used in current times of record-breaking floods to forecast flood levels by various hydrological forecasting services. By knowing