Stream water quality is at the core of the biological and ecological health of streams and rivers. Subject to the environmental variation and in-stream metabolic activities, key water quality indicators such as dissolved oxygen (DO), water temperature (Tw), and chlorophyll a (Chl a, an indicator of primary productivity) typically follow a concave-shaped diurnal cycle. Variation and changes in climatic drivers (e.g., solar radiation, air temperature, and pressure), land uses, and hydrologic variables lead to a high spatiotemporal variability in stream water quality. The spatiotemporal variation challenges the development of robust models for predictions across large scales (e.g., county, state, country), which are relevant for environmental management and policy making. Scaling resolves the spatiotemporal variation of water quality variables and thereby widens the model applicability. This dissertation focuses on developing scaling-based, robust empirical models to predict the diurnal cycles of hourly DO, Tw, and Chl a in streams and rivers across the contiguous U.S. For each of the periodic water quality indicators, scaling of the diurnal cycles representing different streams and days by the corresponding reference observations resulted in a generalizable (i.e., robust) dimensionless cycle, which was estimated with an extended stochastic harmonic algorithm. Hourly observations of the extended growing season (May ‒ October) during 2008 ‒ 18 from numerous streams and monitoring stations across the contiguous U.S. were obtained from the U.S. Geological Survey for model evaluations. The study sites represented gradients in latitude, elevation, drainage area, and land uses. The daily estimated set of parameters in model calibrations collapsed within a comparable range across the growing season, leading to a temporally ensemble parameter set for each stream. Linkages of the site-specific model parameters with the corresponding climatic, hydrologic, and land use variables were then investigated over the contiguous U.S. Results suggested similarity of parameters across various streams and rivers based on similarities in latitude, elevation, drainage area, and land use types of the monitoring sites. Accordingly, category-specific and continental-scale generalized sets of parameters were computed to evaluate the spatiotemporal robustness of the scaling models. The hourly observations of independent validation periods were then predicted using the corresponding site-specific, category-specific, and generalized sets of parameters. As expected, the site-specific parameter sets provided the most accurate predictions of the water quality indicators. Remarkably, the category-specific sets of model parameters provided predictions similar to the site-specific sets. Further, the continental-scale generalized single set of parameters resulted in less accurate, but acceptable predictions for most streams and rivers across the contiguous U.S. The prediction performance reflected the spatiotemporal robustness of the scaling-based modeling framework and estimated parameters. The robustness of model parameters was further evaluated by quantifying analytical sensitivity and uncertainty measures. The scaling-based empirical models can predict the entire diurnal cycles of periodic water quality variables from the corresponding single reference observations in minimally gauged and ungauged streams. The models can also be used to fill gaps in existing time-series of continuously monitored streams. Availability of high resolution (e.g., hourly) information of water quality indicators would aid a dynamic assessment of ecosystem health in streams and rivers across the contiguous U.S. and beyond.