Remote sensing provides an efficient and effective tool to monitor inland water quality. Various empirical algorithms have been developed to retrieve water quality constituents in inland waters from remotely sensed imagery. The common practice of previous studies has been to calibrate a single empirical model for the entire study area. However, the performance of a single empirical model is often limited for optically complex inland waters. Additionally, traditional empirical models are not spatially or temporally extensible, which impose strict and expensive demand for in situ water truth data during the overpasses of space-borne or airborne sensors. To overcome the limitations of traditional empirical models, my dissertation research focuses on the development of novel remote sensing algorithms for deriving water quality parameters in inland waters from multispectral and hyperspectral imagery. First, I present a geographically adaptive algorithm that addresses the adverse effect of spatial heterogeneity and are able to produce much better water quality parameter estimates than conventional global models. Second, I develop a multi-predictor ensemble model that exploits the comparative advantages of a set of diverse empirical models based on spectral space partitions. The multi-predictor ensemble model has significantly enhanced the water quality prediction accuracy and particularly possessed the desired model extensibility in space and time. Given its strong spatial extensibility, I finally apply the multi-predictor ensemble model to rivers in a large basin for regional water quality analysis. The spatial and temporal extensibility of the multi-predictor ensemble model greatly decreases the operational cost and difficulty, hence facilitating regional scale and long-term water quality monitoring and assessment with remote sensing data.