Degraded physical habitat is a common stressor affecting river ecosystems and a primary focus of management activities, including stream restorations. In order to assess regional conditions and help prioritize management efforts, there is an ongoing need to provide estimates of different aspects of instream physical habitat conditions at spatially continuous scales. We utilized over 16,000 unique habitat assessments from multiple jurisdictions across the Chesapeake Bay watershed and created a spatially continuous nontidal habitat assessment using predictive random-forest modeling based on landscape attributes. Through this work, we produced predictions for the twelve rapid habitat metrics contained within the EPA Rapid Habitat Protocols (Barbour and others 1999), with a climate-normalized signal at timesteps corresponding to 2001, 2006, 2011, 2016, and 2019 landscape conditions. We also produced two summary habitat metrics, based on principal component analysis, that captured the majority of variability across the Chesapeake Watershed, which was also produced for 2001, 2006, 2011, 2016, and 2019. This data release contains tabular model inputs and outputs of the predictions for the twelve original rapid habitat metrics, plus two summary metrics, for all nontidal NHDPlus v2.1 1:100k catchments/stream reaches for the Chesapeake Watershed for 2001, 2006, 2011, 2016, and 2019. Data are provided in both .csv format and Apache Parquet (.parquet), a free and open-source column-oriented data storage format that is well-suited for large datasets, which enables smaller file sizes, faster reading, faster queries and data manipulations, and the ability to be read in multiple languages (e.g., R, Python, Rust, C++, Java, and more). This data release is the companion to the journal article Cashman and others (in review).