The article presents the first national-scale compendium of lake trophic state (LTS) derived from remotely sensed data in the contiguous United States from 1984 to 2020. The dataset, referred to as LTS-US, is constructed using Landsat surface reflectance data and aims to provide standardized, machine-readable observations of lake trophic states. The LTS-US dataset is built with FAIR principles in mind, ensuring data is findable, accessible, interoperable, and reproducible. The study uses a four-part pipeline: identifying parent datasets, creating classification models, applying predictions to lakes outside the training data, and assessing model performance and prediction validity. The classification models include multinomial logistic regression, extreme gradient boosting regression, and a neural network using multilayer perceptrons. The models are trained on in situ measurements from the U.S. EPA's National Lakes Assessment and matched with Landsat surface reflectance data. The final models are applied to predict LTS for 55,662 lakes, providing insights into the spatial and temporal patterns of lake trophic states. The study highlights the importance of considering both nutrient availability and water clarity in classifying LTS, and the results suggest that dystrophic lakes are more prone to misclassification near trophic state thresholds. The LTS-US dataset offers critical data for addressing research questions about lake water quality and ecosystem dynamics.The article presents the first national-scale compendium of lake trophic state (LTS) derived from remotely sensed data in the contiguous United States from 1984 to 2020. The dataset, referred to as LTS-US, is constructed using Landsat surface reflectance data and aims to provide standardized, machine-readable observations of lake trophic states. The LTS-US dataset is built with FAIR principles in mind, ensuring data is findable, accessible, interoperable, and reproducible. The study uses a four-part pipeline: identifying parent datasets, creating classification models, applying predictions to lakes outside the training data, and assessing model performance and prediction validity. The classification models include multinomial logistic regression, extreme gradient boosting regression, and a neural network using multilayer perceptrons. The models are trained on in situ measurements from the U.S. EPA's National Lakes Assessment and matched with Landsat surface reflectance data. The final models are applied to predict LTS for 55,662 lakes, providing insights into the spatial and temporal patterns of lake trophic states. The study highlights the importance of considering both nutrient availability and water clarity in classifying LTS, and the results suggest that dystrophic lakes are more prone to misclassification near trophic state thresholds. The LTS-US dataset offers critical data for addressing research questions about lake water quality and ecosystem dynamics.