Existing in situ global sediment monitoring limits our potential to answer regional and continental scale sediment transport questions. Remote sensing of rivers has an established history addressing these spatial and temporal data gaps but suffers from resolution and overpass timing tradeoffs. Satellite spatiotemporal data fusion can help alleviate these tradeoffs by “teaching” a coarse resolution sensor what it would have seen if it were a fine resolution satellite. In this study, we fuse Landsat/Sentinel-2 with MODIS to produce 30 m spatial data with Landsat bands and MODIS repeat frequency. We then create a database of 1.8M large river sediment concentrations derived from fused and nonfused images made over 1253 Continental United States sites, as validated with over 25 000 in situ measurements across these same sites. Combining fused and original images using a globally scalable workflow more than doubles training and prediction image data density with marginal loss in overall performance (combined model relative error 36% versus 33% landsat versus 39% fusion) and no loss in stability. Furthermore, we test our global model's ability to generate meaningful time series for individual large and small rivers included in and exogenous to the training data. Results indicate that global model skill holds when applied at a site, even out of sample, which paves the way for future sediment mapping in the MODIS era and shows the value of fusion for addressing hydrologic measurement challenges.