Remote-sensing relative density (RD) for the conterminous United States: direct prediction from AlphaEarth satellite embeddings

Forest relative density (RD = SDI/SDImax) predicted directly from Google AlphaEarth annual satellite embeddings (10 m) at U.S. Forest Inventory and Analysis (FIA) plot locations, with a national quick-look RD surface for the conterminous United States. This version (1.0.0) provides the reusable mosvr R package (multi-objective support vector regression with a random-forest baseline), the full extraction and modeling pipeline, the anonymous plot-level model inputs and out-of-fold outputs, the trained model fit, cross-validated skill tables, the head-to-head against the TreeMap 2022 RD product (Chivhenge et al. 2025), the disturbance dual-role substitution results, a red-team and stress-test report, and a coarse national RD quick-look raster. The full-resolution 30 m and multi-year (2017-2025) RD, RD-change, and RD-trend GeoTIFFs are planned for a subsequent version. Honest national skill is spatial-block cross-validated R2 ~0.18; the surface is a standardized structural-predisposition layer, not a pixel-level forecast. RD uses the units-consistent SDI/SDImax basis. FIA plot coordinates are never included; all plot-level tables carry an anonymous id plus predictor values and the RD response only.