Remote-sensing-assisted small area estimation of forest relative density (CONUS): multi-resolution SAE products, uncertainty, and SAE-calibrated 30 m surface

Version 1.3.0 adds the small area estimation (SAE) product line to the remote-sensing relative-density (RD = SDI/SDImax) deposit. Area-level Fay-Herriot EBLUP fuses the design-based FIA RD estimate per area with a remote-sensing RD auxiliary (RS-RD, from AlphaEarth satellite embeddings) and site-productivity covariates, with analytic Prasad-Rao MSE and model-based 90% intervals. Includes: multi-resolution estimates (county, 10 km sub-county grid, and 6/9/12/18/30 km hexagons); a unit-level Battese-Harter-Fuller comparison; a per-plot quantile-regression-forest prediction-interval layer (90% empirical coverage 0.90); a fair auxiliary-contribution comparison showing RS-RD reduces Fay-Herriot MSE more than the TreeMap product (county variance reduction 41% vs 29%); and a SAE-calibrated 30 m RD surface with a matching credible-interval-width uncertainty surface. Variance reduction versus the direct estimate is approximately 45% at county scale and 68% at the 3-5 plot hexagon scale. FIA plot coordinates were used server-side only and are not included; all tabular products are area-aggregated. See README_SAE_products.md for the data dictionary and methods. The honest limitations of the underlying pixel RD surface (interpolative, not transferable out-of-region; spatial-block cross-validated R2 about 0.18) still apply and motivate the aggregate/SAE framing. Companion to the One Earth dual-role manuscript and the RS-RD method and red-team deposit (v1.0 to v1.2).