Composite Site Productivity Index (CSPI) for the conterminous United States: 30 m soil and terrain and 1 km combined surfaces

Modeled forest site index (meters) across the conterminous United States from freely available soil (SoilGrids 2.0), terrain (SRTM), canopy (Hansen Global Forest Change), and (for v4) climate (ClimateNA 1991-2020 normals) covariates, using random forests trained on USDA FIA condition level site index (SICOND).

Surfaces included: CSPI v3 at 30 m (soil + terrain + canopy; plot blocked cross validated R-squared 0.602; values 5.98 to 55.83 m); CSPI v3 at 1 km (companion preview); CSPI v3 forest masked at 1 km (Hansen tree cover 2000 >= 10%); CSPI v4 at 1 km (combined soil, terrain, canopy, climate; plot blocked CV R-squared 0.660); CSPI v4 forest masked at 1 km; CSPI v4 prediction quantiles at 1 km (p50, p05, p95, 90% interval width).

Key findings: (1) soil and terrain at 30 m match climate at 1 km in SICOND prediction; the two families are complementary. (2) FIA coordinate fuzzing destroys high resolution canopy bands (r 0.13 to 0.24 between true and fuzzed extractions); predictors at 250 m or coarser are robust (r > 0.85). (3) Adding the surface to the FVS diameter growth (Kuehne) residual gives no improvement because that model already includes ln(SI) as DGSITE; CSPI's contribution is to supply the SI input wherever measured dominant height and age are unavailable.

See README.md for full methods, data dictionary, citation, and load examples.