Assessing depth-resolved tropical soil organic carbon with ERA5-land and SoilGrids-conditioned dielectric inversion

Reliable soil organic carbon (SOC) estimates below 0.3 m are difficult to obtain in tropical regions because deep sampling is sparse and field campaigns are expensive. This study tests whether public model products can provide useful depth-specific SOC information at Central and East African TropSOC sites. We combine ERA5-Land soil moisture and temperature with SoilGrids texture and organic matter priors in a dielectric SOC inversion (DSOC) workflow. The workflow converts ERA5-Land moisture to a mineral-soil dielectric proxy, compares that proxy with an organic matter-aware dielectric model, and estimates SOC at four sampled depth intervals. DSOC was evaluated against 304 independent TropSOC observations from the Democratic Republic of the Congo, Rwanda, and Uganda at 0-0.1, 0.3-0.4, 0.6-0.7, and 0.9-1.0 m. Because ERA5-Land represents soil moisture below 0.28 m as a single 0.28-1.0 m layer, the three subsoil validation depths (0.3-0.4, 0.6-0.7, and 0.9-1.0 m) share one modeled moisture input; "depth-resolved" here therefore refers to validation against separately sampled horizons rather than to an independent moisture layer at each depth. SoilGrids was used as a map prior and calibration target, not as ground truth; laboratory TropSOC SOC was used only for evaluation. The overall gain relative to depth-matched SoilGrids was small after map-based calibration (r = 0.733 vs. 0.727; root mean square error = 14.5 vs. 15.2 g kg-1). The clearest and most stable support occurred at 0.3-0.4 m (Δr = + 0.15); the positive point estimates at 0.6-0.7 and 0.9-1.0 m were less certain, and there was no surface correlation gain. Residual tests showed that raw DSOC retained information associated with observed SOC at 0.3-0.4 and 0.6-0.7 m after the SoilGrids component was removed. Raw DSOC remained strongly correlated with the SoilGrids prior overall (r = 0.91) and carried a large positive bias before calibration, so its useful signal lies in how it reorders samples rather than in its absolute values. These results support DSOC as a cautious, SoilGrids-conditioned refinement for subsoil-focused SOC assessment-not an independent, observation-only SOC retrieval.