Land models typically represent the entire fine-root system as a single homogeneous pool, collapsing structural and functional complexity into one set of parameters. This simplification weakens the belowground feedbacks that balance source-driven carbon dynamics. Here, we integrate the TAM (Transport and Absorptive fine roots with Mycorrhizal fungi) framework into the E3SM Land Model (ELM), replacing the single fine-root pool with three pools — transport roots, absorptive roots, and mycorrhizal fungi — defined by their roles in resource conduction, nutrient acquisition, and symbiotic exchange, each governed by pool-specific C<span> </span>:<span> </span>N ratios, longevity, allocation coefficients, and litter chemistry (21 new parameters). Sobol sensitivity analysis at 13 FLUXNET sites spanning 13 natural PFTs in ELM reveals a two-strategy parameterization dichotomy: deciduous and grass plant functional types (PFTs) concentrate variance in 2–3 allocation parameters, while evergreen PFTs distribute sensitivity across 5–6 parameters spanning chemistry, longevity, and allocation. Surrogate-assisted Bayesian calibration produces emergent fine-root properties that diverge from ELM defaults, with system-level nitrogen demand increasing at 10 of 14 PFTs (up to 4.3<span> </span>×) and absorptive-root longevity spanning 0.5–3.5 yr. A three-tier validation framework—monthly evaluation, temporal out-of-sample, and spatial out-of-sample—demonstrates that these changes yield transferable improvements: improved seasonal amplitude fidelity (mean GPP amplitude ratio 0.99 vs. 1.38 for the baseline), near-elimination of ER bias at a deciduous site (MBE from +<span> </span>515 to +<span> </span>33 gC m⁻² yr⁻¹), and cross-continental parameter transfer for C4 grasses (81<span> </span>% RMSE reduction). Degradation at evergreen sites where the baseline overestimates productivity identifies a scope boundary where a belowground-only improvement cannot remedy upstream source-side bias. The function-oriented design provides natural coupling points for future extensions, including mechanistic root–water interactions and depth-resolved root profiles. ELM-TAM demonstrates that resolving belowground structural heterogeneity strengthens sink-based controls on the carbon cycle without overparameterization, establishing a foundation for these extensions and for global benchmarking in Earth System land models.