Nutrient cycling via livestock excreta is an important grassland ecosystem service. This study was conducted in 2016 and 2017 and determined nutrient excretion from cattle grazing three systems: (a) N‐fertilized bahiagrass (Paspalum notatum Flüggé) overseeded with rye (Secale cereale L.) and oat (Avena sativa L.) during winter (Grass+N), (b) unfertilized bahiagrass overseeded with rye–oat–clover (Trifolium sp.) mixture (Grass+clover), and (c) rhizoma peanut (Arachis glabrata Benth.)–bahiagrass mixture overseeded with a rye–oat–clover mixture (Grass+CL+RP). Urinary volume excreted in the warm season ranged from 122 to 182 L ha⁻¹d⁻¹, more than double that in the cool season (56–70 L ha⁻¹ d⁻¹). Urinary N concentration during the warm season was greater in Grass+CL+RP compared with Grass+N (4.4 vs. 3.1 g kg⁻¹, respectively). In addition, fecal dry matter and organic matter output was greater for Grass+N than for steers grazing Grass+CL+RP during the warm season (3.8 and 3.2 vs. 2.8 and 2.2 kg hd⁻¹ d⁻¹, respectively). Total annual feces excretion concentrations of P, K, Mg, and N were greater in the Grass+N than in Grass+CL+RP system. The introduction of legumes increased the proportion of N returning via urine. However, when summed across seasons, total N excretion (feces and urine) was greater in Grass+N than in Grass+CL+RP (89 vs. 71 kg ha⁻¹ yr⁻¹), mainly due to the greater stocking rate in the former. Grass–legume systems receiving 34 kg N ha⁻¹ yr⁻¹ recycled 80% of the N recycled in the grass system receiving 224 kg N ha⁻¹ yr⁻¹, indicating the potential of forage legumes to add N to grasslands.

Accurate spatial quantification of soil organic carbon (SOC) stocks and their associated uncertainties is vital for climate mitigation and sustainable grazing management. However, regional SOC mapping remains challenging due to complex soil–landscape interactions, dynamic environmental processes, and sparse and non-representative field observations. This study advances regional SOC mapping by developing a parsimonious, uncertainty-aware digital soil mapping framework to estimate topsoil (0–20 cm) SOC stocks across Florida’s grazing lands at 30 m resolution. Grounded in the STEP–AWBH soil-forming framework, the approach explicitly operationalizes time-invariant and dynamic controls through spatiotemporal feature construction. SOC stocks were modeled using Quantile Regression Forests to generate point estimates and 90% prediction intervals, validated via spatial cross-validation. While a baseline model using contemporary data achieved modest results (R2 = 0.38; RMSE = 9.52 t ha-1), cross-temporal spiking with legacy samples significantly enhanced predictive accuracy and uncertainty calibration (R2 = 0.57; RMSE = 7.93 t ha-1). In contrast, unconstrained spiking across heterogeneous land uses degraded model performance, underscoring the importance of domain-specific data integration. SHAP-based interpretation revealed that SOC variability reflects the influence of pedological properties (e.g., available water capacity), grazing intensity, and temporal signatures—persistence, variability, and long-term trends—of hydroclimate, soil moisture, and vegetation phenology. Total topsoil SOC storage for Florida’s grazing lands was estimated at 36.11 Tg, with a mean stock of 20.3 t ha-1. The resulting high-resolution SOC stock and uncertainty maps are hosted on an open-access interactive GIS platform, providing a robust baseline for monitoring and managing soil carbon in subtropical grazing systems.