Versioned soil organic carbon maps for climate-resilient and sustainable cities: A ridge-penalised linear mixed-effects ensemble of MIR-based estimates

Accurate and transferable estimation of soil organic carbon (SOC) concentration supports measurement, reporting and verification (MRV) and city-region climate strategies under Sustainable Development Goals (SDG) 11 and 13. However, large-scale operational deployment of spectroscopic SOC mapping is often constrained by limited access to, and poor harmonisation of, raw spectral datasets across instruments and regions. We develop a secondary-estimates stacked-ensemble approach that does not require harmonised raw spectra. A ridge-penalised linear mixed-effects model (LMM) stacker combines five routinely generated mid-infrared (MIR) SOC predictions (one calibrated on Walkley–Black SOC and four on dry-combustion SOC) with simple location-type descriptors, while modelling farm-level random effects. Using 15,346 samples from 431 farms across six Australian states, we tune hyperparameters via farm-stratified nested cross-validation and evaluate on held-out test sets. This model consistently outperforms five baselines across states (maximum R 2 improvement of 0.14 and a 29% reduction in RMSE) and achieves national R 2 = 0.874 with RMSE = 0.388%. We then generate versioned, auditable, depth-resolved SOC maps that can be aggregated to jurisdictional and city-region planning units. By relying on shareable secondary estimates rather than harmonised raw spectra, the workflow lowers data-access and harmonisation barriers and delivers MRV-ready SOC maps to support climate-resilient and sustainable cities within integrated SDG 11/SDG 13 strategies. • Provides shareable SOC estimation without requiring harmonised raw spectra. • First employs a ridge-penalised mixed-effects model captures farm-level clustering. • Model trained on 15,346 samples from 431 farms across six Australian states. • Maximum state-level R 2 improvement of 0.14; national R 2 = 0.874 and RMSE = 0.388%. • Depth-resolved SOC maps (0–30, 30–60 cm) support MRV and city-region planning.