The paper discusses the connections between two-stage stochastic optimization and robust statistical estimation. Main question related to statistical predictions is how to use the predictions to optimize the overall decisions and how current decisions can affect predictions. In general problems of decision-making, feasible solutions, concepts of optimality and robustness are characterized from the context of decision-making situations, i.e., systems structure, goals, security constraints, safety norms, supply-demand relationships, thresholds. Robust statistical approaches can be effectively combined with disciplinary or interdisciplinary models, e.g., land use model GLOBIOM, for effective decision-making in the conditions of uncertainty, increasing interdependencies and systemic risks. We discuss a quantile- regression EPIC meta-model for tracking dynamics and uncertainties of Soil Organic Carbon (SOC), which is an important agri-environmental indicator. SOC levels (quantiles) can be controlled with GLOBIOM, to analyze the costs of achieving robust land management practices to sequester SOC and fulfill food-energy-water-environmental nexus security goals. Quantiles identify critical SOC levels signaling how close is a threshold or a targeted level. The SOC-EPIC meta-model is developed using historical observations and results of the bio-physical model EPIC. It enables the analysis of SOC content and respective probabilities as a function of exogenous parameters such as monthly temperature and precipitation and endogenous, decision-dependent parameters, which can be altered by the land management decisions computed with GLOBIOM.