Multi-objective optimization of “human-grassland-wild ungulate-livestock” system in a typical nature reserve on the Qinghai-Tibetan plateau

Nature reserves on the Qinghai-Tibetan Plateau (QTP) play crucial roles in supporting ecological security. Achieving the balance of “human-grassland-wild ungulate-livestock” (HGWL) system is important for sustainable development in these nature reserves. This study developed an indicator-based decision-support framework based on machine learning (ML) methods and the non-dominated sorting genetic algorithm II (NSGA-II) in the Changtang Nature Reserve on the QTP. An extreme gradient boosting model was employed to quantify the imbalance between aboveground biomass (AGB) and crude protein (CP) required by livestock and wild ungulates and that supplied by grasslands. A multi-objective optimization model of HGWL system aiming to increase the ecological and economic benefits and decrease the grassland pressure was constructed using NSGA-II based on the imbalance of AGB and CP. This model was applied to analyze the optimized cultivated grasslands planting structure, livestock structure and other variables in three scenarios. Ecological benefits decreased by 0.50% in economic development priority scenario, and economic benefits decreased by 16.22% in ecological protection priority scenario. By contrast, ecological benefits and economic benefits increased, and grassland pressure decreased by 5.06%, 5.05%, and 0.74% respectively in ecological economic balance scenario, which was identified as the optimal scenario. More areas of alfalfa, higher sheep slaughter volume and lower goat and yak slaughter volume were recommended to achieve the HGWL balance. The combination of ML and NSGA-II can help achieve the multi-objective optimization of HGWL system, and this framework has potential to be promoted to other nature reserves around the world.