Abstract: To elucidate the multilevel drivers of stocking rate increase on plant community state transition in desert steppe， this study was performed in a long-term grazing manipulation platform （2004-present） in Stipa breviflora desert steppe of Inner Mongolia. Four grazing intensity treatments were established： no grazing （CK）， light grazing （LG，0.91 sheep units·hm^-2·half-year^-1）， moderate grazing （MG， 1.82 sheep units·hm^-2·half-year^-1）， and heavy grazing （HG， 2.71 sheep units·hm^-2·half-year^-1）. Based on data from 145 monitored quadrats in representative plots， a state transition probability matrix was constructed， integrating species interaction networks， functional group succession， and plant family closure analysis. The results showed that： （1） Plant population network complexity peaked during the stage of LG transition to MG （with only 4 isolated species）， whereas core species （e.g.， the disappearance of Teloxys aristata） reduced and isolated species sharply increased to 15 at the HG stage； （2） As stocking rate increased， the transition probabilities from shrub-subshrub （SS） and perennial forbs （PF） to perennial grasses （PG） rose， while the transition probability from PG to annual-biennial plants （AB） increased by more than 30% under HG； （3） The receiving capacity of the Poaceae （POA） strengthened with increasing stocking rate， whereas the biomass proportion of Asteraceae （AST） declined. In summary， a stocking rate exceeding 1.82 sheep units·hm^-2·half-year^-1 triggered the critical threshold for community degradation. In contrast， moderate grazing （LG-MG） maintained community stability through species redundancy， optimized resource allocation among functional groups， and the formation of closed-loop transitions between plant families. This study provides a multidimensional theoretical foundation for early warning of desert steppe degradation and targeted restoration.

Key words: Desert steppe, Stocking rates, State transition, Functional group succession, Multistable maintenance

摘要： 为阐明载畜率升高对荒漠草原植物群落状态转移的多层级驱动机制，本研究依托内蒙古短花针茅荒漠草原长期放牧控制平台（2004—今），设置无牧（CK）、轻度放牧（LG，0.91羊单位·hm^-2·半年^-1）、中度放牧（MG，1.82羊单位·hm^-2·半年^-1）和重度放牧（HG，2.71羊单位·hm^-2·半年^-1）4个载畜率梯度，基于代表性样地145个监测样方数据构建状态转移概率矩阵，整合物种互作网络、功能群演替及植物科闭合性分析。结果表明：（1）LG→MG阶段植物种群网络复杂性达峰值（孤立物种仅4种），而HG阶段核心物种减少（如Teloxys aristata消失）、孤立物种激增至15种；（2）随载畜率升高，灌木半灌木（SS）和多年生杂类草（PF）向多年生禾草（PG）转移概率增加，而PG→一二年生植物（AB）转移概率在HG阶段增幅超30%；（3）禾本科（POA）接收能力随载畜率增大而增强，但菊科（AST）生物量占比下降。总之，载畜率>1.82羊单位·hm^-2·半年^-1触发群落退化临界点，而适度放牧（LG~MG）通过物种冗余、功能群资源优化及植物科间转移闭环维持群落稳定性。本研究为荒漠草原退化预警和精准恢复提供了多维度理论依据。

关键词: 荒漠草原, 载畜率, 状态转移, 功能群演替, 多稳态维持