Abstract：
Grassland degradation threatens ecosystem function and livestock production, partly induced by soil nutrient deficiency due to the lack of nutrient return to soils, which is largely ascribed to the intense grazing activities. Therefore, nitrogen (N) fertilization has been widely adopted to restore degraded Qinghai-Tibetan Plateau (QTP) grasslands. Despite numerous field manipulation studies investigating its effects on alpine grasslands, the patterns and thresholds of plant response to N fertilization remain unclear, thus hindering the prediction of its influences on the regional scale. Here, we established a random forest model to predict N fertilization effects on plant productivity based on a meta-analysis synthesizing 88 publications in QTP grasslands. Our results showed that N fertilization increased the aboveground biomass (AGB) by 46.51 %, varying wildly among plant functional groups. The positive fertilization effects intensified when the N fertilization rate increased to 272 kg ha-1 yr-1, and decreased after three years of continuous fertilization. These effects were more substantial when applying ammonium nitrate compared to urea. Further, a machine learning model was used to predict plant productivity response to N fertilization. The total explained variance and mean squared residuals ranged from 49.41 to 75.13 % and 0.011-0.058, respectively, both being the highest for grasses. The crucial predictors were identified as climatic and geographic factors, background AGB without N fertilization, and fertilization methods (i.e., rate, form, and duration). These predictors with easy access contributed 62.47 % of the prediction power of grasses\' response, thus enhancing the generalizability and replicability of our model. Notably, if 30 % of yak dung is returned to soils on the QTP, the grassland productivity and plant carbon pool are predicted to increase by 5.90-6.51 % and 9.35-10.31 g C m-2 yr -1, respectively. Overall, the predictions of this study based on literature synthesis enhance our understanding of plant responses to N fertilization in QTP grasslands, thereby providing helpful information for grassland management policies. Conflict of interest: The authors declare no conflict of interest.
摘要：
草地退化威胁着生态系统功能和畜牧业生产,部分是由于土壤养分缺乏导致土壤养分缺乏，这在很大程度上归因于激烈的放牧活动。因此,氮（N）施肥已被广泛用于恢复退化的青藏高原（QTP）草地。尽管进行了许多现场操作研究，以调查其对高山草原的影响，植物对氮肥的反应模式和阈值尚不清楚，从而阻碍了其对区域尺度影响的预测。这里,我们建立了一个随机森林模型来预测氮肥对植物生产力的影响，基于荟萃分析，综合了QTP草原上的88篇出版物。我们的结果表明，氮肥使地上生物量（AGB）增加了46.51％，在植物官能团之间变化很大。当施氮率增加到272kgha-1yr-1时，正施肥效应增强，连续施肥三年后下降。与尿素相比,当施用硝酸铵时,这些效果更显著。Further,机器学习模型用于预测植物生产力对氮肥的响应。总解释方差和均方残差范围分别为49.41~75.13%和0.011~0.058,两者都是最高的草。关键的预测因素被确定为气候和地理因素，背景AGB没有氮肥，和施肥方法(即，rate,形式,和持续时间)。这些易于访问的预测因子贡献了62.47%的草响应预测能力，从而增强了我们模型的泛化性和可复制性。值得注意的是,如果30%的牦牛粪便被送回QTP的土壤中，预计草地生产力和植物碳库分别增加5.90-6.51%和9.35-10.31gCm-2yr-1。总的来说,本研究基于文献综合的预测增强了我们对QTP草地植物对氮肥的反应的理解，从而为草地管理政策提供有用的信息。利益冲突：作者声明没有利益冲突。
