中文摘要：

青藏高原沼泽化草甸是土壤有机碳密度最高,对气候变化最敏感的高寒生态系统。对其生态系统总初级生产力（GPP）持续准确的量化是掌握站点到全球尺度的碳循环的关键。涡度相关技术（EC）是测量生态系统碳通量的最佳途径,而遥感模型可以实现从生态系统迹点（footprints）到区域乃至全球的尺度扩展。但是,大多遥感估算模型的适用性在这种高寒沼泽化草甸上并没有得到验证。本研究选取了四个近年来被广泛运用的遥感估算GPP的模型,即MODIS算法（MOD）、植被光合模型（VPM）、光合能力模型（PCM）和高寒植被模型（AVM）对青藏高原中部的一个典型高寒草甸生态系统的GPP进行了估算。结果显示：所有遥感模型对GPP的年内季节变异都有很好的解释（R2〉0.89,P〈0.0001）,但很难解释其年际变化。与日均EC_GPP相比,VPM严重的低估了该生态系统的GPP,其估测值大约仅为EC观测值的54%。但是,其他三个模型可以较准确地进行GPP估算：相比之下,AVM可以反演94.5%的EC观测,相对于EC观测的均方根误差（RMSE）最小（1.47 g C m~（-2））;PCM对EC_GPP有微小的高估（约12.0%的EC观测值）,而MODR对EC_GPP有微弱的低估（约8.1%的EC观测值）,但二者的偏差都不显著。本研究表明AVM对该高寒沼泽化草甸的GPP估算比其他较复杂的GPP估算模型更有优势。

英文摘要：

Alpine swamp meadows on the Tibetan Plateau,with the highest soil organic carbon content across the globe,are extremely vulnerable to climate change.To accurately and continually quantify the gross primary production（GPP） is critical for understanding the dynamics of carbon cycles from site-scale to global scale.Eddy covariance technique（EC） provides the best approach to measure the site-specific carbon flux,while satellite-based models can estimate GPP from local,small scale sites to regional and global scales.However,the suitability of most satellite-based models for alpine swamp meadow is unknown.Here we tested the performance of four widely-used models,the MOD17 algorithm（MOD）,the vegetation photosynthesis model（VPM）,the photosynthetic capacity model（PCM）,and the alpine vegetation model（AVM）,in providing GPP estimations for a typical alpine swamp meadow as compared to the GPP estimations provided by EC-derived GPP.Our results indicated that all these models provided good descriptions of the intra-annual GPP patterns（R〉20.89,P〈0.0001）,but hardly agreed with the inter-annual GPP trends.VPM strongly underestimated the GPP of alpine swamp meadow,only accounting for 54.0% of GPP_EC.However,the other three satellite-based GPP models could serve as alternative tools for tower-based GPP observation.GPP estimated from AVM captured 94.5% of daily GPP_EC with the lowest average RMSE of 1.47 g C m~（-2）.PCM slightly overestimated GPP by 12.0% while MODR slightly underestimated by 8.1% GPP compared to the daily GPP_EC.Our results suggested that GPP estimations for this alpine swamp meadow using AVM were superior to GPP estimations using the other relatively complex models.