Permafrost degradation profoundly affects carbon storage in alpine ecosystems, and the response characteristics of carbon sequestration are likely to differ at the different stages of permafrost degradation. Furthermore, the sensitivity of different stages of permafrost degradation to climate change is likely to vary. However, related research is lacking so far on the Qinghai-Tibetan Plateau (QTP). To investigate these issues, the Shule River headwaters on the northeastern margin of the QTP was selected. We applied InVEST and Noah-MP land surface models in combination with remote sensing and field survey data to reveal the dynamics of different carbon (vegetation carbon, soil organic carbon (SOC), and ecosystem carbon) pools from 2001 to 2020. A space-for-time analysis was used to explore the response characteristics of carbon sequestration along a gradient of permafrost degradation, ranging from lightly degraded permafrost (H-SP) to severely degraded permafrost (U-EUP), and to analyze the sensitivity of the permafrost degradation gradient to climate change. Our results showed that: (1) the sensitivity of mean annual ground temperature (MAGT) to climatic variables in the U-EUP was stronger than that in the H-SP and S-TP, respectively; (2) rising MAGT led to permafrost degradation, but increasing annual precipitation promoted permafrost conservation; (3) vegetation carbon, SOC, and ecosystem carbon had similar spatial distribution patterns, with their storage decreasing from the mountain area to the valley; (4) alpine ecosystems acted as carbon sinks with the rate of 0.34 Mg ‧ha  1 ‧a  1 during 2001-2020, of which vegetation carbon and SOC accumulations accounted for 10.65 % and 89.35 %, respectively; and (5) the effects of permafrost degradation from H-SP to U-EUP on carbon density changed from promotion to inhibition.

The Qinghai-Tibetan Plateau (Q-TP hereafter) has experienced dramatic warming in recent decades, resulting in severe effects on the ecosystems and downstream. However, none of previous studies investigated elevationdependency temperature trend with the high resolution over the long-term period. Based on monthly temperature dataset with 0.1 developed by generative adversarial network, elevation-dependency temperature trend over the Q-TP and 5 climate zones (humid, humid-semihumid, semihumid, semiarid, arid region) during 1901-1946, 1946-1965, 1965-1997 and 1997-2015 are investigated. Snow cover (SNC), high cloud cover (HCC), middle cloud cover (MCC), specific humidity (SHUM) and soil moisture content (SOILM) are introduced to analyze possible mechanism. There are 4 cases of elevation-dependency temperature trend, which are positive/negative elevation-dependency warming (EDW+/EDW-) and cooling (EDC+/EDC-). These patterns (EDW+, EDW-, EDC+ and EDC-) are identified as warming/cooling trends that become stronger/weaker with increasing elevation. EDW- signal is found during 1901-1946 due to the influence of SOILM. The most prominent EDW- signal occurs over the arid region. EDC+/- is presented during 1946-1965 under the dominance of SHUM and SOILM. The stronger EDC signal is shown over the arid and semiarid region than over the humidsemihumid region. The subtle EDW+ signal is shown over the semihumid, semiarid and arid regions during 1965-1997 when SOILM has a relatively large contribution to temperature trend. The robust EDW+ signal is exhibited from 1997 to 2015 when SNC plays a vital role in regulating the temperature change. There is a more significant EDW+ over the humid, humid-semihumid, and semihumid regions than that over the semiarid and arid regions during this period. Above all, SNC, SHUM and SOILM are found to be the primary contributors to elevation-dependency temperature trend. SOILM and SHUM are associated with hydrological effects and control temperature variations over the Q-TP during 1901-1997. SNC is related to snow/ice-albedo feedback and dominates temperature variations over the Q-TP during 1997-2015.