The Tibetan Plateau holds the world’s largest alpine permafrost and is undergoing  an acceleration of warming. Phenological shifts over alpine permafrost in a warmer world have been little studied and are greatly underrepresented in current syntheses. Here, we conducted seasonal and gradient temperature-controlled experiments in a permafrost-affected meadow to evaluate how warming drives shifts in spring and autumn phenology, and associated growing-season length at both community and species levels. Our results showed that there is no sign of slowdown in spring advance with warming under a higher year-around warming treatment, aligning with a future medium warming scenario. This finding can be attributed to the possibility that winter warming is insufficient to reduce chilling accumulation, which would not delay spring phenology and then lead to a non-slowdown in spring phenological advancement. Although spring advance led to an advance in autumn senescence according to spring-only warming experiments, the advance could not offset the delay due to concurrent warming. As a result, year-around warming significantly delayed autumn senescence, although there was a deceleration in delay with warming under high temperature treatment than under the low one. Taken together, there is no slowdown in an extension of growing season length with warming under a higher year-around warming treatment, with an increase of length by 9 and 21 days at the end of this century under a CO₂ stabilization and medium warming scenarios, respectively. Our results suggest that a continued growing season extension at least under the medium warming scenario would help permafrost-affected meadow ecosystems to mitigate permafrost carbon release on the Tibetan Plateau.

Experimental design

The manipulative warming experimental platform in typical permafrost-affected region was established in April 2018. Hexagonal open-top warming chambers (OTCs), which were used extensively in the International Tundra Experiment (Wahren et al., 2013), were adopted to simulate the effect of warming on ecosystems. For the low warming gradient, the OTCs are 1.3 m in diameter at the base, 1.1 m in diameter at the top, and 0.4 m high. For the high warming gradient, the OTCs are 1.3 m in diameter at the base, 0.8 m in diameter at the top, and 0.8 m high  (Figure S2). All chambers use the same light panels with 8mm thick and the light transmittance rate > 95%.

There were three warming treatments in our experiment: spring warming (from 10th February to 10th May), winter warming (from 10th November to 10th February) and year-round warming (throughout the whole year), each warming treatment set two warming gradient (40cm and 80cm height OTC, respectively) and four replicates. In total, 28 plots (4 replicates × 2 levels × 3 treatments + 4 control replicates) were set up to explore the impact of seasonal warming on the sequence of phenological events.

In addition, we also set up an active infrared warming experiment started in the year 2021 to explore the response of ecosystem dynamics to a gradient of warming. Specifically, it consists of the control and three warming treatments (constant warming of 1, 2 and 4℃ with respect to the control), each with four replicates. Here we used ceramic heating elements (1000W, 240V, Campbell Scientific, Logan, USA) to produce and emit infrared radiation that could transfer energy from the infrared source to plots. For each plot, three stainless steel posts were anchored about 30 cm into the ground and three crossbars were then attached to the posts at a height of 1.6 m above the ground. The ceramic heating elements were bolted to the crossbars along the 1.6 m side, and sixteen triangular plots were then set up. The dummy heaters with the same shape and size were set up in control plots to mimic any possible infrastructure impact of infrared warming systems. We separated adjacent plots using a buffer zone of around 5-10 m to avoid interferences, and used a Proportional-Integral-Derivative control system (Campbell Scientific, Logan, USA) to ensure an accurate control of pre-set warming magnitude over plots (Kimball et al., 2008). This active warming system was deployed on the same site, given that passive warming systems such as OTCs were often criticized for the following reasons. First, OTCs often reduced soil moisture by passively heating a small vegetation plot by capturing more solar radiation and intercepting more rainfall events, which would potentially bias phenological responses to warming. Second, OTCs might act as barriers to wind flow and pollinator movement (Dong et al., 2023), although some studies found no significant effect of OTC installation on insect flower visitors (Robinson et al., 2018). Third, in contrast to active warming system that has a uniform warming magnitude on ecosystems, the impact of OTC on the magnitude of warming varies seasonally, with a lower magnitude in summer and higher ones during winter and spring seasons on our permafrost-affected site.

Manual phenological observations

Phenological events were consecutively monitored for the three years (2018–2020) at both community and species levels with 2~3 times per week interval depend on weather conditions. For community phenology observation, we set up in situ and permanent quadrat (1×1 m²) in each chamber, then meshed the quadrat into 100 cross points to record the individual phenological status under each projection of the cross points. Here we defined two threshold (10% and 90%, respectively) to identify the green up and green down days of community (Chen et al., 2022; Li et al., 2016; Meng et al., 2017). For instance, the start of the growing season (SOS) in community level is defined as 10% individuals at least under the cross point turn green, while the end of growing season in community level is defined as more than 90% individuals under the cross point into complete senescence and highly dehydrated. The start and the end of reproductive season in community level are defined as the date of 10% individuals turn into first flowering (FF) and 90% individuals turn into first fruit set (FFS), respectively.

    For species level phenology observation, eight native perennial species which maximally represented the plant community assemble is recorded during the study period. These species accounted for 86.89% and 83.26% of the relative contributions to coverage and community biomass, based on their leaf out time, they were classified into three functional groups: early spring, mid spring and late spring species (Table 2). The intermediate date between first recorded date and the date of the preceding survey when the event was not recorded as the start date, and vice versa for determining the end date. For example, if the initial observation of leaf-out occurred on 26 April, and the preceding survey on 22 April did not record this event, then the estimated leaf-out date would be 24 April, which is the midpoint between the two dates (Li et al., 2016). The duration of each phenological event was the number of days between the start and ending date. Here the growing season was defined as the period from leaf onset to complete leaf senescence, while the reproductive season was defined as the interval from the first flowering to first fruit set.

Based on previous study (Li et al., 2016; Meng et al., 2017) and well-known BBCH (Biologische Bundesanstalt, Bundessortenamt und CHemische Industrie) guidance (Meier et al., 2009), we selected 7 sequential and hierarchical phenological events in monitoring (Table 1). Because of low growing season temperature and rare pollinators, the flowering rate of native species is relatively low, most species also reproduce clonally (Lagotis glauca Gaertn for instance), in order to ensure the quality of phenological monitoring, more than 10 individuals for each species were marked in previous year. Once flower buds were noticed, the individual will be monitored as follows, if the individual fails to enter the reproductive phase, we choose another individual in monitoring, the data was averaged of all the selected individuals for each species we choose.

In addition, air temperature near soil surface, soil moisture and temperature at 10 cm depth were recorded at an interval of 1 min using Decagon 5TM sensors (Decagon Devices, Pullman, Washington, USA). The logger auto calculated and saved the data on 30-minute averages across the whole year, and three replicates were designed for each treatment in this study. Mean annual solar radiation data are derived from the WorldClim Version 2.0 dataset (www.worldclim.org).

For each phenological event at both community and species levels, its temperature sensitivity is calculated as the difference of phenological date to that of air temperatures between control and treatment plots (Δdays ℃^-1). In addition, for a given phenological stage, we also computed phenological niche breadth as the length between the earliest species to that of the latest species in community.