ability to predict how plants and their communities respond to climate change. We aim to help fill this gap through metaanalysis of plant trait and community property responses to in situ warming.

We focus on plant traits and plant community properties because these are important mediators for plant interactions with their environment (Diaz et al. 2004;Violle et al. 2007) and they can vary across multiple contexts (Gratani 2014). Plant traits are defined as any morphological, physiological, or phenological (morpho-physio-phenological) feature measurable at the individual level, whereas community properties are defined as emergent features measurable at the community or ecosystem level (e.g., community biomass, percent cover, etc.) (Violle et al. 2007). Though there is debate about the importance and usage of traits in ecology (van der Plas et al. 2020), they can provide a meaningful, taxon-independent view of overall community structure and function (Funk et al. 2017;Hagan et al. 2023).

While plant traits and community properties may be useful for predicting overall responses to climate change, traits themselves can also change as a result of climate stressors. For instance, SLA has been seen to increase in response to warming (Descombes et al. 2020). The ability of a single genotype to produce a range of phenotypes as a function of the environment is known as phenotypic plasticity (Bradshaw 1965;Nicotra et al. 2010). Plasticity in plant traits can assist plants in responding to changes in the environment, such as temperature increases associated with climate change (Gratani 2014;Matesanz et al. 2010;Nicotra et al. 2010). Due to the potential for plastic trait responses to abiotic stressors, we would ideally first determine the extent of trait plasticity before we can begin to use mean trait values to potentially predict overall community responses to climate change stressors. Evaluating a variety of environmental, experimental, and plant-level contexts can help quantify the potential causes for inter-and intraspecific trait variation and provide insights into the potential for plastic responses to climate change. For example, Stotz et al. (2021) determined that phenotypic plasticity for five trait types (leaf morphology, physiology, plant allocation, size, and performance) can be associated with both climate (e.g., mean annual temperature) and non-climate (e.g., nutrient availability) related contexts in the environment. However, while their study looked at overall plant plasticity associations with environmental contexts, they did not consider how climate warming may alter these plastic trait associations. For instance, though traits may be affected by temperature differences across a species' range, how might novel warming stress affect species' traits within their current environment? Remaining questions such as this demonstrate that variation in plant and community responses to warming has not yet been well defined between environmental, experimental, and plant-level contexts. Furthermore, while previous studies have investigated in-depth both the effect of experimental warming on numerous plant traits in specific regions (Bjorkman et al. 2018;Elmendorf et al. 2012), and how global contexts contribute to specific plant responses to warming (Lin et al. 2010;Stuble et al. 2021), to our knowledge no study has combined these avenues of research to investigate multiple plant responses to experimental warming for plants across the globe.

Our focus on in situ field experiments enables the inclusion of multiple environmental contexts such as latitude, mean annual precipitation and temperature, and species distance from their range edge-all of which can affect plant traits and community properties (Nicotra et al. 2010). By focusing solely on passive warming experiments using open-top chambers (Marion et al. 1997), we limit methodological differences between studies, which can lead to large differences in experimental outcomes (Wolkovich et al. 2012). With these studies, we performed a meta-analysis investigating how 9 different environmental (e.g., latitude), experimental (e.g., length of warming study), and plant-level (e.g., plant functional type) contexts contribute to variation in plant responses to warming. We address two main questions: 1. How does in situ experimental warming affect various plant traits and community properties, and 2. How do different environmental, experimental, and plant-level contexts contribute to variation in plant trait and plant community responses to in situ warming? Our core hypotheses regarding these main questions are as follows:

1. Trait and property responses: a. Phenology: warming will lead to earlier spring phenophases, later fall phenophases, and longer flower lifespans as warming shifts the timing of cues that trigger these phases (Parmesan and Yohe 2003;Peñuelas and Filella 2001;Zhou et al. 2022). b. Growth: warming will lead to increases in overall plant growth (e.g., increases in overall percent cover, above and belowground biomass, leaf growth, and plant growth), likely due to longer growing seasons and increased enzyme activity (Elmendorf et al. 2012;Lin et al. 2010;Villén-Peréz et al. 2020;Wangchuk et al. 2021). c. Reproductive traits: warming will lead to increases in reproductive trait outputs (e.g., number of flowers, number of fruits, and fruit weight), likely due to longer growing seasons and increased growth rates (Fernández-Pascual et al. 2019). d. Chemical traits: warming will lead to decreases in above and belowground N content, likely due to the dilution effect (An et al. 2005;Jarrell and Beverly 1981;Yang et al. 2011).

2. Contexts: a. Environmental: plants further from the equator and individuals closer to their species' northern range edge will demonstrate stronger responses to warming, likely due to increased climate variability at high latitudes (Anderson and Song 2020;Ghalambor et al. 2006;Janzen 1967) and individuals at their northern margin being more responsive to climate warming (Buizer et al. 2012;Kilkenny and Galloway 2013). b. Experimental: year-round, warmer, and shorter duration warming experiments will lead to stronger responses to warming, likely due to the presence of winter and early spring warming when warmed year-round (Kreyling et al. 2019), hotter temperatures leading to stronger changes (Wang et al. 2019), and strong initial plastic responses in the short-term which level out over time (Franks et al. 2014;Kristensen et al. 2020). c. Plant-level: non-vascular species, such as bryophytes and lichens, will show more negative responses to warming than vascular species, potentially due to increased litter production of vascular plants (Elmendorf et al. 2012), and non-native species will show more positive responses to warming than native species as discussed in the vacant niche and priority effect hypotheses (Zettlemoyer et al. 2019).

In total, we identified 126 in situ passive open top chamber studies around the world (Figure 1, Table S1). ). This search returned 24,516 peer-reviewed papers published between 1971 and 2020 (Figure S1). 13,660 papers were successfully downloaded after removing corrupted or otherwise inaccessible papers. Text mining, via the R package "tm" (Feinerer and Hornik 2023), was used to search through each of the given papers for any combination of the terms "open-top," "itex", "(passive) warming," and/or "chamber", resulting in 2990 papers.

The Scopus search was repeated on 5 August 2022 to capture any papers published between 2020 and 2022 that may contain relevant data, resulting in 8955 new papers meeting the search criteria (Figure S1). The original 2990 and additional 8955 papers were manually screened to check for relevance using the following criteria: (1) the article utilized an open-top chamber design, (2) a field experiment was performed using the open-top chamber, and (3) the experiment was used to determine relevant plant response(s) to elevated temperatures. The original search returned 177 relevant papers and the reconducted 2022 search returned 54 relevant papers for a total of 231 papers.

For data extraction, we chose a subset of relevant plant responses that were most commonly measured in experimental warming studies. These plant response variables included: above-and belowground biomass, above-and belowground nitrogen content, spring phenology, fall phenology, flower lifespan, flower number, fruit number, fruit weight, percent cover, plant growth, and leaf growth. To determine the most commonly measured traits and properties, we recorded each plant response that was included as a measured response in the original 177 relevant warming studies. The total number of studies a specific response was measured in ranged from 1 to 49. We determined that a trait or property response present in at least 13 studies was common and included only these common plant responses in our data collection. We did not collect data on responses that were rarely measured (e.g., water use efficiency, which was measured by one study) due to low replication across studies. We also ensured that the selected responses were measured in experiments in varying regions across the globe to reduce geographic biases in which responses were collected where (Figure S2).

The 231 relevant papers were narrowed down to 126 papers that met the plant response criteria and contained data we were able to collect (Figure 1, Figure S1, Table S1). There were several reasons why a paper may not have contained usable data: not listing plant measurements in both warmed and control treatments, not including variation, not including sample size, or not presenting the data in a format in which it was accessible. If we were unable to collect data from the publication, we contacted the lead author of the study to solicit the needed data; 14 of 25 contacted authors responded to our inquiry and provided the requested data.

Free-air concentration enrichment (FACE) experiments were not considered because differences between this method and open-top chambers have been extensively documented (Hendrey and Kimball 1994;Macháčová 2010). We also did not include papers utilizing any active warming method, such as heating cables, due to differing experimental designs potentially leading to different experimental outcomes (Wolkovich et al. 2012). Reviews and meta-analyses were excluded due to potential overlap with the primary articles being reviewed in this study. The third criterion for inclusion, which stated that the experiment was used to determine plant response(s) to elevated temperatures, excluded articles utilizing open-top chambers that monitored processes such as CO 2 flux. If a study included a fully factorial design with warming crossed with a secondary stressor (CO 2 increase, herbivory, etc.), we only collected data from the warmed-only and control treatments. The majority of the included studies were in treeless ecosystems; only 9 studies contained a treatment for closed canopy species responses, in which shading may lead to differential vegetation responses (Table S1).

For each of the 126 studies, we collected metadata on environmental, experimental, and plant-level contexts. We were not able to obtain data on each of the data types listed below for all studies, so these data were collected when available.

For each study, we collected data on the following environmental contexts: latitude and longitude, elevation, mean annual precipitation and temperature, and the distance of each species to its northern range edge. Most studies published the latitude and longitude within the manuscript, but if only a general region was given (e.g., the name of a research station), we pulled the coordinates for the centroid of the given region. Absolute latitude was calculated as the absolute value of latitude in order to obtain total degrees away from the equator. Elevation was calculated for each experiment's coordinates using the elevatr R package (Hollister, Elphinstone, et al. 2023;Hollister, Robitaille, et al. 2023). WorldClim mean annual temperature and precipitation was extracted for each coordinate using the geodata and terra R packages (Hijmans 2023;Hijmans et al. 2023). We quantified each species' distance from its northern range edge to determine if individuals further from their 'leading edge' had differential responses to warming compared to individuals closer to the edge (Angert et al. 2011). To determine the distance to the northern range edge, we used the GBIF and BIEN R packages (Chamberlain et al. 2024;Maitner et al. 2018) to collect all occurrence records for each species included in our dataset as of June 2023. We used both GBIF and BIEN to ensure a comprehensive search of species' occurrence records. We then determined the maximum recorded latitude for each species, either from GBIF or BIEN, and subtracted the study latitude for each species from its maximum latitude to determine the distance (in degrees) to that species' northernmost occurrence, as a proxy for the species' northern range edge. The range edge analyses were only performed for species from experiments in the Northern Hemisphere, as species located in the southern hemisphere may track the climate southward, rather than northward (Angert et al. 2011).

For experimental contexts, we collected: publication number (1-126), publication information (primary author, year published, and journal), start year of the warming experiment, error type recorded (standard error, standard deviation, etc.), timing of warming (year-round or seasonal), site information (if one study contained multiple sites), average amount warmed by the chambers (°C), how average chamber warming was determined (growing season or annual average), and number of years warmed in the study.

For plant-level contexts, we collected: plant traits and community properties measured (Table S2), plant family, genus and species, plant functional group (Table S3), plant native status in the region of the study (native or non-native), and tissue type measured for aboveground biomass and N content (leaf, shoot, etc.). To standardize species names, we used Plants of the World Online (POWO 2023) to ensure all species are listed under their currently accepted name. The native status of a species was determined using the species distribution maps on POWO, in which a species was marked as "native" for a study if the experiment resided within the noted native range. Any species outside of the native range on the distribution map was designated as "non-native". Similarly to Stuble et al. (2021), we grouped some response variables into broader categories of similar plant function (Table S2). For example, all early phenophases (e.g., bud break, flowering, etc.) were grouped under "spring phenophases" due to their association with a similar plant function (in this case, the beginning of a life cycle phase). The response variables measured in this study could be categorized as either a plant trait or a community property, as discussed, but some variables could be both a trait and a property. For example, aboveground biomass may have been measured for a single individual of a species, making it a trait, or it may have been summed across multiple individuals/species in a plot, making it a community property. We parse out these differences when we test for differences between plant functional groups, in which we include "total community" as a grouping.

To generate effect sizes for each study's response variables, we extracted the mean, variation, and sample size of a given response variable in warmed and control treatments. Data were primarily collected from tables and figures, with some data coming from openly available data or the publication text including figure and table captions. Figures were imported into Fiji (Schindelin et al. 2012) for data extraction. To ensure data collection from figures was standardized amongst the three individuals extracting data, each individual independently extracted data from three of the same randomly selected figures to ensure their measurements were similar. After all measurements were completed, we visually compared each figure to its extracted data to ensure that the measurements were accurate for all figures.

All analyses were conducted using R (R Core Team, 2024). We first transformed all variation measurements recorded as standard error or 95% confidence intervals to standard deviation to ensure we could calculate effect sizes. Using the metafor R package (Viechtbauer 2010), we then calculated Hedges' g effect sizes to determine the effect of warming on each response variable from each study (Table S2). Hedges' g is a commonly used effect size in meta-analytic studies, as it standardizes the mean differences between groups while also accounting for uneven group sample sizes (Hedges 1981). Each study may have multiple effect sizes due to multiple traits and properties being measured across different contexts (e.g., multiple species, different levels of warming, etc.).

To test for potential publication bias, we generated funnel plots of the relationship between effect size and standard error, sampling variance, inverse standard error, and inverse sampling variance. In all plots, we found a standard funnel distribution, indicating an absence of strong publication bias (Figure S3).

All models were conducted using the rma.mv function from the metafor R package (Viechtbauer 2010). We ran an initial random-effects univariate model to test for the effect of warming on each response variable. This initial model contained species nested within site nested within publication number as random effect (Table S4). Because metafor does not allow for NAs in random effects, and not all measurements had species-level information, we substituted functional group for species when species-level designations were missing, in order to include all data in models.

We then ran separate random-effects univariate models for each response variable to test if the warming response depended on: absolute latitude, mean annual precipitation, mean annual temperature, species distance from range edge, timing of warming, amount warmed by the experiment, number of years warmed, plant functional group, and plant native status. These models also contained species nested within site nested within publication number as random effects (Tables S5-S16).

The effect of elevation was not analyzed as we found that latitude and elevation were negatively correlated (t 1254 = -27.4, p < 0.001; Figure S4). We also include Holm-corrected comparisons for contexts that contain multiple levels; while the univariate model for each variable tests if each level within a context differs from 0, the Holm-corrected comparisons compare each level to each other. For example, the univariate models for plant native status test if native and/or non-native species' effect sizes differ from 0, while the Holm-corrected comparison tests if native and non-native species' effect sizes are different from each other.

For models that tested the effects of a grouped variable (e.g., spring phenophases) or grouped plant functional type (e.g., shrubs) on the warming response, we also ran an initial model to ensure that the finer-scale levels within the grouped variable did not affect model outcomes. For example, we tested to see if deciduous and evergreen shrubs differed in their response to warming, and if not, we ran our model with the broader grouping (i.e., shrub). Similarly, for grouped variables, we tested to ensure that each finer-scale variable did not differ from the variables in its grouping. For example, we ensured that height and shoot length did not have opposite responses to warming (e.g., a positive and negative estimate) in order to group them under the broader "growth" response category. We tested for the effects of these finer-scale levels for all variables (Table S2) and functional groups (Table S3) that contained multiple levels. The models testing for differences between these finer-scale levels can be found in the supplement (Tables S17 and S18). Similarly, because aboveground biomass and N content contained multiple different tissue types that were measured across studies, we ensured that tissue-specific responses to warming were similar for both variables (Table S19).

For all models, we limited data to the last year of data collection for each study to avoid temporal pseudoreplication for experiments that followed the same plots over time (Stuble et al. 2021).

For two variables (fruit number and percent cover), we found that the models containing all years detected an overall response to warming, whereas our year-limited models did not (Table S20). This difference may be due to multi-year experiments containing some years with a warming response, but the effect was lessened when the dataset was reduced to the final year for those experiments.

We found positive, negative, and neutral plant responses to warming overall (Figure 2, Table S4). Aboveground N content decreased in response to warmer temperatures (Hedges' g = -0.41, z 13 = -4.88, p < 0.0001), and leaves showed a stronger change in N content compared to other measured tissue types (Table S19). There was also a marginal negative effect of warming on spring phenophases, meaning spring phenological events occurred earlier when warmed (Hedges' g = -0.12, z 13 = -1.76, p = 0.08). Aboveground biomass (Hedges' g = 0.25, z 13 = 3.64, p < 0.001), belowground biomass (Hedges' g = 0.60, z 13 = 3.73, p < 0.001), fruit weight (Hedges' g = 0.58, z 13 = 4.04, p < 0.0001), plant growth (Hedges' g = 0.65, z 13 = 10.0, p < 0.0001), and leaf growth (Hedges' g = 0.54, z 13 = 8.24, p < 0.0001) all increased in response to warming. For aboveground biomass, total plant biomass showed a stronger warming response compared to specific plant tissues (Table S19). All other traits and community properties did not demonstrate an overall effect of warming (Figure 2, Table S4).

When looking at environmental contexts that contributed to variation in responses to warming, we tested for effects of absolute latitude, mean annual precipitation, mean annual temperature, and species distance from its range edge. As latitude increased, the effect of warming on the number of fruits (β = 0.03, FIGURE 3 | The effect of absolute latitude (°) on Hedges' g effect size. Shown are the response variables that demonstrate an effect of latitude; the effect of latitude on all traits and properties, as well as sample sizes, can be found in Table S5 and Figure S5. Lines represent the derived regression from the mixed-effects model with the shaded region as the 95% confidence interval.

z 1 = 1.99, p = 0.05), fruit weight (β = 0.05, z 1 = 2.36, p = 0.02), and belowground N content (β = 0.06, z 1 = 3.32, p < 0.001) increased (Figure 3, Table S5). However, the trend for belowground N content is driven by one study around 80° latitude (Figure 3). There was also a marginal increase in the effect of warming on the number of flowers as latitude increased (β = 0.01, z 1 = 1.68, p = 0.09). On the other hand, there was a marginal negative effect of warming on spring phenophases as latitude increased (β = -0.02, z 1 = -1.81, p = 0.07).

As mean annual precipitation increased, the effect of warming on flower lifespan increased (β = 0.07, z 1 = 2.19, p = 0.03; Table S6, Figure S6). Species closer to their northern range edge were more sensitive to warming in terms of increased plant growth (β = -0.014, z 1 = -2.11, p = 0.03) and percent cover (β = -0.036, z 1 = -2.40, p = 0.02; Table S7, Figure S7). Mean annual temperature did not influence the effect of warming on any plant traits or community properties (Table S8, Figure S8).

For experimental contexts, we tested for an effect of the timing of warming, the amount warmed by the experiment, and the number of years warmed. Only when warmed year-round did aboveground biomass (Hedges' g = 0.65, z 2 = 3.29, p = 0.001), belowground biomass (Hedges' g = 1.06, z 2 = 2.75, p = 0.006), fruit weight (Hedges' g = 0.94, z 2 = 2.38, p = 0.02), and aboveground N content (Hedges' g = -0.52, z 2 = -3.78, p < 0.001) demonstrate an effect of warming (Figure 4, Table S9). In contrast, fall phenology (Hedges' g = -1.70, z 2 = -2.15, p = 0.03) and fruit number (Hedges' g = 0.63, z 2 = 1.78, p = 0.08) only demonstrated a warming effect when warmed seasonally (Figure 4, Table S9). Plant growth, leaf growth, and spring phenology demonstrated similar effects of warming, regardless of the timing of warming (Table S9, Figure S9). The Holm-corrected comparisons demonstrated differences between year-round and seasonal warming for aboveground biomass and fall phenology (Table S10).

Belowground N content demonstrated a response to the amount warmed by the experiment, with warmer temperatures leading to a stronger decrease in belowground N content (β = -0.98, z 1 = -1.67, p = 0.05; Figure S10). The amount warmed by the experiment did not affect any other plant traits or community properties (Table S11, Figure S10). As the number of years warmed increased, the effect of warming on flower lifespan (β = 0.57, z 1 = 2.02, p = 0.04), spring phenophases (β = 0.13, z 1 = 1.77, p = 0.08), fruit number (β = -0.23, z 1 = -1.95, p = 0.05), and leaf growth (β = -0.03, z 1 = -1.83, p 0.07) became weaker (Table S12, Figure S11). That is, the effect size became closer to 0 as the number of years warmed increased.

For plant-level contexts, we tested for an effect of plant functional group and plant native status. Graminoids were the most common functional group to experience an effect of warming, with 9 out of 13 traits/properties demonstrating a warming effect on graminoids (Figure 5, Table S13). Forbs and shrubs were the second most common functional groups to show an effect of warming for 5 out of 13 traits/properties (Figure 5, Table S13).

Although bryophytes and lichens were not commonly measured plant types, they appear to have opposing responses compared to vascular plants for some traits. For example, bryophytes (Hedges' g = -0.46, z 7 = -3.34, p < 0.001) and lichens (Hedges' g = -0.39, z 7 = -2.72, p = 0.01) have reduced percent cover when warmed, whereas graminoids (Hedges' g = 0.22, z 7 = 2.34, p = 0.02) and shrubs (Hedges' g = 0.40, z 7 = 3.45, p < 0.001) have increased percent cover (Figure 5, Table S13). Holm-corrected comparisons also demonstrated that warming decreased percent cover for bryophytes and lichens when compared to shrubs and graminoids (Table S14).

Non-native species measurements were uncommon compared to native species measurements (Table S15); therefore, we limited our results to the traits with a sample size of n ≥ 10 for both native and non-native species. Native and non-native species showed similar responses to warming for leaf growth (Native: Hedges' g = 0.33, z 2 = 3.82, p < 0.001; Non-native: Hedges' g = 0.35, z 2 = 2.31, p = 0.02) and spring phenology (Native: Hedges' g = -0.56, z 2 = -3.14, p = 0.002; Non-native: Hedges' S13. Mean values are estimates from the mixed-effects model which accounts for species, site, and publication number. Points and error bars represent mean ± 95% confidence intervals.

g = -0.53, z 2 = -2.55, p = 0.01; Figure S12). For aboveground biomass, native plants were positively affected by warming (Hedges' g = 0.39, z 2 = 2.06, p = 0.04), whereas non-native plants had no response (Hedges' g = 0.07, z 2 = 0.25, p = 0.80; Figure S12). We found differences between native and non-native species for other traits and community properties (Table S15), but, as stated above, sample sizes were too low for non-native species to allow for meaningful statistics (e.g., aboveground N content had n = 100 for native, and n = 5 for non-native; Table S15). The Holm-corrected comparisons demonstrated differences between native and non-native species for belowground biomass, fruit number, and fruit weight (Table S16); however, these comparisons contained n < 10 non-native measurements.

Across 126 studies around the world, we identified how various plant traits and community properties responded to warming, as well as which contexts contributed to variation in these plant responses to warming. In terms of overall trait and property responses, we found support for the majority of our first set of hypotheses outlined in the introduction. The support for these hypotheses demonstrates that our findings generally align with the findings of previous warming studies; this shows that, globally, most plant traits and community properties have similar responses to warming when compared to more local-scale experiments. However, a variety of environmental, experimental, and plant-level contexts explained these responses to warming, including latitude, distance from range edge, the timing of warming, the amount warmed, the length of the warming experiment, and plant functional group.

In terms of environmental contexts, latitude explains much of the variation in some traits and community property responses to warming across studies. As latitude increases, the effect of warming becomes stronger. Depending on the trait type, warming either has a more positive effect at higher latitudes (e.g., number of flowers, number of fruits, and fruit weight) or a more negative effect (e.g., spring phenophases). These changes may be linked in that at high latitudes, warming leads to even earlier spring phenophases, which could contribute to greater input into reproductive traits later in the season (Sercu et al. 2021). These results demonstrate that plants at higher latitudes may show a more pronounced warming response than plants at lower latitudes. This finding could be attributed to the fact that as climate variability increases, plasticity may also increase (Anderson and Song 2020). Plants that reside in regions that experience greater variations in temperature, such as plants at higher latitudes, may have stronger thermal tolerances and therefore greater plasticity compared to plants that experience less variation in temperature (Ghalambor et al. 2006;Janzen 1967). Therefore, due to greater temperature variability at high latitudes, plants residing in those areas could have greater phenotypic plasticity, and accordingly, more pronounced responses to warming. Future studies could expand this idea by empirically calculating climate variability between sites and testing if warming responses depend on the amount of variability.

Interestingly, many of the traits that experienced an effect of latitude were traits related to reproduction (number of flowers, number of fruits, and fruit weight), demonstrating that reproductive traits may be more susceptible to latitudinal gradient effects than other traits. Unfortunately, there is a strong Northern Hemisphere bias in our dataset, which may limit some of our understanding of climate effects across a broad range of latitudes (Hansen and Cramer 2015).

Species closer to their northern range edge experienced a more positive effect of warming in terms of increased plant growth and percent cover. In contrast, species further away from the range edge experienced lessened or potentially more negative effects of warming (i.e., decreased plant growth and percent cover). Similar to our results for latitude, this effect may also be due to increased climate variability leading to increased plasticity for plants that reside at higher latitudes (Ghalambor et al. 2006;Janzen 1967). Plants closer to their northern margin may also be undergoing selection for traits that enhance colonization, therefore potentially contributing to differential trait responses to warming across a species' range (Buizer et al. 2012;Kilkenny and Galloway 2013). However, more research is necessary to uncover how plant traits and community properties may relate to species range shifts and edge expansion with further climate warming. Our use of GBIF and BIEN occurrence records also may not have captured the true northernmost extent, and therefore the leading edge, for some species.

The timing of warming, which is an experimental context, also plays a large role in some plant responses to warming. For example, warming only affected aboveground biomass, belowground biomass, fruit weight, and aboveground N content when warming was year-round (Figure 4). In contrast, warming only affected fruit number and fall phenophases when warming was seasonal. The results for belowground biomass and fruit number may not be as meaningful due to low sample sizes with seasonal experiments (n = 2 and n = 9), compared with larger sample sizes for year-round experiments (n = 42 and n = 25) (Table S9). Interestingly, we also see that fall phenophases have a negative effect size when warmed seasonally, but a slightly positive effect size when warmed year-round. Most climate studies find a delay in fall phenology when plants are warmed (Collins et al. 2021;Peñuelas and Filella 2001;Walther et al. 2002), shown by a positive effect size in this study. Climate change projections show that winter warming will be significant (Kreyling et al. 2019); therefore, year-round warming experiments may more accurately reflect the effects of climate warming. These results may also be due to plant responses to varying conditions associated with year-round versus seasonal OTC studies, such as potential differences in soil moisture, snowpack, etc.; therefore, more research is needed to parse apart year-round vs. seasonal warming responses.

Other contexts associated with experimental design, such as the amount warmed by the experiment and the length of the study, affected a few traits and properties. Belowground N content decreased as the amount warmed by the experiment increased; however, the sample size for belowground N in this analysis was somewhat low (n = 7; Figure S10). We were surprised to find that no other traits or properties were affected by the amount warmed by the experiment, as the amount warmed ranged from 0.10°C to 4.60°C (Figure S13). This finding demonstrates that plant responses to warming may be similar across a range of temperatures. On the other hand, certain species or species types may exhibit more pronounced changes in stronger warming, which this multi-species analysis does not capture. The plants in these experiments may have also reached the extent of their plastic capabilities, meaning that even if temperatures were to increase further, a trait may not continue to change if factors in the environment limit its plasticity (Valladares et al. 2007). Furthermore, at extreme temperatures, plants may be limited in their ability to plastically respond due to high amounts of stress (Valladares et al. 2007). However, natural anthropogenic climate change is often associated with other abiotic and biotic changes, such as changes in precipitation regimes (IPCC 2021) or biotic interactions (Zarnetske et al. 2012). Therefore, increased temperatures coupled with other changes are likely to produce novel plant responses that may not be seen with temperature increases alone (Xu et al. 2013;Young et al. 2024).

In terms of the length of the experiment, we found that long-term experiments led to weaker warming effects on flower lifespan, spring phenology, leaf growth, and fruit number when compared to short-term experiments (Figure S11). Over time, warming stress may not have as strong an effect on a plant's phenotype if the plastic response is reversible (Kristensen et al. 2020). Plastic responses may be immediate when plants experience novel, rapid stress (Franks et al. 2014), but the effect may lessen over time as the community acclimates to a new regime.

Finally, in terms of plant-level characteristics such as growth form and native status, the results varied based on the trait or property measured. Although there were similar sample sizes in this meta-analysis for graminoids, forbs, and shrubs, graminoids most often demonstrated a warming response (Figure S14, Table S13). We found some contrasting results between nonvascular (bryophytes and lichens) and vascular growth forms, with non-vascular species often being more negatively affected by warming (e.g., decreased percent cover; Figure 5). This finding corroborates other research on this topic in tundra ecosystems (Elmendorf et al. 2012;Walker et al. 2006), but our study highlights this warming effect on non-vascular species globally (Figure S15). However, non-vascular species had relatively small sample sizes (ranging from n = 39 to 46) compared to other growth forms (graminoids, forbs, shrubs, etc.; ranging from n = 327 to 338; Figure S14). Our analyses comparing native and non-native species were hindered by low representation of non-native species in warming studies (Table S15). Contrasting responses between plant types may explain why some traits and properties do not exhibit an overall effect of warming. For example, the positive effect of warming on vascular plant percent cover and the negative effect of warming on non-vascular plant percent cover may lead to no change in total community percent cover (Figure 2). These variations in plant responses to warming highlight the importance of considering differences between plant types (e.g., growth forms, provenance, etc.) in climate change studies.

In this meta-analysis, we find some clear relationships that increase understanding of climate warming impacts on plants around the world and guide future experiments. Below we highlight three broad conclusions and recommendations based on our findings.

First, researchers conducting warming experiments should carefully consider their experimental design in the context of interpretation of their results. Seasonal warming studies may not fully capture trait responses to climate warming due to the lack of winter and potentially early spring warming in those studies. While OTCs are typically employed for monitoring summer warming responses, previous OTC studies have noted winter warming effects (Hollister, Elphinstone, et al. 2023;Hollister, Robitaille, et al. 2023;Welshofer et al. 2018). We recognize that seasonal warming studies may be the only option in locations that receive heavy snowfall; therefore, researchers should recognize that the plant responses observed in summeronly warming experiments may be weaker than responses to year-round warming (Sanders-DeMott and Templer 2017). This finding demonstrates how experimental warming studies may be underpredicting the severity to which climate warming affects plants, which has been corroborated in studies such as Wolkovich et al. (2012).

Furthermore, in terms of open-top chamber experiments specifically, it is known that the chambers can influence other variables such as soil moisture, wind, humidity, etc. (Hollister, Elphinstone, et al. 2023;Hollister, Robitaille, et al. 2023); however, all in situ temperature manipulation experiments harbor their own unintended effects, so these issues are not unique to chamber studies specifically. The design of the chambers themselves and the underlying plant communities may also influence the amount of warming achieved (Hollister, Elphinstone, et al. 2023;Hollister, Robitaille, et al. 2023;Young et al. 2024).

While it can be difficult to generalize plant responses to warming due to the many nuanced factors that play a role in determining how traits and properties respond to climate stress, warming experiments are still necessary for us to understand the mechanisms that underlie plant responses to climate change. Many efforts are being made to synthesize similar warming experiments across the globe in order to parse apart these plant responses to warming, such as the WaRM Network (Prager et al. 2022).

Long-term studies are also essential in aiding our understanding of plant community change over time; networks such as the Long-Term Ecological Research network contain valuable information on community changes over long time scales (Cusser et al. 2021;Knapp et al. 2012). Coordinated experiments such as these are an effective method for mechanistically understanding how varying contexts contribute to warming responses.

Second, more studies are needed that investigate warming effects on different plant types, such as native vs. non-native or vascular vs. non-vascular species. Our study highlights the dearth of data on non-native and non-vascular species responses to warming compared to their counterparts (Figure S12, Table S15). Studies have demonstrated that under future climate regimes, nonnative species may benefit over natives (Zettlemoyer et al. 2019), but more research is needed on this topic to determine which specific responses may be beneficial for population persistence and invasion success (Chen et al. 2024). Similarly, our study corroborates past findings on negative responses of non-vascular species to warming (Elmendorf et al. 2012;Walker et al. 2006), suggesting that those species types could decline globally in the future. By including varying species types in climate studies, we could gain a stronger understanding of which species may persist or perish under future climate regimes.

Third, we highlight several contexts that contribute to specific trait and property responses to warming and what the implications of these findings may be. We found that spring phenology and reproductive trait responses to warming are more sensitive to latitudinal gradients than other plant responses. With ongoing climate warming, this shows that plants further from the equator may experience increased input into these reproductive traits, and therefore potentially increased fitness, which could promote species persistence in these regions. We also found that the length of the warming experiment can alter the magnitude of trait change; shorter experiments demonstrated stronger trait changes when compared to longer experiments, which shows that initial plastic responses to warming stress may be strong, but under the same level of warming over time, the community may acclimate to the stressor and return to a baseline level. We identified several other contexts that contribute to variation for specific plant responses, but overall, with more warming and the increased pace of warming as is predicted with ongoing anthropogenic climate change (IPCC 2021), we can expect to see further plant trait changes and community shifts in response to warming that may vary according to the contexts defined here.

In conclusion, warming experiments are necessary for us to be able to test causal relationships between plants and climate. We highlight several important findings regarding plant and community responses to experimental warming, such as the effects of latitude and experiment length. More coordinated experimental network studies across multiple contexts, including underrepresented regions (e.g., the Southern Hemisphere), varying habitat types, and multiple species types would improve our understanding of how plants and communities may respond to climate stress and what contexts are important for defining these responses.