Investigation of the Relationship Between Treeline Migration and Changes in Temperature and Precipitation for the Northern Hemisphere and Sub-regions

Numerous studies have reported that treelines are moving to higher elevations and latitudes over the last few decades. Climate change is assumed to be the main driver behind the observed migration of treelines. Predicting how treelines are likely to move in the future is important to identify impacts on sensitive high-alpine ecosystems and help model potential climatic feedback mechanisms associated with the expansion of alpine and arctic forests. Here, we examine the correlation between treeline movements and recent climate change to better understand the drivers of treeline migration. Changes in monthly minimum and maximum temperature and precipitation between 1961 and 1970 and 2009 and 2018 for 256 treeline sites in the Northern Hemisphere were analysed through logistic regression modelling. The model was run at various spatial scales to assess if globally significant climatic drivers are applicable at regional scales. Our analysis revealed that the rate of temperature change during the northern hemisphere autumn, particularly in October, was a significant predictor of treeline movement. At the end of the growing season, increased minimum temperatures corresponded with treeline migration. At warmer maximum summer temperatures, treelines were more likely to remain stationary. While statistically significant correlations were identified at global scales, there were also substantial regional correlations of treeline movement regarding warmer temperatures. Changes in precipitation correlated with treeline migration were not statistically significant. Continued warming is likely to further extend the growing season for alpine forests, promoting continued encroachment of forests into alpine grassland and tundra in areas where moisture availability does not present a limiting factor for tree survival.


Introduction
Treelines, which represent the upper bio-climatic limit for trees, are highly temperature-dependent and indicate the boundary beyond which trees are no longer capable of establishing. Treelines are predominantly located in cold areas at high elevation or latitude and are restricted by cold temperatures during the growing season [1]. Visually, treelines can either be characterised by an abrupt limit where the difference between forest and non-forest is easily determined or by gradual transition zones where trees become progressively smaller with increased altitude or latitude [2] and can be scattered with a combination of young trees and old dwarf trees [3]. The total aboveground biomass tends to decrease rapidly as the vegetation transitions from forest to non-forested areas [4,5].
Trees are in some instances absent from their bio-climatic limit due to non-climatic factors such as fires, topography, or human land use. In this paper, we use the treeline definition proposed by Körner [6], which excludes forest limits caused by factors unrelated to tree biology or those related to the absence of suitable treeline taxa. These climatically limited treelines tend to follow a common isotherm where the seasonal mean ground temperature is close to 6.7 °C [7]. The establishment of trees requires a minimum growing season duration of 94 days with a mean daily temperature of > 0.9 °C (i.e., cold temperatures during the growing season primarily limit treelines) [1]. The widely accepted theory of temperatures being the primary factor determining treeline positions is supported by the correlation between historical changes in treeline positions and temperature change [8,9] and recent advances of northern hemisphere treelines associated with global warming [10, 11••]. Elevational treelines, meaning those that form on mountains, have so far been migrating at a greater frequency, while those forming at high latitudes are likely to migrate over greater distances [11••]. Treeline migration primarily occurs either as a result of successful establishment of new trees above the present treeline, or by height growth of old established trees with stunted growth form, also referred to as krummholz [12•].
Changes to the world's treelines' position may be significant even under a moderate warming scenario, with recent evidence suggesting that the majority of the world's studied treelines (66%) have already seen an increased elevational or latitudinal extent [11••]. Advances of treelines are dependent on the successful survival of new seedlings, which may be influenced by various factors [13]. Climate change can affect the rate of successful seedlings through warmer summers which can improve seed viability and emergence of seedlings, and by increased winter temperatures which cause reduced mortality rates of seedlings and trees [14]. Recent warming trends have increased the average and maximum recorded temperatures and increased the length of the growing season [15] by up to 20 days in some regions [16,17]. In addition to treeline migration, warmer temperatures can lead to several changes to the phenology of trees, such as earlier leaf unfolding, earlier spring bloom and later leaf fall [17]. Other climatic responses have also been noticed, such as increased radial growth [18] and more successful seedling recruitment leading to increased forest densities [19].
Noting that temperature is the primary limiting growth factor for treelines, an increase in temperature could rapidly advance the current treeline [3]. Predicting treeline migration under future climate change is important because of the effect changes in alpine and high latitude vegetation have on both sensitive alpine ecosystems and the carbon cycle [11••]. Predicting migration is often difficult, as treeline migration is not an instant response to climate change but a slow-changing process that often lags temperature increases. Globally there is a relationship between the advance of treeline positions and increased winter temperatures, where treelines located in areas experiencing a high degree of warming during winter are observed to be more likely to advance than other treelines [10]. However, treelines around the world are comprised of a range of different species, whose successful regeneration and early survival may be influenced to a varying degree by increased temperatures [13]. Despite global temperatures rising, it is clear that treeline advancement has not occurred uniformly worldwide. This may be due to substantial seasonal and spatial variability in global temperature changes [20] or strong ecological or physiological differences between treeline taxa in different parts of the northern hemisphere.
The literature about correlating temperature change to modern treeline migration is limited, with the most recent being the 2009 study by Harsch et al. [10]. They identified the rate of warming in winter as the most important climatic predictor for treeline movements. However, their study was limited as only correlation with changes to summer and winter temperature was assessed, and changes to neither spring nor autumn temperatures or changes to precipitation were included. In addition, there has been a substantial increase in the literature available on treeline movements since this study. Here, we examine the literature on the link between modern climate change and the movement of treelines in the northern hemisphere. First, we examine whether changes in temperature and precipitation have changed across 256 treeline locations between 1961 and 2018. Secondly, we utilise logistic regression models to investigate the correlation between changes in climatic parameters and treeline migration in various geographic regions to establish if regional trends may be distinct from global trends. The climatic parameters are the minimum and maximum recorded temperatures and total precipitation. Lastly, we discuss climatic and non-climatic variables important to treeline migration that may not be captured in large-scale analyses and highlight several factors that should be considered when trying to predict future treeline migration.

Treeline Data
The treeline data used for the climate analysis were published by Hansson et al. [11••] and cover 477 locations from 142 publications where treeline migrations have been studied. Due to the wide variation in the rate of treeline advancement and the lack of data reported on this, data were analysed as binary categories for "migrated" or "stationary" treelines. We removed sample sites from the database where human disturbance such as land clearance had been noted (99/477). This left 378 study locations spread across 23 countries. Most of these sites were in the northern hemisphere, with only 15 study sites being located across two countries in the southern hemisphere. Samples sites from the southern hemisphere were removed from the analysis due to the poor representation of southern hemisphere treelines and the different ecological tolerance of southern hemisphere treeline taxa such as Fuscaspora cliffortiodies [21]. Data from Southern and Central America were also removed, as only two samples from that continent were in the Northern Hemisphere. This left us with a total of 361 sample sites.

Combine the Treeline and Climate Data
Modelled high-resolution historical monthly weather data from 1960 to 2018 were obtained from WorldClim Historical Monthly Weather Data [22••]. However, the data in 1960 were incomplete and excluded from the analysis. The variables included average minimum temperature (°C), average maximum temperature (°C) and total precipitation (mm) for each month of the year. The spatial resolution of this dataset is 2.5 min, which equates to an average area of approximately 21 km 2 . The climate data were split into two decades: the first complete decade of 1961-1970 and the last entire decade of 2009-2018. These time periods were selected as they represent the longest timespan on which climate data has been measured, including recent years during which warming has been the most prominent in many parts of the world.
The climate data were converted into six raster layers, one for each time period and climate variable used in the analysis. The treeline data, now consisting of 361 sample sites, were also imported into ArcGIS v10.6 [23] as point data using the coordinates from Hansson et al. [11••]. The average monthly recorded value for each decade and climate variable was then extracted using the "extract values to point" feature in ArcGIS. Using this tool, we extracted the monthly average minimum and maximum temperature and the average total precipitation occurring for each month between 1961 and 1970 and 2009 and 2018 at all treeline locations.
The treeline data were further cleared of study sites located near each other, as an overrepresentation of samples within a local area can skew both regional and global trends. The climate data are at a resolution of 2.5-min grids, resulting in sample sites sometimes being located within the same climate grid and therefore being attributed to the same climatic data. First, we categorise all treeline data as being either stationary or migrating. We then identified climatic grids which had multiple sample points within them. One data entry was kept for each climate grid where all samples were classified as either migrating or stationary. Two data entries were kept for grids where treelines had been recorded to both migrate and remain stationary within the grid. The final dataset contained 256 sample locations sourced from 99 published records. Within the dataset, 68% of sites had experienced treeline advancement to higher altitudes or latitudes. The data were broken down based on continents, as summarised in Table 1. The region with the greatest proportion of migrating treelines is Asia, closely followed by the USA, while the geographic region with the least migrating treelines is Europe.

Define Changes of Climate Parameters and Geographic Regions
Monthly recorded data for each variable were averaged across the two time-series, resulting in a dataset containing average monthly minimum temperature (°C), maximum temperature (°C) and total precipitation (mm) for 1961-1970 and 2009-2018. To examine how climate has changed across the various sites, the recorded monthly data of the historical period of 1961-1970 was subtracted from the average recorded climatic data of 2009-2018. By doing so, the change in average minimum temperature (dMinTemp) (°C), average maximum temperature (dMaxTemp) (°C) and total precipitation (dPrec) (mm) between these two decades was determined.
The data were categorised based on geographic distribution and treeline form ( Table 2). Data were assessed both at a hemisphere scale and at a continental scale. This was done to identify regionally significant trends that may not be present at a hemisphere level. The treeline form was classified as elevational or latitudinal treelines, with elevational treelines growing on mountains, while latitudinal treelines defining the highest latitudes at which trees can establish in the area irrespective of elevation. This separation was done as altitudinal treelines are known to more often show signs of treeline migration [11••]. For the geographic region Northern Hemisphere, data were modelled using logistic regression for three sub-categories, firstly for all samples and secondly for altitudinal and latitudinal samples separately.
The Asian treeline data represent treeline positions with a broad distribution across both latitudes and longitudes. Data are most highly represented from the Ural Mountains and in the Himalayas (Fig. 1a). As with the Northern Hemisphere data, the data were categorised into three groups: all data, altitudinal data and latitudinal samples ( Table 2). The latitudinal sites were not used for logistic regression modelling due to the small number of samples (5 in total) and show that no latitudinal treeline in Asia had experienced migration.
Data points in Europe are mainly located in central parts of Sweden, the European Alps and the Carpathian Mountains (Fig. 1b). The European data were first categorised based on if data were altitudinal or latitudinal. Logistic regression modelling was not performed on the latitudinal data, as only one sample in Europe was classified as latitudinal. The European   All data  256  173  83  Altitudinal data  227  159  68  Latitudinal data  29  14  15  Asia  All data  67  52  15  Altitudinal data  62  52  10  Latitudinal data*  5  0  5  Europe  Altitudinal data  81  43  38  Latitudinal data*  1  1  0  Scandinavia  49  22  27  Continental Europe  32  11  21  North America  All data  107  77  30  Altitudinal data  84  64  20  Latitudinal data  23  13  10  Above 60°latitude  66  49  17  Below 60° latitude  41  28  13 data were further allocated to sub-categories depending on their geographic position, with treelines in Scandinavia and Continental Europe (here defined as all of Europe excluding Scandinavia) being modelled separately ( Table 2). Migration of North American treelines has mainly been studied in the Rocky Mountains and Sierra Nevada of the contiguous USA (Fig. 2). Sample sites across Alaska and Canada are broadly distributed across the continent. For modelling purposes, the data for North America were divided into five categories: all North American data, latitudinal treelines, altitudinal treelines, treelines below 60°N and treelines above 60°N ( Table 2). The split of data in accordance with their latitude was done to identify if treelines close to and within the arctic circle showed a different response to warming than those below.

Logistic Regression Modelling
Logistic regression was used to identify the relationship between the probability of treeline migrated and the defined climate variables. Let y i be a binary variable of treeline migrated at the i th location in a geographic region or a subcategory. When y i = 1 , the treeline was migrated; when y i = 0 , the treeline was stationary. By assuming y i follows Bernoulli distribution, the expectation of y i is equal to the probability of treeline migrated (i.e., E y i = p i ) at the location i . The logistic regression models p i in the form of logit functions to linearly correlate with the climate variables. The logit function is the logarithm of the odds; that is, McCullagh and Nelder [24] and Schabenberger and Pierce [25] for detail). The model can be represented as follows where is the intercept, x i the column vector of the climate variables and the vector of the associated coefficients.
The regression coefficient describes the amplitude and direction of the relationship between a climate variable and treeline migration (i.e., the probability of treeline migration to occur). A positive coefficient can be interpreted as increasing the chances of treeline migration, while a negative coefficient instead reduces the chances of migration. The coefficient also describes the change of the odds regarding treeline migration. Given other variables are fixed, a one unit increase in the k th element of x i alters the odds by factor e k [26], and this factor implies (e k − 1) × 100% change in the odds of treeline migration. The logistic regression was fitted independently to the data of each month in each geographic region and subcategories. The model fitting was conducted in the environment of R version 4.1.1 [27].
We note that logistic regression mixed models [28,29] were applied to the data with climate variables as fixed effects and study levels as a random effect. However, the Europe. Samples that were closely distributed have been displaced to reduce overlapping of symbols. The displacement is visualised in a circular pattern, with a black dot indicating the central position of the displaced symbols mixed models struggled with convergence issues and produced results less reliable than the logistic regression model. The cause of the issue was that the data were highly unbalanced; 60% of study levels (about 60 studies) provided only one sample, 14% (about 14 studies) provided two samples, and 7% (about seven studies) provided three samples, but five studies (about 5%) contribute 8,9,11,14 and 32 samples, respectively. A simple synthetic simulation study (see the R code in Supplementary Material) showed that the logistic models provided more robust results than the mixed models, and caution is necessary to interpret the mixed model results. The results of the mixed models are presented in the Appendix.

Results
The results are presented according to the geographic regions specified in Table 2. Firstly, the changes of the three climate variables within the geographic region are summarised in violin plots (see Hintze and Nelson [30] for detail). Secondly, results from the logistic regression regarding the changes of dMinTemp, dMaxTemp and dPrec, respectively, are presented for each sub-category. The model was also applied to dMinTemp and dPrec as well as dMaxTemp and dPrec concurrently. These results are presented in the Appendix. We note that interpreting the coefficients in the Appendix should be cautious because the correlations between temperature and precipitation in some months by region are moderate or even strong. These results have been included in the appendix. In general, dPrec have a limited correlation with treeline migration for most locations, and the amount of warming appears to be far more important in predicting treeline migration.

Northern Hemisphere
dMinTemp has on average increased by 1.5 °C across all months of the year, while dMaxTemp has increased by an average of 1.2 °C (Fig. 3). Overall, there has been a greater temperature increases in both maximum and minimum temperature during the months of November to March, where the median temperature has increased around 2 °C. In the remaining months, median temperatures increased around 1 °C. The highest minimum temperature increases recorded were up to 6.5 °C in January. The median maximum temperature change shows fewer seasonal variations, with the only month recording less than 1 °C increase being October, which has only increased by 0.44 °C. Neither dMinTemp nor dMaxTemp data primarily has a unimodal distribution. Several months show a bimodal or trimodal distribution of data. Because the distribution of dMinTemp and dMaxTemp is highly skewed and not unimodal, the means of these two variables is an inappropriate and biased statistics to represent them. For example, results of dMaxTemp in February show that most sites have warmed either less or more than median, with the highest frequency of data occurring at the quartiles. Many months are also characterised by large variation of the data and are skewed to the right (Fig. 3).
Median dPrec is for most months close to zero, averaging 0.05 mm across all months. A few sites recorded changes greater than ± 50 mm (Fig. 4). A slight majority of sites (54%) have experienced an increase in precipitation since 1961. Precipitation shows a great variability across treeline sites; some locations experienced monthly variations of ± 100 mm. The dPrec data mostly display a unimodal distribution, with first and third quartiles including outliers that have recorded changes far greater than most sites.
Regression analysis showed that temperature change had a statistically significant effect on treeline migration across several months (Table 3). Higher dMinTemp is significantly correlated to greater odds of treeline migration (P value of < 0.05). For every 1 °C of additional warming in June, September and October, the treeline migration's odds increased by 43%, 88 and 84%, respectively. More increased dMinTemp are not associated with increased dOdds of treeline migration across all months; however, none of the months with a negative coefficient has a P value of < 0.1. Changes in dMaxTemp in July, September, October and November are also significantly correlated to treeline migration. In contrast to minimum temperatures, dOdds in July, September and November was reduced by − 50%, − 51% and − 39%, respectively. The only month showing a clear trend between more rapid rates of warming and treeline migration is October, where the dOdds increases by 66%. dOdds of treeline migration associated with dPrec is relatively minimal across all months, corresponding to a change of dOdds in the range of − 4 to 3%. This means that a relatively large change in precipitation is required to correspond to an overall likelihood of treeline migration.
Altitudinal treelines are somewhat more influenced by changes to maximum temperatures compared to latitudinal treelines (Table 3). Warmer dMaxTemp during mid to late summer (Jul-Sept) reduced the dOdds for treeline migration. Increased dMinTemp and dMaxTemp positively influenced the dOdds of migration in June and October. Warmer dMinTemp across latitudinal treelines had the most significant influence on the dOdds of treeline migration during the summer. October stands out for altitudinal treelines as the most influential, with dOdds increasing by 119% for dMinTemp and 71% for dMaxTemp. dPrec has a statistically significant relationship with treeline migration in several months, but dOdds are overall minimal (at most − 5% for dPrec in May).
For latitudinal treelines, warming dMinTemp in July and August is associated with a dOdds of 2393% and 1662%, respectively. More rapid rates of warming during winter (February and December) increased the dOdds of treeline There was a slight trend for increased dPrec to negatively correlate to treeline migration in spring and early summer but positively influence the late summer and autumn dOdds. dPrec was significantly correlated with treeline migration in January, March and July. It is essential to point out that there are relatively few sites where latitudinal treelines have been measured (29 samples), which likely contributes to the relatively high standard error seen during many months.

Asia
Across the Asia data, there was a mean increase in dMinTemp of 1.6 °C across all months, while dMaxTemp has seen a rise of just 1 °C. Warming has been most significant during winter and spring, with summer months experiencing less rapid warming rates (Fig. 5). There is also less variability in temperature changes during the summer months, with July, August and September having close to a unimodal distribution. The most prominent temperature increases occurred during the first 6 months of the year, with the most rapid rise of dMinTemp of 5.8 °C occurring in January. Spring and early summer temperatures generally show a bimodal or trimodal distribution and are skewed to the right, indicating that several sites have seen increases far above median during these months. Some locations experienced a cooling of temperatures in certain months, but no single site has experienced a cooling of dMinTemp or dMax-Temp during all months.
dPrec has seen very small changes of − 0.3 mm. However, there is great monthly variability, with dPrec increasing by up to 105 mm in August and reducing by as much as 100 mm in September at some locations ( Fig. 6). First and third quartiles are for several months no greater than ± 10 mm different than median values. The temperature data mostly showcase a unimodal distribution, with the greatest variation of dPrec seen in early summer.
Logistic regression analysis revealed only a few significant correlations between climate change and treeline migration in Asia (see Appendix). Treeline movements were not statistically significantly influenced by either dMinTemp or dMaxTemp. While not statistically significant, a trend was apparent where increased dMinTemp was associated with positive dOdds. The opposite trend was evident for dMaxTemp, where increased dMaxTemp generated negative dOdds. Precipitation does again appear to have limited influence on dOdds.
Correlations between climate change and treeline migration are more apparent when the latitudinal data were removed (Table 4). While consisting of few samples (only 5), all latitudinal treelines have remained stationary. Increasing dMinTemp had a statistically significant positive effect during summer and autumn when these samples were removed. Warming in August and September is associated with dOdds of 763% and 419%, respectively. Removing latitudinal treelines did not bring out any statistically significant correlation between dMaxTemp or dPrec and treeline movements. Europe dMinTemp increased by a mean of 1.5 °C in Europe across all months, while dMaxTemp increased by 1.4 °C. The most significant increase in dMinTemp and dMaxTemp occurred during the winter months and July (Fig. 7). Several months have close to unimodal distribution, where most sites experienced similar levels of warming. This includes dMinTemp and dMaxTemp in April, May and September. Some months,  such as June and August, have a bimodal distribution of dMaxTemp. There was almost no rise in temperature during October and no change to median dMaxTemp in June. Several sites experienced a cooling of July and October temperatures, contributing to the low median change during these months. While the lack of change observed in October, dMaxTemp appears fairly homogenous amongst sites, June temperatures show a great variation in distribution of temperature change between sites. dPrec in Europe has increased by less than 1 mm if averaged across all months, and variability between sites ranges from − 35 mm in August to + 39 mm in September (Fig. 8). Most months have a unimodal distribution of dPrec, with changes not exceeding ± 10 mm across most sites and months. Data from August indicated that a large frequency of sites experienced a small increase in precipitation, and that most sites have seen a reduction in dPrec during October.
The logistic regression modelling for Europe was performed excluding the only sample classified as latitudinal in this area. The model found little correlation between treeline migration and temperature changes in Europe at large (see Appendix). Warmer dMinTemp in spring was negatively correlated with the dOdds of treeline migration, with every 1 °C warming in March, April and May being associated with a reduction in the odds of treeline migration by − 65%, − 73% and − 17%, respectively. Cooler dMinTemp and dMaxTemp also appeared to be associated with treeline migration in October, a month where several sites experienced a cooling of temperatures (see Fig. 7).
Logistic regression modelling was also performed on the continental treelines of Europe and Scandinavian treelines. While no significant correlation was identified between temperature change and migration in Scandinavia (see Appendix), the influence of dMinTemp became more prominent once Scandinavian samples were removed from the European dataset (Table 5). Increased dMinTemp between March and August was significantly associated with reduced dOdds of treeline migration. These results, while significant, were associated with standard errors ranging from 1.3 to 2.0. Warming of maximum temperatures in the months before and after March and August increases the dOdds of treeline migration by 350% in February and 1358% in September.

North America
dMinTemp in North America increased by a mean of approximately 1 °C in most months (Fig. 9). During January, March and December, dMinTemp increased more than average, with increases of 2.7 °C, 1.6 °C and 1.7 °C, respectively. The smallest variation in temperature increase was seen during summer months, and the greatest variation was seen for dMinTemp during winter. The data of these months were also skewed to the right. dMaxTemp increased by a lesser extent of around 1 °C or less in all months except January, where temperatures rose by 2.2 °C. Some locations have experienced a substantial increase in January and December maximum temperatures, with sites recording warming of up to 6.5 °C. dPrec has, in most places, changed no more than ± 10 mm/ month, with a handful of sites seeing changes of up to ± 50 mm in certain months (Fig. 10). The most variability in precipitation was recorded in summer (June-August). Most sites where dPrec was greater than ± 20 mm recorded these changes during the winter months of November to January, with some sites also experiencing a significant increase in March.
Increases in dMinTemp during summer and fall are important indicators in predicting treeline migration in North America (Table 6). Increased dMinTemp in July, August and October increased dOdds of treeline migration by 189%, 193% and 92%, respectively. While warmer dMax-Temp during the shoulder months of March and September reduced the dOdds of migration, warmer dMaxTemp in October increased the odds of treeline migration. Changes in dPrec are less important in predicting treeline migration, but a significant correlation was identified between increased precipitation in April and May and slightly reduced odds of treeline migration (− 11%).
The North American data were divided further after both classification of treeline type (altitudinal or latitudinal) and based on their latitudinal position (as per Table 2). There was very little correlation between dMinTemp, dMaxTemp and dPrec and treeline migration for altitudinal treelines (see Appendix). dOdds of latitudinal treeline migration was, on the other hand, strongly correlated to more rapid rates of warming minimum temperatures during summer and autumn (Table 6). Increased dMinTemp in July, August and September increased the dOdds of treeline migration by 1741%, 2203% and 50,271%, respectively. While these odds are extremely high, they are also associated with a quite significant standard error. Again, it is important to point out that the sample size for this sub-category can be considered very small (23 samples). Warmer dMaxTemp in March and September reduced dOdds, while warming during April and October increased dOdds of treeline migration. Increased dPrec is also linked to negative dOdds during most of the year, and significantly so in January, March and May.
North American samples were also separated based on their latitudinal positions, with sites over and below 60°N being modelled independently. All treelines below 60°N were altitudinal, while those further north consisted of both altitudinal and latitudinal samples. Altitudinal treelines displayed very little correlation with temperature change at a continent scale (see Appendix). When separated, altitudinal treelines below 60°N appear to have a more uniform response to climate warming compared to when all altitudinal treelines are modelled together. Migration of altitudinal treelines below 60°N is strongly influenced by dMinTemp, particularly during the growing season (including shoulder seasons). Increased dMinTemp in July, August and September is associated with dOdds of 320%, 384% and 278%, respectively. dMinTemp in August and September reduced dOdds by 78% and 71%. There was also a significant correlation between reduced dPrec and treeline migration between January, February and April, and with increased dPrec in March. However, these changes have minimal influence on the change in odds of migration.
Both latitudinal and altitudinal treelines located below 60°N are strongly influenced by dMinTemp in summer and fall. The logistic regression modelling indicated that the altitudinal treelines above 60°N respond differently to climate change than their southern counterparts, as the trends seen for latitudinal treelines in Table 6 are no longer visible when these data are combined (see Appendix). Instead, reduced dMaxTemp

Discussion
Our findings suggest that temperature change has a measurable effect on treelines. In contrast, the correlation of changes in precipitation and the change in odds of treeline migration are overall small with low statistical significance. There was a stark difference between the influence of changes in maximum and minimum temperatures. Warming minimum temperatures are positively correlated with treeline migration, while increasing maximum temperatures are more often associated with stationary treelines. A summary of the odds of treeline migration that are associated with 1 °C of warming across all geographic zones used for logistic regression modelling is provided in Table 7. A shared characteristic across most sites is the importance of autumnal temperatures on treeline migration. More rapid rates of warming of minimum and maximum October temperatures appear to be particularly important for predicting treeline migration. These findings differ from conclusions made by Harsch et al. [10] for obvious reasons, as neither spring nor autumnal temperatures were assessed in their study.
We believe that there are a range of reasons why warmer minimum temperatures during summer and autumn stand out as being correlated with treeline migration at a hemisphere scale and why warmer overall temperatures in October significantly increase the odds of treeline migration. Firstly, warmer minimum September, and particular October, temperatures, are likely capable of extending the growing season in areas of marginal climates for tree growth. The onset of heavy frosts is often a driver for tree dormancy [31] and defines the length of the growing season. The establishment of trees requires a minimum growing season length of 94 days, with the mean temperature reaching at least 6.4 °C across all days [1]. Warmer minimum temperatures in September and October could increase the altitudinal or latitudinal limit where this threshold is met, thereby enabling trees to establish and grow. Secondly, the timing of temperature change is essential, as temperature changes in spring and autumn more strongly influence the growth of trees at the treeline than they do during the rest of the year. This is due to the length of the growing season being one of the most influential parameters controlling treeline positions [1]. Not only could warmer minimum temperatures in October increase the growing season, but warmer maximum temperatures could also contribute to increased tree growth with consequential effects on the robustness of seedlings through the following winter, as cool autumn temperatures are known to significantly inhibit growth conditions [4].
Spring temperatures were also strongly associated with treeline advancement at many sites and were particularly important in Europe and at higher latitudes in North America. Increasing temperatures in March, which is considered a shoulder season at many treeline locations, could be assumed to positively influence treeline establishment by prolonging the growing season. However, our results indicate that more rapid rates of warming temperatures in March instead significantly reduce the odds of migration at several locations. One of the reasons for this could be the increased risk of frost damage to plants. Snow acts as a protective layer for seedlings [32], and without a protective layer of snow, trees are more susceptible to frost damage during the early start  of the growing season [33]. This can increase the chances of extreme soil temperatures and frost events occurring once trees have entered their growing season [34]. The rate of temperature increase during winter is likely less important than changes occurring during the rest of the year due to temperatures during winter being below the physiological limit of tree growth. This does not mean that winter temperatures are insignificant, as milder winters have been known to reduce tree mortality [14]. When altitudinal and latitudinal treeline data are distinguished, it is apparent that each is influenced differently by climate change. For latitudinal treelines, changes to minimum temperatures appeared to be overall more important than changes to maximum temperatures. Latitudinal treelines in particular benefit from rapid warming of December temperatures, which may reflect increased tree and seedling survival from less harsh winters. The positive impact of warmer winters on tree survival may be why Harsch et al. [10] identified winter temperatures as more influential on treeline migration than summer temperatures in their study. But it may also be from the lack of separating maximum and minimum records, meaning the negative influence from increased maximum temperatures as compared to increases in minimum temperatures could not be assessed.
Increased maximum temperatures in late summer and autumn are more likely to reduce the odds of treeline migration. Most evaporation at treelines occurs during summer [4], which causes enhanced drought stress towards the end of the growing season. While temperature is the most common limiting factor for tree growth, drought-stressed trees can experience a negative impact from increased temperatures [35,36]. Drought stress combined with other pressures, such as insect outbreaks and herbivore activity, can exacerbate the stress of trees further and thereby increase tree mortality [37,38]. While some models exist on the relationship between these parameters (see, for example, Anderegg et al. [39] and Gely et al. [40]), these interactions are complex and less understood in the treeline environment. Insect outbreaks could reduce the climate sensitivity of trees and may become a limiting factor where droughts are becoming more prevalent, such as in Alaska [41]. This could indicate that increased temperatures potentially only accelerate tree growth in areas with sufficient moisture availability. As an example, Bunn et al. [42] identified that trees growing in the same region responded differently to trends in precipitation and temperature depending on the local soil moisture availability. Trees with ample water availability responded more strongly to changes in temperature, while trees with limited water supply responded more strongly to changes in precipitation. Most mountain regions have relatively shallow soils, which will cause altitudinal treelines to be more prone to water stress when compared to latitudinal treelines. This may explain why warmer summer temperatures less negatively impact latitudinal treelines compared to their altitudinal counterpart. Models only considering changes in precipitation without inclusion of more complex variables, such as existing moisture stress, are likely to not be able to accurately predict future treeline migration. In addition, the scale of precipitation change at many sites has been small to date. More significant changes would be needed for precipitation-based effects to become visible.
It is not only the total amount of precipitation that may be important but also the form of precipitation. As treelines are located in cold climate areas, precipitation falling during the winter months is likely to fall as snow. It is unlikely that precipitation falling during a single month in winter has any significant impact on treeline migration; it is more likely that the cumulative change in precipitation falling as snow during the dormant season impacts water availability during the growing season. While some depth of snow cover can protect seedlings [32], too much snow cover can cause physical damage from the loading of branches and leaves [34]. The very weak correlation and small overall influence of change in precipitation have on the odds of treeline migration could be because the modelling fails to account for both prevalent moisture stress and extreme precipitation events. These may be more important to treeline migration than changes in average precipitation.
Climatic correlations to treeline migration identified at a hemispheric scale are not necessarily reflected across all geographic regions. Rather, no single month and climatic variable were significantly correlated with treeline movements for all geographic sub-regions assessed in Table 2. There were several instances where locally significant trends were not identifiable globally, perhaps most notably in Europe and Asia. There was also a lack of any climatic correlations for some sub-regions in these areas. In Asia, this was most likely because altitudinal and latitudinal treelines had a substantially different response to warming, which became evident when latitudinal treelines were removed from the dataset. Europe was the only region in which warmer October temperatures were not statistically significantly correlated to increased odds of treeline migration.
The lack of such correlation is difficult to explain, but potential factors could be the importance of local settings influencing the microclimate surrounding the tree [43] or a lag effect between temperature changes and responses of treelines. Seedlings generally require multiple years of favourable growing conditions to fully establish [14], and a lack of good growing conditions in consecutive years could limit treeline advances despite trends of increasing temperatures. It is also important to point out that successful seedling establishment and survival depend on a range of biotic and abiotic factors and that increased temperature may not always positively influence seedling survival rates [44••]. In addition, soil temperatures have been connected both to tree growth and carbon cycles in treeline environments [45]. Hence, datasets of soil temperatures, such as those presented by Lembrechts et al. [46•], can beneficially be used to study their relative importance on treeline migration at larger scales. Outside of temperature change, light, water availability, nutrition, shelter from nearby vegetation, and soil depth all play a role in influencing seedling survival rates. In addition, the rugged topography present across altitudinal treelines may present an obstacle in treeline migration [47], which could cause trees to be absent from their climatic limit. If such topographic limitations are not identified at affected treeline sites, it can be challenging to pinpoint any climatic drivers behind treeline migration at a   0.86 regional scale. It is also essential to consider the fact that treelines around the world are comprised of different species and, while limited by the same growing temperature [7], may respond to a varying degree to increased temperature, depending on the overall success of seedling regeneration and survival [13].

Conclusion
Across most of the northern hemisphere's treelines, an observable pattern indicates that increased minimum temperatures during the autumn are a strong predictor for treeline migration. Warmer minimum temperatures during these months appear to extend the growing season, which can, in turn, extend the range of suitable habitat for trees into higher altitudes and latitudes. More rapidly warming summer temperatures seem to reduce the chances of treeline migration, perhaps likely because warmer summer temperatures influence evaporation and reduce moisture availability. Treelines in different parts of the world have shown varied responses to climate warming. In some regions, there is little evidence that changes in either temperature or precipitation have been essential factors influencing treeline migration. The lack of an apparent response to these variables may be either due to the coarse geographical scale at which the climate data available for this analysis had been measured or could be due to the range, local variability, and complexity of other biotic and abiotic factors that influence seedling survival. Ideally, factors such as frequency of frost during the growing season, moisture availability and physical barriers for migration should be considered when analysing treeline migration drivers. In addition, higherresolution climate data that takes into account that the elevation of treelines would support more precise prediction of the relationship between climate variables and treeline migration. Finally, more data on treeline positions and their movements must be collected in the Southern Hemisphere, as such data are currently very scarce.