Impacts of anthropogenic climate change on tropical montane forests: an appraisal of the evidence

In spite of their small global area and restricted distributions, tropical montane forests (TMFs) are biodiversity hotspots and important ecosystem services providers, but are also highly vulnerable to climate change. To protect and preserve these ecosystems better, it is crucial to inform the design and implementation of conservation policies with the best available scientific evidence, and to identify knowledge gaps and future research needs. We conducted a systematic review and an appraisal of evidence quality to assess the impacts of climate change on TMFs. We identified several skews and shortcomings. Experimental study designs with controls and long‐term (≥10 years) data sets provide the most reliable evidence, but were rare and gave an incomplete understanding of climate change impacts on TMFs. Most studies were based on predictive modelling approaches, short‐term (<10 years) and cross‐sectional study designs. Although these methods provide moderate to circumstantial evidence, they can advance our understanding on climate change effects. Current evidence suggests that increasing temperatures and rising cloud levels have caused distributional shifts (mainly upslope) of montane biota, leading to alterations in biodiversity and ecological functions. Neotropical TMFs were the best studied, thus the knowledge derived there can serve as a proxy for climate change responses in under‐studied regions elsewhere. Most studies focused on vascular plants, birds, amphibians and insects, with other taxonomic groups poorly represented. Most ecological studies were conducted at species or community levels, with a marked paucity of genetic studies, limiting understanding of the adaptive capacity of TMF biota. We thus highlight the long‐term need to widen the methodological, thematic and geographical scope of studies on TMFs under climate change to address these uncertainties. In the short term, however, in‐depth research in well‐studied regions and advances in computer modelling approaches offer the most reliable sources of information for expeditious conservation action for these threatened forests.


I. INTRODUCTION
Tropical montane forests (TMFs) are typically an evergreen ecosystem constrained to a narrow altitudinal belt, with an uneven canopy layer frequently enveloped by orographic clouds (Hamilton, 1995;Loope & Giambelluca, 1998;Still, Foster & Schneider, 1999;Foster, 2001;Richter, 2008). These forests often harbour high abundance and diversity of epiphytes (Loope & Giambelluca, 1998;Foster, 2001;Collin, 2001). TMFs are thus commonly referred to as cloud forests, mist forests or mossy forests, as well as numerous names in other languages, especially in Latin America (Brown & Kappelle, 2001) where most studies of this ecosystem have been conducted Soh et al., 2019). Unlike other vegetation types defined by their taxonomic affiliation (e.g. coniferous forest, oak forest, etc.) or their phenological structure (e.g. deciduous forest, xerophilous shrubland, etc.), TMFs are characterised by the intersection of an atmospheric phenomenon (fog incidence) and a topographic feature (mountain slopes). Such restrictive features highlight the potential vulnerability of TMFs in the face of climate change (Hamilton, 1995), but also make impacts difficult to isolate from these ecosystems' intrinsic climatic variability (Vuille et al., 2003). In this study, we focus on anthropogenic climate change, as defined by the United Nations Framework Convention on Climate Change (UNFCCC) in its Article 1 (Sands, 1992): a change of climate which is attributed directly or indirectly to human activity that alters the composition of the global atmosphere and which is in addition to natural climate variability observed over comparable time periods.
Despite recent advances in our understanding of the impacts of ongoing anthropogenic climate change on TMFs, concrete empirical evidence (i.e. compelling proof of climate change-driven impacts occurring in real time) remains scarce, partly because such assessments require long-term data that are difficult to collect and analyse (Wauchope et al., 2021). For instance, it has been observed that plant communities can shift their distributions, but these movements lag behind the velocity of climate change (Feeley et al., 2011;Corlett & Westcott, 2013;S aenz-Romero et al., 2016). Therefore, key ecosystems like TMFs help advance our knowledge of climate-driven distributional change because the microclimatic variability and island-like distribution of mountainous biomes not only results in high levels of endemism, but also high extinction vulnerability (Freeman et al., 2018), making them natural laboratories for climate change (Silveira et al., 2019;Tito, Vasconcelos & Feeley, 2020).
Numerous literature reviews on TMF research have been conducted, focusing on various aspects, including environmental determinants (Oliveira et al., 2014;Fahey, Sherman & Tanner, 2016), ecosystem functions (Dalling et al., 2016), ecosystem services (Buytaert, Cuesta- Tropical montane forests and climate change conservation (Peh et al., 2011) and restoration strategies (Christmann & Menor, 2021), specific geographical regions (Rosas Rangel et al., 2019;S aenz-Romero et al., 2020a;Tovar et al., 2022) or taxonomic groups (Gotsch, Nadkarni & Amici, 2016;He, He & Hyvönen, 2016), and use of remote sensing technologies (Altarez, Apan & Maraseni, 2022). Our study goes beyond a systematic review on the effects of climate change on TMFs by critically evaluating the quality of the available evidence. We applied a replicable evidence-appraisal methodology to synthesise the current knowledge of the impacts of climate change on TMFs, with a focus on the published studies yielding the strongest evidence. This will serve as a guide to identify knowledge gaps and inform future research on this highly threatened biome.
Specifically, our research questions are: (i) what are the general thematic and methodological trends in the published studies investigating the impacts of climate change on TMFs? (ii) What are the implications of those trends in terms of our understanding and application of the current knowledge? (iii) How reliable is the literature in terms of the quality of the evidence it provides? (iv) What are the general forecasts for TMFs globally based on the most reliable evidence and the direction of future research?

II. METHODS
(1) Systematic review We carried out a literature search on the effects of climate change on TMFs following a standard systematic review protocol (PRISMA-P; Shamseer et al., 2015). The search terms were based on the most common names of TMFs and terms associated with long-term changes in the climate. The final search terms used were '(trop* monta* forest* OR cloud forest* OR trop* high* elevation* forest* OR trop* mid* elevation* forest* OR trop* mount* forest*) AND (climat* chang* OR glob* warm* OR temperature ris* OR clima* vari*). ' We conducted the literature search in July 2022 using four academic databases (Web of Science, Scopus, ScienceDirect and Google Scholar) and the Google search engine for the period 1994-2021. Relevant studies published during 2022 were omitted from the statistical analysis but are included in the discussion. The search string was run choosing the 'all fields' option if available, hits were sorted alphabetically, and relevant publicationsi.e. studies that explicitly reported observed or projected effects of anthropogenic climate change on TMFs (or any component thereof)were chosen based on title and abstract. Duplicated records were removed, and studies that mentioned climate change only as a potential future threat but did not investigate its effects on TMFs were excluded. Likewise, studies that described diversity patterns along elevational gradients but did not link them to environmental factors to allow inferences of potential future shifts were also discarded ( Fig. 1). For Google searches, only the first 100 hits were reviewed to ensure that no relevant studies had been missed. A search in Spanish was conducted in the Redalyc database (www.redalyc.org), which did not yield any relevant studies, however, publications in Spanish and French listed under an English title, and thus captured by the search string, were retained. Since we aimed to rank the quality of the available scientific evidence, only peer-reviewed primary research articles were retained (see Section II.2). However, we checked reviews for 'snowballing' purposes (i.e. pursuing references of references; see Greenhalgh & Peacock, 2005). We excluded book chapters, reports and other grey literature. Palaeoclimatic and palaeoecological studies were excluded because our focus was on current anthropogenic climate change only.
We extracted from individual studies information on study region, country, methodology, study duration, publication date, ecological level, taxonomic group, research topic, as well as measures of climate change and their observed effects in TMFs. We considered patterns of changing climate (rising temperature and increased or reduced precipitation) as binary variables (i.e. whether these measures were reported or tested in the study or not), and the observed impacts as categorical variables: occurrence of extreme events [strong El Niño Southern Oscillation (ENSO), cyclones, fires, frost, landslides], species' distributional changes [habitat losses (range contractions or fragmentations), and upslope or downslope shifts], impacts on biodiversity (increased vulnerability, population decline, extinctions, reduced genetic diversity, invasion of neighbouring communities and lowland biotic attrition), and other local-scale effects (alterations to carbon and other nutrient fluxes and stocks, soil functions, and phenological/physiological patterns).
To elucidate general trends in TMF research related to climate change, we investigated if studies were skewed towards a particular geographical region, taxonomic group, ecological level, or research topic; and if there were any associations between these parameters. We also investigated if studies reporting or testing an impact on TMFs tend to look at a particular measure of climate change or taxonomic group in their research. Only categories represented by at least 10 studies were analysed using chi-squared tests to avoid inferring spurious associations. These associations should be interpreted with caution because they are based solely on what individual studies reported (observations or historic records) or tested (experimentally or through modelling). Statistical tests were carried out in R 4.1.1, using ggplot2 v.3.3.5 (Wickham, 2011) and corrplot v.090 (Wei et al., 2017) packages.
(2) Evaluation of evidence quality The evidence assessment tool devised by Mupepele et al. (2016) is designed to rank the quality of evidence provided by individual reviews and studies based primarily on their study designs. Studies that are poorly conducted (e.g. unclear research questions, inadequate sample sizes, lack of controls, etc.) are downgraded in the evidence hierarchy. The discrete categories of evidence quality and assessment criteria of the evidence assessment tool were specifically designed for conservation studies, but can be adapted for other fields of research. Our evidence assessment was an adaptation of Mupepele et al. (2016). We removed the 'review' (systematic and conventional) category, ranked highest by Mupepele et al. (2016), as its inclusion would result in double-counting of studies (i.e. studies captured by our search string are likely to occur in other reviews; Fig. 2). In addition, we carried out a methodological appraisal of each study. Studies with flaws and biases identified in terms of data collection and analyses or that employed outdated methodologies (mainly applicable to modelling studies) were downgraded to a lower level in the evidence hierarchy. Studies that combined multiple methodological approaches were classified according to their highest level in the evidence hierarchy [e.g. a study with both ex-situ experiments and modelling would be ranked as Level of Evidence 1b (LoE1b); Fig. 2, Table 1].
To define the hierarchical LoEs, we considered the following features: use of references and controls, execution of the study in the field (i.e. in-situ), long-term collection of data, high regularity of survey, and corroboration of findings derived from models with empirical observations (Table 1). Experimental study designs with a control carried out in the field provide very strong evidence (LoE1a). Experiments conducted ex situ yield only strong evidence (LoE1b) because artificial environments introduce potential biases to the study (Tito et al., 2020). Observational studies spanning a minimum of 10 years with observations at regular intervals (weekly, monthly, etc.) also provide strong evidence by accounting for the long-term nature of climate patterns (Bruijnzeel, 2004;Chapman et al., 2018), and are included in LoE1b. We prioritised the highest-ranked study design in studies that had employed both an ex-situ experiment and a modelling approach and therefore assigned these to LoE1b. None of the studies captured by our search string had conducted both field experiments and modelling.
LoE2 comprises short (<10 years) longitudinal and all cross-sectional observational studies, as well as the bulk of studies that relied on computer-based simulations (hereafter 'modelling'), i.e. projections of future climate conditions and especially ecological niche models, which include species distributions, life zones and relative abundance projections. Non-modelling studies are subdivided into <10-year long longitudinal studies (LoE2a, moderate evidence), sequential cross-sectional studies, i.e. resurveys (LoE2b, inconclusive evidence), and all other crosssectional studies (LoE2c, circumstantial evidence), including comparative surveys (between different sites) and along altitudinal or latitudinal gradients (space-for-time substitution approach).
Given the requirement for long-term data in climate change research and the urgency to implement wellinformed policies to preserve ecosystems from anthropogenic  Table 1 for detailed explanation of levels.  Fig. 2). In principle, modelling approaches using ideal data and next-generation tools that are still under development (the 'aspirational' gold standard) is unattainable for predictive climate change studies and is therefore not considered in our evidence quality scale. The silver standard corresponds to best modelling practices, i.e. a combination of the best-available data and tools to account for or quantify bias and uncertainty (LoE2a). The bronze standard implies limited acceptable practices that allow inferring implications from their results (LoE2b,c). Lastly, the deficient category corresponds to insufficiently robust data or modelling practices (LoE3), and this level also comprises all speculative studies, i.e., not supported by empirical data. These include expert opinions and speculations based on knowledge of other regions or ecosystem types.

III. RESULTS
(1) Distribution of studies Our initial search string retrieved 10,163 studies. After successive elimination stages ( Fig. 1), the final data set included 395 studies (see online Supporting Information, Table S1 for full list), published between 1994 and 2021 ( Fig. 1). The number of publications increased from less than 10 annually before 2006 to an average of 30 from 2013 onwards, potentially reflecting an increased interest in TMFs.
Inconsistent definitions of TMFs, ambiguous latitudinal limits and different mapping methods have resulted in estimates of their global distribution and extent that vary by as much as an order of magnitude. However, an estimate of 2M km 2 globally (Mulligan, 2010;Los et al., 2019) has been reported relatively consistently, and is compatible with other recent regional estimates (e.g. 401,300 km 2 TMF cover in the Neotropics; Helmer et al., 2019). Around 75 nations are thought to have some TMF cover, however, there is no consensus regarding many countries such as Bhutan, East Timor, Eritrea, Eswatini, Lesotho, Nepal, Oman, Somalia, South Africa, South Sudan and Sudan (Fig. 3), and some authors count subnational territories separately, e.g. New Caledonia, Tahiti and La Réunion Island (see Karger et al., 2021;Mulligan, 2010).

Tropical montane forests and climate change
The vast majority of studies took place in tropical regions (0 to 23.5 N and S), whereas exclusively subtropical (23.5 to 30 N and S; sensu Corlett, 2013) studies were only those conducted in the Canary Islands (Fig. 4A). Some studies included tropical-subtropical overlapping regions, mainly in Taiwan, Argentina and Brazil.
(2) Trends in the literature We found that research from certain regions tended to focus on a particular taxonomic group (χ 2 = 129.77, df = 81, P < 0.001). Research efforts in North America (i.e. Mexico) were skewed towards trees; in Oceania (mainly in the Wet Tropics Bioregion, northeastern Australia) were skewed towards insects; and in Central America (mainly in Costa Rica and Panama) towards both bryophytes and fungi. Conversely, studies on trees were scarce in Oceania, with only one study from Papua New Guinea (Venter et al., 2017).
Some research topics were associated with certain geographic regions (χ 2 = 104.25, df = 45, P < 0.001). For example, studies in the Caribbean tended to focus on the effects of changing abiotic conditions, such as changes in fog immersion, air temperature and streamflow. We also found an association between research topics and taxonomic groups (χ 2 = 268.83, df = 45, P < 0.001). This was mainly driven by studies on ecosystem functions whose focus was on trees and bacteria. Tree studies focused on large-scale processes such as primary productivity, carbon sequestration and distributional shifts, whereas bacteria studies mainly examined soil functions, such as nutrient cycling and decomposition. There was also a dearth of diversity (e.g. abundances and community composition) studies on trees, compared to other taxa, such as birds and insects.
We did not find evidence that studies reporting or testing climate change impacts tend to use a particular measure of climate change or consider a particular taxonomic group in their research (i.e. no association between focus on climate change measures and on taxonomic groups). Studies reporting habitat losses (range contractions or fragmentations) focused more frequently on mammals (χ 2 = 41.07, df = 10, P < 0.001), and were also more likely to report biodiversity losses in terms of abundances, species richness or species turnover, and extinctions (χ 2 = 10.32, df = 4, P = 0.035).
(3) Assessment of the evidence on the impacts of climate change in TMFs We found 30 (7.6%) field experimental studies (Fig. 5A), i.e. providing a 'very strong' LoE (LoE1a). They reflected the general geographical trend described in Section III.1, with five studies from Mexico, although four of these were conducted by one research team (García-Hern andez et al., 2019; Toledo-Aceves, García-Hern andez & Paz, 2019; García-Hern andez & Toledo-Aceves, 2020; Toledo-Aceves & del-Val, 2021). Four LoE1a studies were from Peru, followed by three each from China, Costa Rica and Hawaii, two each from Puerto Rico, Taiwan and Tanzania, and one study for Colombia, Ecuador, India, Panama, the Philippines and Rwanda. The scarcity of field experimental studies could be due to the relative inaccessibility of tropical montane regions, often in poorly connected rural areas of low-income countries, which makes field experiments impractical and costly in the absence of wellestablished research groups. By contrast, strong evidence (LoE1b; ex-situ experiments and long-term data sets), was found in 94 studies (23.8%). Around 13% of these studies (12, 3% of total) included the use of modelling methods. Studies that scored as moderate evidence (LoE2a) accounted for 58 studies (14.7%), including mainly <10 year data sets and a small contribution of modelling studies (6, 1% of total) that supported their forecasts with field observations. A third of the reviewed studies provided inconclusive evidence (LoE2b; 132 studies). The reliance of climate change studies on modelling methods was evident: modelling studies accounted for 81.8% (108, 27.3% of total) of studies in this level. The remaining studies in LoE2b were non-longitudinal observational resurveys that provided two or more snapshots suggesting temporal changes but did not allow identification of trends. Studies yielding circumstantial evidence (LoE2c) were common (73, 18.5%), mostly consisting of crosssectional observational studies, and a small contribution from modelling approaches (11, 2.8% of total). Finally, speculative studies (LoE3a) were the least frequent (8, 2.0%).
Modelling methods were used by 144 (36.5%) studies in total. Climatic envelope models [mainly species distribution models (SDMs), as well as projections of species' population decline, extirpation or extinction under climate change] comprised over two thirds (102, 70.8%) of these; 20 (13.8%) were future climate projections (mainly estimations of future temperature and precipitation regimes), and the remainder (22, 15.3%) were a diverse array of computer-based simulations of changes in biomass, evapotranspiration, albedo, erosion, water runoff, etc. From 1994 to 2005, we found 14 studies that employed modelling methods, with no more than three studies annually, and zero modelling studies for some years, including 2005. Of these, only two were SDMs (Williams, Bolitho & Fox, 2003;Miles, Grainger & Phillips, 2004). An association between the LoE strength and research topics (χ 2 = 116.11, df = 20, P < 0.001; Fig. 5B) revealed some methodological trends. This association partly was an artefact of the ranking of most modelling studies in LoE2b (i.e. distributional shifts as a main research topic was strongly associated with LoE2b). However, the association persisted after removing distributional studies (χ 2 = 68.45, df = 16, P < 0.001) because studies on biodiversity also relied more Biological Reviews 98 (2023)  Tropical montane forests and climate change on LoE2b approaches that yield inconclusive evidence (modelling and resurveys). Conversely, studies on ecosystem functions contained the most study designs providing strong evidence (LoE1b, experiments and long-term dendrochronological records).
Considering only the studies at species, community, and ecosystem or landscape levels (i.e. ecological levels with >100 studies each; Fig. 5C), we found an association of these studies with certain LoE strength (χ 2 = 23.44, df = 10, P = 0.009). Whilst provision of inconclusive evidence (LoE2b) was most frequent in studies at these three ecological levels, the species and ecosystem or landscape levels also included a non-negligible proportion of studies yielding strong evidence (LoE1b). For the species level, these corresponded mainly to ex-situ experiments (e.g. thermal and drought tolerance experiments). For the ecosystem or landscape level, these were based on long-term data sets.
Regarding taxonomic groups, trees and other vascular plants, birds, amphibians, and insects were the only taxa represented in more than 20 studies each (Fig. 5D). By retaining only these groups and grouping together the remaining taxa, we found that studies of some taxonomic groups were associated with certain LoE strength (χ 2 = 79.49, df = 20, P < 0.001). Studies of trees commonly provided evidence ranging from strong (LoE1b) to inconclusive (LoE2b), whereas studies of birds were associated more often with inconclusive evidence (LoE2b, modelling studies) and those of insects more with circumstantial evidence (LoE2c, comparative cross-sectional surveys). Very strong evidence (LoE1a) was provided most often in studies of other taxa (i.e. fungi, mammals, bacteria, reptiles, bryophytes, lichens, and others), driven by the use of field transplant experiments in studies of fungi, bryophytes, lichens, and bacterial (soil) communities.
(4) Evidence-based synthesis of impacts of climate change on TMFs Ranking the published studies according to LoE strength allows us to outline better the current state of knowledge on the present or expected future impacts of climate change on TMFs. In this section, we summarise the findings on climate  Table 1  change impacts on the abiotic environment, biodiversity and ecosystem functions of TMFs, focusing primarily on the studies ranked highest in our evidence hierarchy in each case.

(a) Changes in atmospheric conditions
None of the published studies on atmospheric conditions employed field experiments (i.e. no LoE1a studies). Albeit scarce (7, 1.8%), evidence based on long-term studies (≥10-year observations; LoE1b) suggested that atmospheric changes in TMFs were a local or regional effect, rather than a global trend. A study analysing 100 years of meteorological data in Mexico showed that the subtropical (Nearctic) northern mountainous regions have experienced more consistent increments of both atmospheric temperature and precipitation since 1970 than the (Neo)tropical southern mountainous areas (Cuervo-Robayo et al., 2020). Elsewhere, significant reductions in precipitation have been reported over recent decades (e.g. the Indian Western Ghats region; Murugan et al., 2009). However, a decrease in rainfall does not necessarily affect the capacity of TMFs to intercept water from the atmosphere (e.g. La Hispaniola; Comarazamy et al., 2015). TMFs in Puerto Rico had a higher likelihood of fog immersion during the dry periods of the year compared to the rainy season (Van Beusekom, Gonz alez & Scholl, 2017). In addition, over 40 years of observations in Puerto Rico revealed no significant change of cloud base levels in the mountains (Miller et al., 2018). Therefore, these data suggest that water availability within TMFs may not be adversely affected by reductions in precipitation.

(c) Distributional effects on flora
Atmospheric variables play a key role in shaping the distributions of species and whole communities, thus shifts in species' ranges are one of the expected consequences of climate change. The presence of clear boundaries between forests and other vegetation types (treelines) is primarily defined by temperature, precipitation and fog incidence, and has been confirmed in various tropical montane regions, including the Cordillera Central in the Dominican Republic (Martin, Sherman & Fahey, 2007;Martin & Fahey, 2014), the Afromontane forests in Ethiopia (Schmitt et al., 2013), and in Hawaii (Crausbay & Hotchkiss, 2010). Upslope displacements of treelines have been observed over periods of 10 years or more (LoE1b) in Mexico (Jiménez-García et al., 2021), protected areas in the tropical Andes (Lutz, Powell & Silman, 2013), Taiwan (Greenwood et al., 2014), Hawaii (Koide et al., 2017) and Mount Kilimanjaro (Shugart et al., 2001), as well as changes in community composition that reflect upslope migrations of lowland plant species in Costa Rica (Feeley et al., 2013).
Yet, many species' responses to spatial changes in temperature and precipitation may not be occurring at a sufficiently fast pace to keep up with the rate of climate change (Feeley et al., 2013(Feeley et al., , 2011Lutz et al., 2013; but see Lu et al., 2020). An analysis on avian seed dispersal in the Peruvian Andes concluded that several long-distance dispersal events would be necessary for the treeline to keep up with warming rates (Nowak et al., 2022). And even if that dispersal occurs, experimental studies have found that species-specific thermal and drought tolerances might influence seedling recruitment rates at higher altitudes (Esper on-Rodríguez & Barradas, 2014; Rehm & Feeley, 2016;Fadrique et al., 2018), potentially stymieing treeline expansion. However, an analysis of historic data of Taiwanese montane trees showed dissimilar responses at both intra-and interspecific levels; species already adapted to higher elevations moved upslope at higher rates, but these responses varied among life stages (O'Sullivan et al., 2021).
The main bulk of evidence of distributional changes comes from forecasting modelling studies (LoE2b), which Biological Reviews 98 (2023)  Tropical montane forests and climate change overwhelmingly project range contractions, population declines, local extinctions or a combination of these in TMF tree communities (e.g. John et al., 2020; Neto dos Santos, Silva & Higuchi, 2020;Rojas Briceño et al., 2020) and herbaceous plants (Setyawan et al., 2020). These predicted range contractions are partially explained by the topography of mountains themselves because as species migrate upwards, the available area decreases. However, upward area reduction does not happen monotonically in over half of the world's mountainous regions (Elsen & Tingley, 2015), and the influence of topography is complex, in some cases potentially leading to horizontal rather than vertical displacements (Lippok et al., 2014).
Several studies have looked at the potential limiting factors for treelines to track new climatically suitable areas. These factors include frost (Rehm & Feeley, 2015;Joshi, Ratnam & Sankaran, 2020;Rehm, Yelenik & D'Antonio, 2021), hydraulic stress (Song et al., 2016b), fruit or seed production (Chapman et al., 2018), seed dispersal (Hillyer & Silman, 2010;Rehm & Feeley, 2013;Nowak et al., 2022), germination rates (Centre et al., 2016), and even the absence of nurse plants (Soto-Correa et al., 2013). By contrast, a field transplant experiment carried out in the Peruvian Andes concluded that soil was not a limiting factor for the establishment of trees at higher elevations (Tito, Vasconcelos & Feeley, 2021). It has been suggested that plants that have evolved in nutrient-poor soils might be 'pre-adapted' to cope with other environmental stressors (Whitman et al., 2021). Thus, intra-and interspecific differential migration rates, in combination with other processes such as higher mortality of cold-resistant species and intrusion of lowland species into montane areas (e.g. de Gasper et al., 2021), may lead to the formation of new communities adapted to warmer regimes (Wright, Muller-Landau & Schipper, 2009), i.e. thermophilisation (Duque et al., 2015;Fadrique et al., 2018).
While experimental studies to test responses of tree species to climate change rarely went beyond seed dispersal and seedling establishment, another defining component of the TMF flora, epiphytes, has been researched more often through experimental manipulations (LoE1a and LoE1b). Field transplant experiments to different elevations to simulate changing climatic regimes on bryophytes (Nadkarni & Solano, 2002;Song, Liu & Nadkarni, 2012;Wagner, Zotz & Bader, 2014) and ferns (Hsu, Oostermeijer & Wolf, 2014) consistently found slower rates of growth and leaf production, and higher mortality, even if some species or individuals displayed some plasticity. Similar results were obtained from ex-situ experiments with both bryophytes and vascular epiphytes (Zotz et al., 2010;Gotsch et al., 2015). Epiphytes' reliance on different water sources seems to be linked to their taxonomic affiliation; Liu et al. (2021) found that in a Chinese subtropical montane forest, bryophytes and ferns obtained water both from humus and fog, whereas lichens and seed plants relied almost exclusively on fog. Although significant tolerance to desiccation (Bader et al., 2013) and temperature rise (up to an average of 3 C) has been observed for some epiphytic species (Müller, Albach & Zotz, 2017), their ability to track new climatically suitable areas was not experimentally tested.

(d) Effects on fauna
Birds were the best-studied taxonomic group after all vascular plants (Fig. 4D). There was some empirical evidence (LoE1a and LoE1b) for the effects of climate change on avian species, such as recorded cases of upslope migrations in Honduras (Neate-Clegg et al., 2018), the tropical Andes (Forero-Medina et al., 2011b;Hermes et al., 2018a;Hayes, Lecourt & del Castillo, 2018;Freeman et al., 2018), Tanzania (Neate-Clegg et al., 2021b) and New Guinea (Freeman, 2016). Elevational shifts have also been reported for moth assemblages on Mount Kinabalu, Malaysia (Chen et al., 2009), and bats in Costa Rica (LaVal, 2004). However, these responses might be species specific (Anderson et al., 2013) and not all species can successfully expand their ranges Neate-Clegg et al., 2020). Even if elevational shifts do occur, they may result in intense competition for space and resources near mountaintops, triggering aggressive behaviours (Jankowski, Robinson & Levey, 2010), or leading to higher morbidity (Freed & Cann, 2013) and mortality rates (Shiao et al., 2020).
Modelling studies (LoE2b) outline similar trends, overwhelmingly predicting range contractions, population declines, local extinctions or a combination of these in TMF birds (Colyn et al., 2020), mammals Raman et al., 2020a,b), amphibians (Cruz-Elizalde et al., 2020;Cordier et al., 2020) and scorpions (de Araujo-Lira et al., 2020). However, the influence of climate change on upslope migrations could be challenging to distinguish from that of land-use change (see Jacob et al., 2015a,b).
Empirical evidence showing effects of climate change on other major animal taxa, such as mammals, reptiles and most invertebrates, is too limited to identify any clear patterns. For instance, a resurvey of tropical montane ants in Costa Rica concluded that over a decade, the community became less diverse, with upland areas becoming more similar to lowland ones (Warne et al., 2020) suggesting that thermophilisation of communities is not limited to florahowever, this finding was promptly contested (Klimes et al., 2021).

(e) Effects on ecosystem functions
Ecosystem functions are commonly studied through experimental manipulation in the field (LoE1a) or laboratory (LoE1b). Multiple soil transplant and litter decomposition experiments have found that increasing temperatures, altered water status or both can change decomposition rates and have a negative influence on the capacity of TMF soils to retain organic matter, potentially turning the systems into carbon emitters (e.g. Becker & Kuzyakov, 2018;Looby & Treseder, 2018;Nottingham et al., 2019bNottingham et al., , 2016. Higher temperatures also make nutrients more readily available, with potential cascading effects on vegetation and other soil properties (Dantas de Paula et al., 2021). Some studies, however, have reached opposite conclusions or found no clear relationship between temperature or hydric regime and soil properties (e.g. He et al., 2010;Scowcroft, Turner & Vitousek, 2000). Such equivocal conclusions could be due to the heterogeneity and localised nature of soil properties.

IV. DISCUSSION
This review of the literature on impacts of climate change on TMFs shows that: (i) the rates of climate change are generally intensified rather than attenuated by elevation, making montane communities more susceptible to their effects; (ii) tropical montane communities might be able to respond by shifting their distributions primarily upslope, but (iii) not all species seem able to shift their distributions and the factors preventing them from tracking or establishing in new climatically suitable areas are unknown or not well understood; (iv) even if montane species are able to track suitable conditions fast enough, they risk running out of physical space; (v) the impacts of climate change on genetic diversity and species interactions within tropical montane ecosystems remain largely unknown; and (vi) the loss of biodiversity and functions of TMFs could result in the loss of valuable ecosystem services for human populations living close to tropical montane regions, with repercussions at a broader scale.
Additionally, this study revealed some significant knowledge gaps in several aspects (i.e. methodological approaches, geographical and taxonomic skews, and research topics) that need to be addressed, but also shows that there are some areas of opportunity, either expanding on the available knowledge, or by employing methodologies and data sources that have not been properly explored.
(1) Methodological gaps and opportunities The nature of climate change as a global and long-term phenomenon limits our ability to produce in the short term abundant empirical evidence of its effects in real time on particular ecosystem types or any of their functions and biotic components. Instead, our assessment of evidence strength showed that there has been greater reliance on study designs that yield moderate, inconclusive and circumstantial evidence. Given that conventional study designs that yield strong evidence tend to be time-consuming, effort-intensive and costly in remote mountainous regions, a practical strategy is to accumulate independent lines of moderate, inconclusive or circumstantial, yet coherent evidence that build up the same narrative. Thus, greater effort is needed to reconcile contradicting findings across different study sites and spatiotemporal scales, as well as attempts to disentangle the synergistic influences of multiple environmental factors on Biological Reviews 98 (2023)  the diversity and functions of TMFs. Albeit likely geographically biassed and difficult to interpret, long-term data sets are becoming increasingly abundant and accessible (Wauchope et al., 2021).
Additionally, environmental gradients along mountain slopes make TMFs ideal locations to conduct field manipulation experiments (e.g. transplant experiments along temperature gradients). Field transplant experiments are recognised as a powerful tool capable of yielding robust evidence by replicating complex projected environmental conditions more accurately than laboratory trials (Nooten & Andrew, 2016;Silveira et al., 2019;Tito et al., 2020). These types of experiments are useful to inform assisted migration programmes (e.g. Castellanos-Acuña, Lindig-Cisneros & S aenz-Romero, 2015; S aenz-Romero et al., 2020b), changes in ecological networks (e.g. Maunsell et al., 2015), adaptation to urbanisation (e.g. Martin et al., 2021), among others. Alternative methods to produce strong evidence in the short term include analysis of historical remote sensing data (e.g. aerial photography and satellite imagery to detect changes in ecosystem boundaries and canopy spectral changes over time), as well as 'natural' long-term records (e.g. dendrochronological studies; Rodríguez- Ramírez et al., 2022).
Even though our review focuses exclusively on anthropogenic climate change, the potential value of palaeoclimatic and palaeoecological studies cannot be disregarded. In principle, studies of Quaternary-time and deep-time face similar challenges as long-term future-modelling studies, namely: 'the impossibility of distinguishing between true and false' (Biondi, 2014, p. 1), and their findings should be interpreted cautiously. For instance, Fitzpatrick et al. (2018) projected that by 2090 (i.e. seven decades), climates in North America will have shifted by as much as they did during the past 13,000 years. Hence, it seems unreasonable to expect that species will be able to replicate in just a few decades past migrations spanning millennia. Conversely, a few decades might simply be too little time to detect distributional changes, especially for long-living, slow-growing organisms such as trees than can live for centuries. In fact, a common criticism of ecological niche modelling approaches is their underlying assumption that present distributions reflect the whole set of conditions in which a species can persist (Feeley & Silman, 2010a;Sax, Early & Bellemare, 2013), which is not necessarily true and needs to be accounted for. Some studies warn that relying exclusively on 'realised distributions' as input for predictive distributional studies could overly restrict potential future suitable habitats and overestimate risks of extinction and extirpation (Veloz et al., 2012;Sax et al., 2013). However, we argue that given the current rate of anthropogenic climate change, it is preferable to avoid overly optimistic assumptions that may lead to inaction, especially for montane ecosystems globally. Moreover, sets of good practices have been suggested to improve the accuracy of palaeoecological reconstructions (e.g. Nogués-Bravo, 2009) and a combination of short-term ecological studies with long-term palaeoecological evidence can help us to understand the impacts of climate change better (Lamentowicz, Słowi nska & Słowi nski, 2016).
These research strategies would help resolve conflicting lines of evidence to enable rapid preventive and adaptive responses to climate change impacts on TMFs.
(2) Geographical and taxonomic gaps and opportunities Studies have been heavily concentrated in the Mesoamerican and Andean regions, both part of the Neotropical biogeographic realm. The fragmented evidence from other world regions suggests that other tropical montane regions may share similar climate change-induced impacts, albeit with some degree of local variation. Research efforts should be refocused on understudied regions to find out if there are any major discrepancies among them. Nonetheless, a few intensively researched areas, such as TMFs in Mexico, Costa Rica and Perurepresentative of subtropical, mid-latitude tropical and nigh-equatorial TMFs, respectively can be considered suitable proxies for environmental management while local studies elsewhere are in progress.
Most research on TMFs focuses on only a few taxa, yet these ecosystems are considered hotspots of biodiversity, much of which remains undescribed. For example, a recent survey of rove beetles (Staphylinidae), one of the largest families of organisms in the world, along an elevational gradient in Honduras found that they reached peak diversity precisely in the highly vulnerable TMF altitudinal belt (Dolson et al., 2021). Such lack of knowledge on biotic components of TMFs obscures our understanding of their ecological networks, ecosystem functions, and the magnitude of potential losses if cascading extinctions occur. Fortunately, surveys of soil, understorey and canopy biota can be carried out relatively quickly and are less costly and effort intensive than long-term monitoring or manipulative study designs. Additionally, knowledge biases towards charismatic taxa can be exploited to set up 'umbrella species' conservation schemes.

(3) Thematic gaps and opportunities
Many studies support the notion that climate change will result in physiological pressures and distributional shifts of tropical montane communities, but assumptions of general climate-driven range shifts should be avoided (Rubenstein et al., 2020). Empirical evidence also shows that climatic conditions can impede the effective establishment of tree communities (Rehm & Feeley, 2013;Song et al., 2016b;Joshi et al., 2020) and likely other components of tropical montane biota. In fact, species' ability to persist or migrate is influenced by their interactions with other ecosystem components, both biotic and abiotic Ramirez-Villegas et al., 2014;Quiroga, Premoli & Kitzberger, 2018;Joshi et al., 2020), but few studies have looked at ecological networks in TMFs (Benning et al., 2002;Jankowski et al., 2010;Hillyer & Silman, 2010;Ornelas, Licona-Vera & Ortiz-Rodriguez, 2018;Tito et al., 2021). Improving our understanding of these interactions would help improve the accuracy of forecasts both in terms of distributional responses and potential future assemblages. Although field studies are needed to elucidate how tropical montane networks respond to climate change and other disturbances, existing databases can be used to construct and conduct robust analyses on ecological networks (e.g. de Almeida & Mikich, 2018;Fricke & Svenning, 2020), and project their responses under climate change scenarios.
More concerning is the paucity of studies at the genetic level in TMF research in relation to climate change, which has been previously acknowledged (Pauls et al., 2013). For example, tropical tree populations are experiencing genetic bottlenecks following intense disturbance events, but tropical montane regions are understudied (Pautasso, 2009). For mountainous regions, this might be crucial because microevolutionary processes operate differently within a population along an elevational gradient, i.e. the leading edge, the central population and the rear edge (Kremer, Potts & Delzon, 2014). Sudden disturbance-induced migrations may lead to decreased phenotypic variability, further jeopardising the plasticity and ability of trees both to reach and to establish populations in new areas (Pertoldi, Bijlsma & Loeschcke, 2007). Also, highly variable conditions may not result in an adaptive response because selection processes are multi-directional and the existing genetic variation in a population might be insufficient to generate the genotypic combinations required for it to persist under new environmental conditions (Alfaro et al., 2014). As the most conspicuous biotic component of forests, declines of trees could trigger negative cascading effects (Bawa & Dayanandan, 1998;Nagel et al., 2019). Thus, the genetic status of trees is a factor that ought to be taken into account for assisted migration programmes (Alfaro et al., 2014), and all these considerations are equally valid for other taxa. Unlike other knowledge gaps, the scarcity of studies at the genetic level is difficult to overcome though indirect and remote methods, however, the increasing accessibility and affordability of sequencing methods should facilitate extensive genetic surveys of tropical montane populations in the short term.

V. RECENT WORK
There have been recent developments since our systematic review was carried out. For example, Rico et al. (2023) reported that the genetic health of a tropical montane tree species in Mexico is threatened by human activity and climate change. Research on treeline expansion has also shown that dry conditions in the boundary between TMF and p aramo in the Venezuelan Andes support a seedling bank, but with slow growth ; and some species may need treatment, such as scarification, to increase germination rates (Liyanage et al., 2022). Thus, understanding the genetic adaptations and environmental influence of seeds and seedlings is crucial for conservation efforts. However, long-term monitoring is also necessary as some abiotic effects, such as soil properties, may not be noticeable for years (Martínez-Ramos et al., 2022). Lastly, new knowledge of diversity patterns and spatial partitioning in various regions (Berrios, Coronado & Marsico, 2022;Morton et al., 2022) emphasise the importance of traditional biodiversity surveys in TMFs, especially in understudied regions.

VI. CONCLUSIONS
(1) We highlight the long-term need to widen the methodological, thematic, taxonomic and geographical scope of studies on TMFs under climate change. In the short term, however, the accumulation of moderate to circumstantial evidence constitutes the most accessible and reliable tool to address uncertainties and gaps in current knowledge. As such, in-depth research in well-studied regions, use of alternative data sources (remote sensing and 'natural' longterm records), palaeoecological supporting evidence and advances and refinements of forecasting modelling techniques offer the most reliable and immediate sources of information for expeditious conservation action for these threatened forests.
(2) Natural variability within TMF regions represents a challenge for the generalisation of the findings of individual studies, but it simultaneously represents an exceptional opportunity to generate high-quality evidence of the impacts of climate change for both tropical montane species and lowland species. Environmental gradients along mountain slopes have been identified as natural laboratories, where field manipulation experiments (e.g. field transplant experiments, rain exclusion experiments, etc.) can be conducted in the short term to simulate complex projected environmental conditions with greater accuracy than can be achieved ex situ.
(3) We highlight the importance of modelling approaches in TMF research and encourage further refinement and development of these methods. To enhance their effectiveness, novel, more robust forecasting algorithms should be developed to account for uncertainties and sampling biases. Additionally, incorporating ecological information, such as species dispersal limitations, biotic interactions, and analyses of ecological networks, can make commonly used modelling approaches more informative. Optimising these approaches with more ecological information is crucial for the success of conservation strategies, such as the design of protected areas that consider future suitable habitats for whole biotic communities, and minimising losses of biodiversity and ecosystem functions.
(4) Despite the undeniable importance of trees in forests, the responses of other taxonomic groups to climate change should not be overlooked, as intraspecific interactions could prove decisive for the success of conservation measures. Similarly, the impacts of climate change at the genetic level remain largely unknown and the loss of genetic diversity can threaten the long-term viability of TMF populations. Tropical montane forests and climate change (5) We urge scientists to conduct similar evidence quality assessments in their respective fields. Experts in each area of research should critically ponder what study designs and data sources yield the most robust body of evidence and take them into consideration when carrying out reviews and planning future research.

VII. ACKNOWLEDGEMENTS
This study was funded by the Mexican National Council of Science and Technology (CONACyT) and the University of Southampton. The authors have no conflict of interest to declare.

VIII. REFERENCES
References marked with asterisk (*) have been cited within the supporting information. Tropical montane forests and climate change