The case for an epiphyte area index

Abstract

Three distinct ecosystems where epiphytes contribute a substantial vegetative area in the canopy.

Disclaimer

The New Phytologist Foundation remains neutral with regard to jurisdictional claims in maps and in any institutional affiliations.

Introduction

Leaf area index (LAI) is the foundational metric for quantifying canopy light interception, photosynthesis, and transpiration across scales from individual crowns to the globe (Park & Jeong, 2021). Structurally defined as leaf area per unit ground area, LAI is operationally retrieved through radiative transfer: to estimate the effective area of light‐interacting surfaces, satellite products (e.g. MODIS MOD15) invert observed light reflectance above the canopy, while field instruments (e.g. ceptometers) convert light transmittance through the canopy (Kucharik et al., 1998; Jonckheere et al., 2004; Fang et al., 2019). Because these retrievals register all light‐intercepting surfaces (leaves, bark, lichens, etc.) many operational ‘LAI’ measurements are more accurately interpreted as an effective plant area index (PAI). What registers as ‘LAI’ is thus determined by the optical properties of vegetation elements; properties that vary with physiological state, including water content and pigment concentrations (Ollinger, 2011; Wong, 2023). This radiative dependence is explicit in retrieval algorithms. The MODIS LAI/FPAR algorithm treats canopy optics as functions of hydration and Chl (Knyazikhin et al., 1998), while vegetation indices used in empirical back‐ups (e.g. NDVI‐LAI relationships) are themselves radiative constructs sensitive to tissue water status (Myneni et al., 2002).

This radiative nature creates a critical ambiguity in epiphyte‐rich canopies. In forests world‐wide, substantial fractions of light‐intercepting surface area are not host leaves but epiphytic assemblages: cryptogams (lichens and bryophytes) and vascular epiphytes, such as Tillandsia and desiccation‐tolerant ferns (e.g. Pleopeltis; Fig. 1a). Reported epiphyte loads can rival or exceed host foliage (Nadkarni, 1984). In tropical montane cloud forests, multilayered epiphyte communities combine upwards of 100 nonvascular species (Tucker et al., 2025) with a hyper‐diverse vascular community of hundreds of herbaceous species and dozens of epiphytic shrubs, treelets and hemiepiphytes (Haber, 2000; Fig. 1b), where their canopy organic matter exceeds the labile, nonwoody host biomass by over fourfold (Nadkarni et al., 2004). In temperate rainforests, bryophyte mats can blanket virtually all bark surfaces (Fig. 1c), with epiphytes reaching four times host‐tree foliar biomass (Nadkarni, 1984).

Fig. 1.

Three distinct ecosystems where epiphytes contribute a substantial vegetative area in the canopy. In the coastal, southern live oak‐dominated forests of the southeastern United States (a), canopies are dominated by the epiphytic resurrection fern (Pleopeltis polypodioides (L.) E.G. Andrews & Windham) and the pendulant Spanish moss (Tillandsia usneoides). In tropical montane cloud forests (b), epiphytic communities consist of a hyper‐diverse and abundant assemblage of bryophytes as well as herbaceous and woody species. In the temperate rainforests of the Pacific Northwest of the United States and British Columbia, Canada (c), the canopy community consists of thick mats of bryophytes and their associated organic materials (inset) together with the occasional vascular epiphyte. Photo credits: Sybil Gotsch and Dan Tucker.

When hydrated, these surfaces absorb photosynthetically active radiation and alter canopy reflectance; when desiccated, many become optically and physiologically quiescent. For example, poikilohydric nonvascular photoautotrophs (NVPs), including bryophytes and lichens (May et al., 2018; Ås Hovind et al., 2020; Morales‐Sánchez et al., 2022) and desiccation‐tolerant ferns like Pleopeltis spp. cease photosynthetic activity and show reduced Chl fluorescence when dry (John & Hasenstein, 2020; Prats & Brodersen, 2021). Similarly, vascular epiphytes, like Tillandsia usneoides (L.) L., exhibit significant decreases in NDVI with dehydration (Moore et al., 2023), demonstrating the optical switching that toggles epiphyte contributions to canopy radiative budgets between hydrated and desiccated states. Hydration dynamics also vary among epiphyte functional types, where NVPs can track atmospheric humidity within hours (Morales‐Sánchez et al., 2022), yet vascular epiphytes with substantial internal water storage may remain optically active for days to weeks following precipitation (Gotsch et al., 2015). Consequently, the same forest structure can yield different ‘LAI’ retrievals depending on epiphyte hydration state: after rainfall, epiphyte absorption is counted toward leaf area then, when dry, desiccation‐tolerant epiphyte area becomes optically muted. Current LAI products thus inadvertently create a time‐varying index where canopy structure remains constant but retrieved values fluctuate with epiphyte wetting‐drying cycles – independent of host phenology.

Beyond radiative ambiguity, conflating epiphytes with host foliage obscures physiological differences. Many epiphytes (i.e. NVPs) are poikilohydric, concentrating gas exchange into pulsed windows following rainfall or fog rather than operating more continuously like host leaves (Porada et al., 2023). Atmospheric bromeliads employ CAM photosynthesis (Silvera et al., 2010; Miller et al., 2021), shifting CO₂ uptake to nighttime and inverting the diurnal patterns embedded in many LAI‐based land surface parameterizations of daytime photosynthesis and transpiration. Moreover, many epiphytes possess specialized water‐storage tissues (hydrenchyma, lignotubers, etc.) that decouple canopy water content from host‐tree transpiration dynamics (Gotsch et al., 2015; Hargis et al., 2019). Per‐unit photosynthetic and transpiration rates of many epiphytes can be lower than those of host leaves (Stuntz & Zotz, 2001), but the issue is less about competing magnitude than about misattribution. Epiphytes can cause optically derived leaf area to be assigned to the wrong organism and the wrong physiological rule set, biasing the timing, moisture‐dependence, and temperature response of modelled fluxes in ways that per‐area rate comparisons alone cannot diagnose.

We propose an epiphyte area index (EAI), defined as projected epiphyte tissue area per unit ground area, to be quantified alongside conventional host‐foliage LAI. Separating EAI from LAI enables models and monitoring systems to: (1) partition biogeochemical processes between host and epiphyte layers; (2) condition physiological activity on hydric state for poikilohydric components; (3) distinguish structural contributions of epiphytes to interception and within‐canopy light regimes; and (4) forecast canopy responses to climate change and other stressors more accurately, since epiphyte communities and their hosts can respond to drought, fog loss, wetting frequency, heat, and host decline on different timescales and with different physiological constraints (Lawton et al., 2001; Helmer et al., 2019; Guzmán Q. et al., 2024; Silver et al., 2024). EAI is intended as a complementary canopy metric to LAI and woody area index (WAI). We illustrate this framework at two scales. First, at the stand scale, we apply terrestrial LiDAR scanning (TLS) to a Quercus virginiana Mill. (southern live oak) stand bearing Tillandsia‐Pleopeltis‐NVP assemblages, scanned immediately after substantial rainfall so that epiphytes were fully hydrated and optically active (see ‘Study site, terrestrial LiDAR scanning, and EAI estimation’ in the Materials and Methods section). Briefly, ground returns were removed, the registered point cloud was partitioned into host wood, host leaves, and epiphytes using geometric criteria, then occlusion‐aware ray casting estimated projected component areas. We emphasize that this TLS workflow is a stand‐scale proof‐of‐concept best suited to relatively open stands and distinctive epiphyte morphologies; denser understories or more complex epiphyte growth forms will require additional validation, sampling, or spectral/geometric training data. Complementary TLS‐based approaches to canopy component separation have been demonstrated (Van Langenhove et al., 2021, D'hont et al., 2024, Gaskins et al., 2026) and the methodological space is growing. Second, at the global scale, we use the Lichen and Bryophyte (LiBry) dynamic global vegetation model (Porada et al., 2014; Ma et al., 2024) to estimate the potential extent of EAI from nonvascular epiphytes across forested biomes (see ‘Lichen and bryophyte simulation’ in the Materials and Methods section).

At the stand scale, the TLS‐derived composite PAI over the optical footprint was 3.26 m² m⁻². This included WAI = 1.39 m² m⁻², host LAI = 0.89 m² m⁻², and EAI = 0.98 m² m⁻², with epiphytes contributing c. 30% of the TLS composite PAI. The corresponding effective stand‐level ceptometer PAI estimate, using an LAI‐2200TC (LiCOR), was 3.9 m² m⁻² (Bittar et al., 2018). Thus, TLS composite PAI was broadly consistent in magnitude with the effective optical retrieval while explicitly partitioning the surfaces that the ceptometer cannot distinguish at this site. Because T. usneoides festoons are internally clumped, this EAI should be interpreted as an effective projected area and likely a lower bound on this site's total epiphyte surface area. A simple sensitivity using a T. usneoides clumping correction factor of (1/Ω) 1.5–3.0 (Chen & Black, 1992; Zhu et al., 2018) would increase EAI from 0.98 to 1.47–2.94 m² m⁻² and raise the epiphyte share of a clumping‐expanded composite PAI from 30 to 39–56%. We retain the uncorrected projected EAI for parity with optical effective PAI, while noting that true epiphyte surface area is likely greater. Validation of TLS‐derived structural decomposition exists along a continuum, from internal point‐cloud consistency checks to cross‐method optical comparison and destructive harvesting. The agreement reported here covers the first two, an internal and a cross‐method check that bounds the TLS estimate against the established optical reference for this site. Destructive harvesting, which would close the loop, was not undertaken but is a clear next step.

At the global scale, LiBry predicts a mean NVP bark‐surface cover fraction of 55% (Fig. 2a). Weighting cover by a WAI field (also called stem area index (SAI)) to express it on a ground‐area basis comparable to LAI yields a global mean NVP‐EAI of 0.36 m² m⁻² ground (Fig. 2b). Spatially, the highest NVP cover fractions (> 80%) occur in tropical montane regions and temperate rainforests (Fig. 2a), while WAI‐weighted NVP coverage also highlights lowland humid forests (e.g. Amazonia and Southeast Asia) as hotspots (Fig. 2b). This value can be interpreted as a lower bound on NVP's contribution to EAI rather than a definitive NVP‐EAI estimate. Three considerations suggest the true value will be greater: (1) NVP tissues are layered (LAI typically ranges from 2 to 10 in bryophyte mats; Bond‐Lamberty & Gower, 2007; Niinemets & Tobias, 2019); (2) the LiBry simulation reported here uses a conservative thallus‐area‐to‐projected‐area ratio of 1.0; and (3) the model represents only nonvascular epiphytes, leaving vascular epiphyte loads in tropical forests as additional, unquantified contributions. Validation of LiBry against in situ NVP cover data is sparse outside temperate, boreal, and dryland systems (Ma et al., 2024), so tropical estimates carry larger uncertainty than other climatic regions.

Fig. 2.

Global distribution of nonvascular photoautotroph (NVP) epiphyte coverage from LiBry simulations. (a) NVP cover fraction on stem and branch surfaces (%). (b) NVP cover fraction weighted by stem area index (m² m⁻² ground), representing a lower‐bound estimate of NVP contributions to an epiphyte area index (NVP‐EAI). Actual NVP‐EAI values are likely severalfold higher due to layering of bryophyte tissues. (c) Maximum potential overestimation of active leaf area if hydrated NVP epiphytes are optically classified as vascular leaves but contribute negligible photosynthesis.

The successful validation of surface cover fraction predicted by LiBry at local scale for various NVP covers in Southern Africa (Porada et al., 2019) and Central Europe (Ma et al., 2024), and the ability to reproduce key traits of nonvascular epiphytes in Mediterranean regions (Porada & Giordani, 2021) show, however, that the model captures dynamics of biomass and cover for a range of strongly differing climatic conditions. Moreover, global estimates of net primary productivity (NPP) (Porada et al., 2017), biomass, and water‐storage capacity (Porada et al., 2018) have been validated against available observational data from the literature at biome level, showing good agreement in general.

We further used LiBry to bracket the functional consequence of misclassifying hydrated NVP epiphytes as host foliage in LAI‐based gross primary productivity (GPP) attribution (Fig. 2c). If wet NVP surfaces photosynthesize at rates and diurnal patterns comparable to vascular leaves, the attribution error approaches zero. If, conversely, NVP surfaces are optically counted as leaves but contribute little photosynthesis, then LAI‐based GPP is overestimated by the fraction of apparent active leaf area contributed by hydrated NVPs. Globally, LiBry predicts this maximum overestimate to be 18.4%, with strong regional variation and higher values in several high‐latitude humid forests (Fig. 2c). A ±20% sensitivity analysis on four key parameters governing NVP cover changed this global maximum by < 0.02 percentage points, indicating that the structural estimate is robust to moderate parameter variation (Table 1). This analysis does not imply a universal 18% GPP bias, rather it provides a defensible first‐order upper bound on one well‐defined misclassification pathway. Collectively, these patterns suggest that EAI‐related biases in LAI retrievals may affect radiative budgets across major forested biomes, especially where humidity and wetting frequency keep epiphytes optically ‘on’ for large fractions of time.

Table 1.

Sensitivity of the maximum leaf area index (LAI)‐based gross primary productivity (GPP) error estimate to ±20% variation in key LiBry parameters controlling nonvascular plant cover, biomass, and productivity.

	Max. err (%)	Cover in canopy	Biomass (g m−2)	NPP (g m−2 a−1)
Control run	18.374	0.130	20.790	6.156
Rubisco per respiration +20%	18.376	0.130	20.674	7.184
Rubisco per respiration −20%	18.370	0.130	20.767	4.956
Max. CO2‐Diffusivity +20%	18.374	0.130	20.838	6.219
Max. CO2‐Diffusivity −20%	18.374	0.130	20.731	6.081
Concavity light response +20%	18.374	0.130	21.604	6.657
Concavity light response −20%	18.374	0.130	20.068	5.754
Disturbance interval +20%	18.383	0.130	20.716	6.181
Disturbance interval −20%	18.360	0.130	20.828	6.143

LAI has long been treated as a property of forests. Perhaps it is better understood as a property of measurements, specifically, of radiative retrievals that register all light‐intercepting surfaces without distinguishing their identity or physiology. In epiphyte‐rich canopies, those surfaces may include some tissues that photosynthesize only when wet, fix carbon only at night, and decouple from host water status for days following rainfall. Our stand‐scale TLS results show that epiphytes can contribute roughly one‐third of the effective canopy PAI in a humid coastal forest, while LiBry indicates that NVP epiphytes alone can produce a first‐order active‐leaf‐area ambiguity at regional to global scales. These are not trace contributions to be absorbed into error terms. Near‐term implementation does not need to rely on a single method. TLS can provide plot‐scale partitioning and validation. Repeated optical or satellite observations can exploit phenological or hydration contrasts. Host allometries can constrain LAI where independent PAI is available. Additionally, airborne or spaceborne LiDAR may help scale structure where epiphyte loads are large.

Our case for EAI therefore offers a pragmatic lens – one that must still contend with canopy complexity – but it provides an entry‐point for creative methods to name, measure, and model a canopy layer that current paradigms render invisible, one that holds a substantial amount of water and nutrients, has a major influence on canopy microclimate and energy balance and decouples from host phenology (Gotsch et al., 2016; Hargis et al., 2019; Vaughan et al., 2026). Quantifying EAI alongside LAI will enable land surface models to assign physiologically appropriate rules to physiologically distinct tissues. Operationally, host LAI should be applied together with appropriate parameterizations of host photosynthesis and transpiration; considering EAI should also include information about epiphyte light interception, water storage and interception, and diurnal patterns of physiological activity. Ideally, epiphyte‐specific rules are applied (e.g. hydration‐gating for poikilohydric taxa, nocturnal uptake for CAM bromeliads, and water‐storage decoupling for tank‐forming species). Composite PAI remains useful for radiative transfer but should not be interpreted as host LAI. The forests where EAI matters most (tropical montane cloud forests, temperate rainforests, and humid coastal woodlands) are also those facing acute climate pressures (Lawton et al., 2001; Helmer et al., 2019; Guzmán Q. et al., 2024; Silver et al., 2024). If we are to predict how these ecosystems will respond to altered precipitation regimes, we must first see them clearly. EAI offers a lens.

Materials and Methods

Study site, terrestrial LiDAR scanning, and EAI estimation

We conducted field measurements at the University of Georgia Skidaway Institute of Oceanography (SkIO) campus, Savannah, GA, USA (31.99°N, 81.02°W), a Quercus virginiana Mill. dominated forest with an effective stand‐level PAI of 3.9 m² m⁻² derived from an LAI‐2200TC Plant Canopy Analyzer (Bittar et al., 2018). Terrestrial LiDAR Scanning (TLS) generated a georeferenced, 3‐D point cloud of the site from 12 scans using a Leica BLK360 scanner (Leica Geosystems, Norcross, GA, USA). Scans were co‐registered in Cyclone Register 360 with a mean error of 0.007 m. The merged, georeferenced point cloud is described in Raffai et al. (2025) archived at Zenodo (record 13356635). All scans were collected (10 September 2022 06:00 h–08:00 h) immediately after rainfall (77.5 mm ending c. 8 h prior; University of Georgia Weather Network, 2026) when epiphytes are saturated to ensure our structural measurements align with the maximum contribution of epiphytes to optical retrievals.

From the merged point cloud, 14 individual Q. virginiana crowns were isolated for analysis. Projected canopy area (PCA) for each tree's ground footprint was calculated using alpha shapes (Edelsbrunner et al., 1983; Wischmeyer et al., 2024), with a curvature (alpha value) of one. To facilitate the segmentation of canopy components, point clouds were processed using a weighted Laplacian contraction algorithm adapted from Cao et al. (2010). In this process, the per‐iteration contraction of each point is determined by solving a weighted Laplacian smoothing system of the form:

$$P^{t+1}=\arg {\min }_x{‖W_L^t{L}^{t}X‖}_F^2+{‖W_H^t\left(X-P_t\right)‖}_F^2$$

where $L^t$ is the Laplacian operator computed from the local neighborhood structure of the point cloud, $W_L^t$ is a diagonal weighting matrix controlling the contraction force, $W_H^t$ is a diagonal weighting matrix controlling attraction to the previous point positions, and X represents the updated coordinates of a given point. This technique modifies traditional Laplacian smoothing, with the control matrices penalizing excessive displacement of points with locally dense neighborhoods. The magnitude of this contraction was used as a quantitative measure of how distal an individual point was relative to points in its local neighborhood, with the resulting contraction used to approximate the woody structure of the scanned tree (Meyer et al., 2023). Points were then classified according to the direction and magnitude of their contraction. Although the contraction‐distance distribution provided the basis for identifying a percentile threshold, the selected cutoff was visually verified against the three‐dimensional point cloud rather than accepted solely from the histogram or summary statistics. Following this validation, points with contraction distances at or below the 85^th percentile were classified as Wood, while points above this threshold and exhibiting positive contraction in the z direction (indicating that they lay beneath the approximated woody structure) were classified as ‘P Epiphytes’ (Fig. 3). Points in each class were then projected onto the XY plane and Delaunay triangulation was used to create surface meshes approximating 2D area covered by each component. From these surfaces, we derive PAI for the entire canopy, host LAI, and EAI.

Fig. 3.

Example of terrestrial LiDAR scanning (TLS)‐based epiphyte classification and plot‐scale representation. (a) Epiphyte‐classified returns (colored points) overlaid on the fitted cylindrical stem/branch model (gray) for a representative Quercus virginiana, illustrating the high‐curvature, high‐density, pendulous point structure used to identify epiphytes. (b) Heat map approximating the extent and depth of epiphyte coverage across the c. 0.5 ha plot (14 trees). Coverage (color intensity) expressed as a count of points sampled using a regular 0.5 m × 0.5 m voxel grid.

The method of segmentation described above was utilized due to the ubiquity of open‐source software packages utilizing such techniques (Cao et al., 2015, Meyer et al., 2023). Though high‐density TLS scans were utilized for this analysis, others have found success applying similar algorithms to point clouds obtained from lower density drone collected scans (Zhang et al., 2023). However, computing EAI at the forest‐scale and in more densely forested environments will likely require machine‐learning (ML) models, such as those put forward by Gaskins et al. (2026).

Lichen and bryophyte simulation

The LiBry model (Porada et al., 2013, 2014) is a dynamic global vegetation model (DGVM), which differs from most other DGVMs in two aspects. First, it focuses on NVPs, mainly lichens and bryophytes, and represents vascular plants, such as trees or grasses, as simple boundary conditions, that is periodically changing values of LAI and SAI at monthly resolution that exert shading effects on NVP and provide space for growth in the case of nonvascular epiphytes. Second, the LiBry model uses the principle of natural selection, combined with ecological and ecophysiological theory, to predict trait distributions of NVP communities at any location in the world. Through the community‐weighted mean values of these trait distributions, the morphological and physiological properties of NVP communities can be determined at the global scale. This also includes their biomass and surface cover. The approach goes beyond the concept of plant functional types (PFTs) used in many other DGVMs and has the advantage that community‐level properties and their potential shifts below the coarse level of a PFT can be captured.

The selection of NVP in LiBry and, therefore, the emergence of trait distributions is driven by environmental filtering and biotic interactions. The model input thus includes environmental forcing, such as time series of climate data and information about disturbance regime. The trait distributions that describe NVP communities in LiBry form a multidimensional space that is sampled at the initialization of the model by a large number of possible trait combinations, so‐called functional strategies. The model then computes at all locations where forcing data is available which subset of the initial strategies are able to survive in the long term by maintaining a positive carbon balance. The carbon balance is coupled to the exchange of energy and water with the atmosphere. In this way, LiBry links functional diversity to biogeochemical cycles of carbon and water.

For this study, we used climate forcing based on the WATCH reanalysis data for the years 1958 to 2001 (Weedon et al., 2011) and periodic maps of LAI and SAI at monthly resolution derived from the Community Land Model (Bonan et al., 2002). The setup is the same as the one used in Porada et al. (2018), but we only include results for epiphytic NVP here in our study. The SAI is used as a measure for the available space for growth; both LAI and SAI determine light availability according to Beer's law (Porada et al., 2013) and the availability of water in the canopy is determined by an empirical scheme of precipitation partitioning described in Porada & Giordani (2021). However, for our study here we switched off the supply of water by uptake from the bark in order to avoid potential overestimation of NVP growth and coverage that may result from poorly constrained estimates on the size of the bark water reservoir at a global scale.

The percentage cover of epiphytic NVP at each grid cell of the model is computed as the average value over the last 5 yr of a 300‐yr model simulation with repeated climatic forcing from the WATCH reanalysis data. The long simulation time is necessary to ensure that the emerging communities in the model are in a steady state, that is all surviving strategies are viable in the long term. The cover fraction is thereby computed as the fraction of stem (and branch) area in each grid cell that is covered by NVP, where the NVP cover is the sum of the cover values of all surviving NVP strategies in that grid cell. The values are usually below 100% since turnover and disturbance lead to continuous losses of cover such that only at the highest growth rates the strategies are able to cover the entire free space. The SAI values used in our simulation are usually below 1.0 m² m⁻² ground. Therefore, the cover fraction of epiphytic NVP is in general reduced when referenced to ground area (Fig. 2b) instead of stem area (Fig. 2a). It should be pointed out, though, that the simulated cover fraction values are a lower‐bound estimate since the individual thallus (body) surface area values of the NVP strategies are not considered here. The LiBry model computes for each strategy the projected area that is covered on the available surface (e.g. bark), but due to the layered or leafy structure of many epiphytic NVP, the actual thallus surface area contained on that projected area is much higher. The model uses an individual ‘NVP‐LAI’ for each strategy to compute light interception and other properties, but observation‐based estimates on this NVP‐LAI are rare for epiphytic NVP at global scale, making a validation of the simulated community‐weighted NVP‐LAI values challenging. Therefore, we use the minimum value of 1.0 for the ratio of thallus surface area to projected area for all NVP strategies here.

To assess the maximum effect of misclassifying nonvascular epiphytes as leaves on global, LAI‐based estimates of GPP, we computed the ratio of NVP cover to LAI for each grid cell of the LiBry model. Subsequently, we assumed that the epiphytes are wet, hence being classified as vascular leaves, but perform only negligible photosynthesis. Additionally, to assess uncertainties associated with the simulated NVP cover, we varied 4 key parameters of the LiBry model by ±20% that have been previously identified as relevant for productivity and biomass estimates at global scale (Porada et al., 2013). These are (1) the respiration costs associated with RuBisCO enzyme turnover; this is the central trade‐off in many mechanistic photosynthesis models, as it determines how much respiration is required to achieve a certain photosynthetic capacity, and thus controls if sun‐plant or shade‐plant strategies are advantageous under any given light regime; (2) the maximum CO₂ diffusivity of the pathway from ambient air to the chloroplasts; this trait is specific for NVP and is similar to stomatal conductance, with the key difference being that diffusivity decreases in NVP with increasing water saturation due to blocking of gas exchange, while vascular plants open their stomata under increased water availability; (3) the concavity of the light response of photosynthesis, which represents self‐shading in the model; a higher value allows for higher light use efficiency; (4) the biome‐specific disturbance interval, which determines the fraction of each grid cell in a given biome (and thus the NVP cover) that is destroyed each year. We evaluate the impact of parameter uncertainty on the maximum error due to misclassification (as described in the previous section) by applying these parameter variations to the control simulation (Table 1).

Competing interests

None declared.

Author contributions

CW conceived the study idea in discussion with dissertation advisor JTVS. CW, JTVS, SGG and PP contributed to the subsequent conceptual development. JTVS provided project leadership, administration and funding support, and conducted the terrestrial laser scanning and co‐registration at the Skidaway field site. SGG contributed expertise on vascular epiphytes and prepared Fig. 1. PP contributed expertise on nonvascular epiphytes, ran the LiBry model, and prepared the Fig. 2 maps and sensitivity analyses in Table 1. CW performed the LiDAR point‐cloud segmentation and prepared Fig. 3. CW and JTVS wrote the first draft of the manuscript. All authors contributed to manuscript writing, review and revision, and to addressing referee comments.

Data availability

All data are publicly available and linked in the methods section where appropriate.