Habitat structural heterogeneity has long been recognized as a key factor for explaining biodiversity patterns at a local scale [1
]. High structural heterogeneity may promote species coexistence by generating varied resources, shelters, and microclimates, thus providing a greater number of niches for species to occupy [3
]. In forests, plant communities represent the main structural component producing habitat heterogeneity, and this heterogeneity has been shown to influence biodiversity and species distributions [5
] of birds [1
], mammals [9
], reptiles [11
], and invertebrates [12
Quantifying forest structural heterogeneity via field measurements is an ongoing challenge because it is both logistically difficult and expensive [13
]. However, recent advances in airborne light detection and ranging (lidar) technologies have enabled three-dimensional (3D) measurement of forest structure with increasing accuracy. Improvements in spatial extent, resolution and cost-effectiveness have turned lidar into an effective tool for studying 3D forest structure at scales relevant to animal diversity [13
]. Despite the ability of lidar pulses to penetrate the top layers of the forest, most animal diversity studies hitherto have focused on building lidar-derived structural metrics using only canopy cover, or the variation in canopy height [14
]. Although these canopy-based metrics often show important relationships with certain animal species or communities, such as birds [16
], structural metrics that capture the entire 3D forest structure may offer greater predictive power for the study of biodiversity across multiple taxa than previous metrics.
Because discrete return lidar instruments can provide several signal returns per pulse and several points per square meter, the vertical distribution of lidar returns is often used to characterize the vertical profile of vegetation, which offers important information about the understory forest structure and, therefore, might be useful for predicting local biodiversity patterns. For example, Goetz et al. [17
] used the height of the understory layer relative to the overstory lidar returns as a predictor of abundance of the Black-throated Blue Warbler.
One shortcoming of metrics focusing on a vertical vegetation profile is that often they do not account for the horizontal variation in the vertical profile [5
]. However, the horizontal variation of the vertical profile is a unique component of the vegetation structural heterogeneity and could have different effects than vertical variation on animal taxa. While there have been many studies examining the effect of horizontal vegetation heterogeneity on species richness [18
], most of them focused on non-structural features such as heterogeneity in plant diversity [19
] or structural features measured at a single vertical stratum such as canopy height or trunk diameter heterogeneity [5
]. Few studies have attempted to examine the horizontal heterogeneity of vegetation structure across the full vertical profile of the forest. With the new capabilities of lidar instruments for characterizing vertical profiles, the identification of structural elements that contribute to heterogeneity and enhance diversity needs further investigation [14
A second shortcoming is that these methods did not consider the attenuation of the lidar pulse passing through the canopy, a factor that can vary for different forest types and whose effects, if disregarded, can lead to a potential underestimation of the vegetation density at the lower layers [20
]. The leaf area density (LAD, [21
]) is a structural attribute based on the concept of the leaf area index (LAI), a common remote sensing metric in vegetation growth studies [22
] that quantifies the ratio between leaf area and per unit ground surface. The LAD can be estimated from discrete lidar in order to study the distribution of plant density across the vertical profile. When appropriately calibrated with vegetation attenuation coefficients for a given lidar pulse density and for a particular grid size, LAD can provide reliable measurements of plant densities for all layers of the canopy [23
]. Structural metrics based on LAD measurements have been used mainly for studying forest inventory attributes such as standing volume and biomass [25
]. Despite its potential, the use of LAD measurements to characterize forest structural heterogeneity (both vertically and horizontally) and relating it to animal diversity has not been explored to the best of our knowledge.
In this study, we built new lidar-derived metrics of forest structural heterogeneity and used them to predict bird species richness across different forest plots in the eastern US. The ability of birds to fly and sample all three dimensions of their habitat means that the 3D vegetation structure is often a strong influence on the ecology of a species [14
], making them an excellent and interesting case-study for examining structural heterogeneity–diversity relationships. We created a suite of LAD-based metrics to capture both the vertical and the horizontal components of structural heterogeneity. Our objectives were to test: (1) whether our LAD-based 3D heterogeneity metrics had better explanatory power for bird species richness than existing canopy height-based metrics, and (2) if different aspects of structural heterogeneity had different effects on bird richness. We used discrete airborne lidar data provided by the National Ecological Observatory Network (NEON) at five forested sites in the eastern US. We used surveys of plant structure and composition to compare our lidar metrics against field-based measurements, and 61 breeding landbird point counts to study vegetation structural heterogeneity–diversity relationships. Finally, we created random forests models of bird species richness and we analyzed the importance and the shape of the relationships between each lidar-derived metric and bird species richness.
Discrete lidar is becoming an effective tool to study forest structure, but creating lidar-derived metrics that reliably characterize the vegetation heterogeneity in 3D is challenging. Our findings suggest that identifying appropriate lidar metrics that account for the different aspects of structural heterogeneity can provide new insights into structural heterogeneity-biodiversity studies. Metrics based on LAD measurements showed better explanatory power with regard to bird richness than those based on the variation of canopy heights. Our results also show how vertical and horizontal heterogeneity can have disparate effects on bird richness.
Our lidar metrics were weakly to moderately correlated with field-based measurements of vegetation structure. Among the latter, variation in stem diameter was most strongly correlated with lidar metrics. Previous studies had shown stronger correlations between lidar-derived vegetation metrics and field-based measurements [13
]. However, most of these studies examined lidar metrics that were unrelated to heterogeneity, such as aboveground biomass, or used spatially located individuals to represent the horizontal distribution of the vegetation in the field. In our study, the weak relationships between lidar metrics and field heterogeneity measurements are likely largely due to mismatches between aspects of heterogeneity quantified by the two types of measurements. For example, data on the heights were not assigned to individual tree crowns, limiting our assessment of metrics related to vertical heterogeneity. Furthermore, the field data comprised many vegetation types ranging from large trees to small bushes; therefore, consideration of the total variation in heights or stem diameter could have caused misleading measurements of vegetation heterogeneity. For instance, large trees might have had the biggest impact on some of the lidar-derived metrics, while the field-based measurements might have heavily weighted the most common individuals such as small bushes or small trees. Future work linking the spatial location of the individuals with the structural measurements are needed to create field-based spatially explicit models of the distribution of the vegetation and to study their relationships with the lidar metrics.
In our study, we used metrics based on LAD measurements as explanatory variables of bird richness across forest plots. To the best of our knowledge, this is the first time these metrics have been used to study the relationships between forest structural heterogeneity and animal diversity. Mean LAD values for the 10 m × 10 m × 1 m voxels were within the ranges of those obtained in previous work that estimated LAD in the SERC site using lidar data [23
] or field measurements [51
]. Mean LAD was closely and positively related to metrics capturing the vertical structural heterogeneity, but did not have a strong relationship with other metrics representing horizontal heterogeneity. Temperature was also closely related to some of the LAD metrics, and, surprisingly, plots with higher temperatures presented lower LAD densities and lower vertical heterogeneity. Although temperature gradients have been linked to differences in forest biomass [52
], these relationships are strongly influenced by the scale of the studies and forest type, while other climatic factors, such as water deficit, may have strong effects at regional scales [53
]. In general, metrics that aimed to capture vertical heterogeneity, albeit built with diverse methods, showed close inter-relationships and the same can be said for the horizontal heterogeneity metrics. Despite their reliability, LAD-based metrics present several limitations. For example, the light attenuation coefficient k could vary slightly between forests with different tree compositions [25
], while specific k values must be estimated for different sensors and plot extents [23
]. LAD values at the lower canopy layers may also be overestimated for low-density point lidar sensors such as NEON’s [23
]. Differences in point densities within a plot, owing to differences in lidar coverage (simple or double flight pass) could also affect LAD measurements for small voxels, such as those created for the vegetation plots, because the dependency of LAD values on point density becomes significant for scales smaller than 10 m × 10 m [23
]. Lidar scan angle might also affect GP and LAI measurements for single coverage regions [54
]. Despite the fact that effects of low angles (single coverage regions generally present low angles) on GP and LAI were shown to be minimal [54
], the influence of the scan angle could depend on the forest type [56
]. Corrections of these possible biases should be addressed in the future.
LAD-based metrics showed, in general, higher explanatory power with regard to bird richness than metrics based on top of the canopy measurements. Models based on LAD metrics alone were found to be more accurate than canopy-based metrics, and LAD metrics were found to be more important than canopy-based metrics in the mixed-metrics models. Standard variation of the canopy height was nonetheless included as the fifth most important variable in the the top-seven variables model. Metrics based on canopy height have long been found to be correlated with species distributions or diversity of many taxa [5
]. Despite very early work linking the distribution of biodiversity with vertical vegetation heterogeneity [1
], canopy-based metrics are most commonly used in studies linking habitat structure to animal ecology [14
]. This is probably because these metrics can be easily extracted from satellite data or from lidar sensors with minimal data processing and with high reliability. In this study, metrics that accounted for the horizontal variation across 10 m × 10 m grids had better explanatory power with regard to bird richness than those metrics measuring the vertical heterogeneity at the plot level. A few animal ecology studies considered the horizontal variation of the vertical forest structure [57
], finding that this horizontal component was key for explaining species distributions. However, most studies that measured structural heterogeneity using vertical vegetation profiles did not consider the horizontal component [5
]. The capabilities of the new airborne lidar sensors in terms of pulse density and extent should allow ecologists to study the variation of the vertical vegetation profiles with greater detail [59
]. Grid-based metrics such as the ones built in this study will enable the addition of this horizontal component to structural heterogeneity-biodiversity studies.
The analysis of the models’ response shapes revealed different components of structural heterogeneity affected bird richness disparately. Metrics measuring horizontal heterogeneity at different canopy layers, or measuring the horizontal variation of the vertical canopy profile, showed positive effects on bird richness. While the positive effects of horizontal structural heterogeneity on bird richness are well documented [3
], very few of those studies considered all three dimensions of the vegetation (but see [6
]). This study did not consider, however, the effects of differences in forest types between sites. The intensity and shape of the effect of horizontal heterogeneity could differ between forests with different vertical configurations, therefore future research considering horizontal heterogeneity across a variety of forest types will help clarifying the absolute effects of this factor on bird richness. The horizontal variation of the mean vertical LAD showed a strong positive effect on bird richness, similar to the effect of canopy height variation. Therefore, variation in canopy height and canopy cover may be closely linked to horizontal heterogeneity in lower canopy layers, which might explain the success of studies that used canopy metrics to explain bird distributions [62
While lower and intermediate values of vertical heterogeneity showed positive effects on bird richness, our analyses indicate that high vertical heterogeneity had a detrimental effect. These results contradict previous studies that found higher bird diversity with increased vertical profile heterogeneity [1
]. However, vertical heterogeneity in many of those studies was calculated using only canopy height variations, or with metrics that did not consider all canopy height layers. The Shannon entropy index has also been used to characterize the vertical distribution of vegetation. For example, Refs. [1
] used a Shannon entropy index to study the foliage height diversity, but the index was calculated as the evenness of the vegetation density across height layers and not as vegetation density heterogeneity across heights. However, Ref. [17
] used full waveform lidar data to create an entropy index similar to our
metric, finding positive associations with a single songbird species. Other studies found negative associations between vertical heterogeneity and bird richness [16
], supporting our results. Nevertheless, we acknowledge that the relationship between vertical heterogeneity and diversity may be highly dependent on how heterogeneity is measured. Our vertical heterogeneity metric based on Shannon-index is strongly correlated with the Shannon-index of overall heterogeneity and mean vegetation density, thus adding difficulties to the isolation and study of this structural factor. Further work is needed to develop and examine vertical heterogeneity metrics that are independent of other structural metrics to test the effect of vertical heterogeneity more conclusively.
Besides the nonlinearity and correlations between components of forest structural heterogeneity, some components may be correlated to other forest characteristics that affect bird diversity, such as forest successional stage [65
] and fire disturbance [66
]. Structural heterogeneity–diversity relationships might also be height- and taxa-dependent, as some bird groups prefer highly dense and homogeneous lower canopy layers, while other groups may rely only on the heterogeneity of the upper layers [6
]. Variation in bird detectability from point counts could also affect the study of forest structural heterogeneity–diversity. Different bird species can be heard from much farther away than others and the same individual of a species can be heard from different distances depending on the forest type [67
]. Denser canopies could decrease the detectability of some bird species, thereby possibly contributing to the negative correlations between bird richness and vegetation density observed in our study. Further work to correct this bias is needed.