Airborne laser scanning, or Light Detection and Ranging (LiDAR), is a proven technique for making precise and accurate three-dimensional measurements of forest and other complex vegetation canopies over large spatial extents [1
]. As such, it is a powerful tool for an increasing range of applications in forestry, ecology and conservation. These include habitat suitability modelling [4
] and the monitoring of carbon stocks, which is an essential requirement of projects for reducing emissions from deforestation and forest degradation (REDD+) [7
At the heart of LiDAR’s effectiveness in modelling vegetation three-dimensional structure is a predictable relationship between return data and the arrangement (e.g., heights and densities) of component parts of the vegetation. However, there is still much to be understood in relation to vegetation–laser pulse interactions [9
]. Repeat LiDAR surveys of deciduous/mixed forests in leaf-off and leaf-on states indicate that foliage development of the tree has a significant influence on the scatter of LiDAR height measurements, and therefore the retrieval of information on parameters of interest for individual trees [10
] and forest plots [11
]. Laser penetration rate through the canopy is greater in leaf-off conditions, providing better information capture on the presence of an understorey (of suppressed trees and shrubs), but potentially less optimal for the modelling of the forest canopy structure itself [16
]. Indeed, a better representation of ground and understorey layers in leaf-off state of Japanese deciduous forest has been shown [17
], and comparable results were obtained for an English mixed deciduous woodland [16
] where as much as 57% of last return height measurements were from the ground layer (<1 m) and 42.5% from the understorey layer (1–8 m). The combination of leaf-on and leaf-off data can help improve tree species classification, and in one example this has been demonstrated using LiDAR intensity information [12
Less attention has been given to the more subtle seasonal effects on the LiDAR modelling of evergreen or mixed canopies. Much of the world’s forest cover, which amounts to some 30–35% of the land surface or around 39–45 million km2
], is dominated by evergreen trees, including coniferous (principally boreal) and broadleaved evergreens. The latter include the majority of Mediterranean forests and woodlands, whose canopies are composed of evergreen oak (Quercus
spp.) and other sclerophyllous trees such as phillyrea, rhamnus and olive. Multi-temporal LiDAR is being increasingly employed for the investigation of tree growth, forest patch, vegetation and biomass dynamics (e.g., [19
]). An understanding of the robustness of LiDAR metrics in the face of seasonal leafing phenology is important for making correct inferences in such studies [26
In this current investigation, we capitalise on a repeat LiDAR survey of an area of mixed oak forest in southern Spain to consider the effect of timing on the LiDAR measurement of tree canopies and understories. We look at whether LiDAR variables differ significantly when captured six weeks apart. The study spans a period in which an evergreen tree species (Quercus suber) is experiencing leaf drop and concurrent new leaf emergence, and when a semi-deciduous tree species (Q. canariensis) is moving from a state of partial to full leaf expansion. We consider the implications of such processes on LiDAR variable retrieval for the application of this technology to the study of Mediterranean and other forest ecosystems.
Comparing LiDAR metrics retrieved at two dates six weeks apart in April and May, we test the hypothesis that they will be significantly affected by phenological changes to leafing state, and that such effects will be greater for Quercus canariensis than for Q. suber because of the leaf expansion of the former tree resulting in significantly less penetration of laser light through the denser mass of May-time foliage.
A chief strength of our approach is that the repeat surveys are conducted using the same sensor configuration and flight parameters, and in the same year. Despite obvious advantages, this has rarely been possible in past multi-temporal LiDAR studies. Data collection in a mixed deciduous woodland in eastern England [16
] was separated by two years, as was the case in a study in southeast Norway [11
]. The experimental design of the latter investigation was also undertaken with different sensors at different flying altitudes, requiring a number of complicated compensations in the analysis. Flying altitude can have a significant effect, as a reduction in peak pulse power concentration can delay pulse triggering within vegetation, thereby increasing laser penetration into the foliage and reducing height metric values [40
]. An increased interval time, on the other hand, allows tree growth to become a factor.
This investigation has found evidence for the effect of seasonality, specifically short-term spring-time leafing phenology in two typical Mediterranean forest species, on the retrieval of LiDAR variables of value for describing tree crown and forest stand vegetation structural attributes. In this way, the specific contribution that the study makes is in the investigation of more subtle effects of tree leafing phenology than the often studied leaf-on/leaf-off dichotomy. The investigation was opportunistic upon an unplanned repeat LiDAR survey, and hence a campaign of field data collection aimed at quantifying the associated phenological changes was not planned. We have, however, been able to draw upon contextual field data to reach some robust conclusions on the effect of seasonal tree leafing processes and the retrieval of LiDAR metrics.
Airborne LiDAR is being increasingly applied in Mediterranean ecosystems (e.g., [41
]). Knowledge of the effect on LiDAR parameters of seasonal changes to these predominantly evergreen canopies—including climatic influences on leafing phenology—is relevant to the design, data-analysis and interpretation of LiDAR surveys in this and other warm-temperate/sub-tropical regions.
Contrasting responses were observed for Quercus canariensis
and Quercus suber
canopies, according to our hypothesis. Field observation and comparison of the digital imagery for April and May (Figure 2
) indicate that the Quercus canariensis
trees develop from a state of partial to full leaf expansion during this period. Under partial leaf expansion, reflectances off branches will presumably represent a relatively high proportion of the LiDAR point cloud. As leaves expand, one can predict that the increased amount of foliage biomass, and canopy closure, will reduce penetration of laser pulses and increase the proportion and concentration of height measurements recorded in the upper strata of the vegetation profile. Our results confirmed this prediction, with an increased mean height, reduced standard deviation and more negative skewness of heights, in May compared to April. These results are analogous to those found for leaf-on/leaf-off conditions. For lowland mixed woodlands in eastern England, a small increase in mean height (13.35 vs. 12.47 m) and decrease in standard deviation of heights (5.10 vs. 5.25 m) was observed in leaf-on conditions [16
]. The height distributions of single returns and last-of-many returns have been shown to shift towards the ground in leaf-off conditions within boreal forests [10
]; skewness values under leaf-off conditions were more positive for single returns (but not for first or last returns) and the variability of the height distribution tended to increase from leaf-on to leaf-off conditions. In another study, the laser interception by the upper parts of a mixed forest canopy were significantly higher in leaf-on conditions [11
]. For the first return data, height metrics were 0.33–0.97 m higher under leaf-on canopy conditions. Analogous results were also shown in a comparison between tropical moist (TMF) and tropical wet (TWF) forest [48
]. At the end of the dry season, leaf loss from canopy-forming trees was pronounced in some TMF areas, and this led to less LiDAR energy being reflected from the upper canopy, thereby reducing the LiDAR median height metric relative to the TWF study area. In an investigation of light availability at different developmental stages of a boreal forest [49
], a direct relationship between skewness and the degree of light entering through crown was established, results that are consistent with the canopy closure between April and May in our study forest.
Maximum height was the least variable of the four metrics for Quercus canariensis
plots. The ratio values of ~1 for maximum height measurement for trees and grid cells is reassuring, in terms of reliable measurement of a variable that should be less affected by tree leafing phenology than the others that were calculated. This is again consistent with the results of the studies that have been reviewed here. Leaf-on/leaf-off values in one investigation were 25.31/25.14 m [16
]. In another, canopy conditions were found to exert little influence on the maximum height obtained for the individual trees, although maximum laser heights of ‘first’ echoes were higher in birch trees under leaf-off conditions [10
]. Meanwhile, plot-level stability of maximum height has been reported [11
]. Using percentile heights, a tendency for canopy height to be underestimated in leaf-off conditions has been observed, but only where the forest was dominated by deciduous compound-leaved trees [14
The tree-level and grid analyses have complemented each other in the comparison of time periods and canopy types. The former provided a number of metrics associated with tree crowns, whilst the latter was unaffected by any error in the tree segmentation process. The results of both give confidence that the LiDAR variables are relatively robust to the seasonality of tree leafing that this study spans. A possible exception is the measurement of skewness, especially in the case of the Q. canariensis
canopy, and the reasons for this have already been discussed. Skewness of heights is a useful summary measure of their asymmetry as affected by canopy closure [49
] and relative density of vegetation in the canopy and understorey layers. It can be used, for example, to differentiate natural forest and plantations, with their varying vertical vegetation profiles [50
]. The reduced skewness values obtained for Q. canariensis
plots during the May survey suggests that this timing is less optimal to capture information on the presence of understorey vegetation. This may also be the case for the evergreen Q. suber
, though evidence for this from our study is equivocal. It may be that a survey undertaken later in the summer, with full leafing out of the cork oak trees, would similarly be less useful for the description of layers below the tree canopy.