Yuxuan Sun / 孙宇轩

Yuxuan Sun / 孙宇轩

he/him

Master's student at Department of Landscape Architecture

Paper Reading: (2021) Integrating terrestrial laser scanning with functional–structural plant models to investigate ecological and evolutionary processes of forest communities-2

11 minute read

Published:

Original paper: O’Sullivan, Hannah, et al. “Integrating terrestrial laser scanning with functional–structural plant models to investigate ecological and evolutionary processes of forest communities.” Annals of Botany 128.6 (2021): 663-684. https://doi.org/10.1093/aob/mcab120

This is a record of the reading process, aimed at personal learning, reading and writing practicing.

4 The realm of TLS

In recent years, TLS has emerged as a revolutionary tool for measuring above-ground 3-D tree architecture and forest structure (Malhi et al., 2018; Disney, 2019; Calders et al., 2020).

Despite the promises and potential, the application of TLS to questions in forest ecology and evolution is still in its early days.

In a recent review, Malhi et al. (2018) outline a number of areas suitable for testing and extending ecological theory on tree form and function in the context of TLS, including seed dispersal, structural mechanics and resource distribution.

  • Malhi, Yadvinder, et al. “New perspectives on the ecology of tree structure and tree communities through terrestrial laser scanning.” Interface Focus 8.2 (2018): 20170052. http://doi.org/10.1098/rsfs.2017.0052

It is also noteworthy that allometric models can be refined towards more flexible functional models using the type of data that is available from TLS (Kaitaniemi et al., 2020; Kaitaniemi and Lintunen, 2021).

  • Alternatively, quantitative structural models (QSMs), which are a hierarchical collection of cylinders (Fig. 4), can be used to compute an array of detailed structural tree metrics, including AGB.
  • There are published and freely available methods to reconstruct QSMs from TLS data; these include:
    • TreeQSM (Raumonen et  al., 2013; Calders et  al., 2015a; Raumonen, 2020),
    • SimpleTree (Hackenberg et  al., 2015), SimpleTree is currently known as SimpleForest and it is a plugin for CompuTree.
    • AdTree (Du et  al., 2019) and CompuTree (Piboule et  al., 2013).

Differentiating between individual trees remains a key challenge when working with scans containing multiple trees.

There are at least two open-source software options available that tackle this issue in different ways.

  • Firstly there is treeseg (Burt et al., 2018) and 3dforest (Trochta et al., 2017), both developed with C++, which use a number of standard point cloud processing techniques;
  • and secondly there is LidR (Roussel et al., 2020), written in the R environment, intended for use with airborne LiDAR, but easily applied to TLS.
  • However, due to the popularity of the R language in biological sciences compared with C++, LidR has the added benefit of accessibility.

Methods for separating leaf and wood fall largely into two groups: geometric-based separation and intensity-based separation.

  • Intensity-based methods: does not perform well in general, may still be useful for classification as one of the information sources (X. Zhu et al., 2018).
  • Geometry-based methods: many approaches available
    • TLSeparation is an open-source Python library offering custom workflows and automated scripts (Vicari et al., 2019).
    • This software performs with a 90 % accuracy for separating leaf and wood in field TLS data, which is a beneficial step towards efficient TLS data processing of large forest scans.
    • There are other geometric-based published methods with similar accuracies (Wang et al., 2018, 2020; Moorthy et al., 2020).

4.1 Species identification

Any ability to automatically identify species across large areas would greatly improve the potential to apply FSPMs to larger ecological questions.

  • Othmani et al. (2013) used bark texture to identify five species from a sample size of 75 individuals using a random forest classification with an accuracy of 85 %.
  • Barmpoutis et al. (2018) classified four species of Caatinga trees using the fast-marching method for tree skeletonization, subsequently classifying skeletons with descriptors that account for a combination of dynamic, appearance and noise parameters.[the sample size (15 individual trees) is too low]
  • Using QSMs to obtain structural features and machine learning for classification.
    • Åkerblom et al. (2017) obtained an average classification accuracy of >93 % in a single-species forest plot, with lower accuracy found in mixed-species plots.
    • A more recent study by Terryn et al. (2020) expanded on this work.
    • The authors find that the greatest factor contributing to classification success is canopy class.

4.2 TLS in studies of vegetation dynamics

TLS + QSM: gain new understanding of forest dynamics and structure.

  • Seidel et al. (2011) showed how crown plasticity in terms of crown asymmetry is used by trees to avoid competition.
  • Van der Zee et al. (2021) introduced a novel method of quantifying the phenomenon of ‘crown-shyness’ by applying a 3-D surface complementarity metric to TLS data.
  • Beyer et al. (2021) investigated whether the branches along tree trunks exhibit a similar constant divergence angle.
  • Calders et al. (2015b) monitored the timing of recurring seasonal dynamics through the plant area index (PAI). TLS was shown to have the potential to study seasonal dynamics not only as a function of time, but also as a function of canopy height.
  • Campos et al. (2021). They outline the creation of an automated and permanent TLS measurement station, able to identify a number of short-term and long-term changes in a boreal forest.

Use TLS to investigate relationships between environmental factors and woody plant structure.

  • TLS data and the derived QSMs of trees have been used for mechanical modelling of the trees under critical wind speeds (Jackson et al., 2019).
  • Van der Zande et al. (2010) combined TLS measurements with a voxel-based light interception model to examine the relationship between variable light conditions and the distribution of leaves in 3-D space.
  • Jackson, Tobias, et al. “A new architectural perspective on wind damage in a natural forest.” Frontiers in Forests and Global Change 1 (2019): 13. https://doi.org/10.3389/ffgc.2018.00013
  • Van der Zande, Dimitry, et al. “Assessment of light environment variability in broadleaved forest canopies using terrestrial laser scanning.” Remote Sensing 2.6 (2010): 1564-1574. https://doi.org/10.3390/rs2061564

4.3 Using TLS to parameterize FSPMs

With TLS it is possible to quickly collect such data from a large number of individuals that form a representative sample.

  • This often requires that the TLS data are first transformed into QSMs, from which the structural tree data can be computed and inferred.
  • However, much of the required structural information can be estimated from TLS data without full QSMs, namely overall measures such as crown dimensions and woody plant height.
  • Moreover, useful information about the leaves, such as total leaf area (Béland et  al., 2011) and leaf orientation (Zheng and Moskal, 2012) together with their spatial distributions, can be estimated from TLS data.
  • Thirdly, the non-destructiveness of TLS allows for repeated measurements over a span of time, for example over a growth season or many years.
  • These time series can similarly be used to validate and test the accuracy of the dynamic modelling of woody plant growth in a given FSPM.
  • Lastly, time series data are useful for initialization, calibration or optimization of the FSPM parameters to make them correspond better to the observed form and function of woody plants.
  • TLS data and QSMs were similarly used by Potapov et al. (2016), where they proposed a stochastic version of LIGNUM for producing tree structures consistent with detailed TLS data.
  • They expanded the idea to general stochastic structure models and showed how to generate data-based morphological tree structure clones (Potapov et al., 2017).
  • stochastic structure models

Lastly, a botanically correct architectural structure, including positions of buds that flush to produce growth (Fig. 2), is preferable for starting of simulation.

5 Discussion

TLS, yet currently largely limited to the detection of static and transient patterns between the diversity of vegetation structure and other community features (Davies and Asner, 2014; D’Urban Jackson et al., 2020).

The availability of structural data for a wide range of species, across a wide range of habitats is key for shifting the focus of FSPMs from a practical use to a more paradigmatic one.

  • Need to establish a standard.

Zhang and DeAngelis (2020) state that although FSPMs are set up in such a way that they should be able to address many of the same questions as population-level agentbased models, it is not yet possible to apply FSPMs at the landscape level. It may not be even desirable to use FSPMs at that level.

  • Opposite attitude?

Of the 21 different architectural forms described by Hallé and Oldemann, only a handful have been featured in published FSPMs (Hallé et al., 1978). However, to exchange architectural sub-models, FSPMs could be better suited to deal with input parameter values. Many models apply a formalism [L-systems (Lindenmayer and Prusinkiewicz, 1990) in the first place] to deal with morphological development.

This opens possibilities of automatic transfer and translation of architectural information. Further, there exists a formalism to code architectural information, the multiscale tree graph (MTG, Godin and Caraglio, 1998), that can be used to transfer architectural information between FSPMs (Boudon et al., 2012).

  • Lastly, increasing the use of TLS-derived data is one way of overcoming the lack of structural forms found currently in FSPMs; however, only one published FSPM, HELIOS, features open-source TLS integration software (Bailey, 2019).
  • For FSPMs to be readily scaled up to the population level, a dedicated TLS pipeline must become an integral part of FSPM development.
  • Benchmarking and comparison have long been a shortcoming of FSPM development, as noted in several reviews (Louarn and Song, 2020; Zhang and DeAngelis, 2020).
  • Indeed, model complexity, inconsistency between sub-model representation, varied input parameters and a lack of open-source code has hindered a thorough evaluation of models (Table 1).
  • In the area of below-ground FSPM development. Schnepf et al. (2020) issued a call for participation for a collaborative effort to compare root FSPMs via a two-step system, Aboveground FSPMs could greatly benefit from a similar approach, and it will be important in upcoming years to improve communication between groups to tackle this issue.
  • Schnepf, Andrea, et al. “Call for participation: collaborative benchmarking of functional-structural root architecture models. The case of root water uptake.” Frontiers in Plant Science 11 (2020): 505466. https://doi.org/10.3389/fpls.2020.00316

Evers et  al. (2019) suggested guidelines for creating mixedspecies FSPMs, noting that any neighbourhood interactions should emerge only as a result of individuals requiring and acquiring resources in a given community, and not as an explicit process in the model. The use of mixed-species FSPMs is a relatively recent direction of research but the empirical observations from TLS studies of vegetation dynamics can hopefully guide model development.

6 Outlook and Conclusions

  • First, FSPM research already spans many disciplines, including life sciences, computer science and mathematics. However, effectively using FSPMs for forest communities requires coordination between plant ecophysiologists, ecologists and evolutionary biologists.
  • Second, efforts must be made to enhance the generality of FSPMs.
  • Third, TLS data have largely been collected for ‘tree-like’ plants. For a comprehensive understanding of woody plant form, TLS campaigns need to take place in a variety of forest environments, including tropical dry forests.

You May Also Enjoy