Vegetation structure represents a complex, extraordinarily effective
and even beautiful solution to the problem of survival and adaptation
on evolutionary time scales. Trees are perhaps the most dramatic
example of this - slow-growing and static and yet immensely resilient
and adaptable across diverse environments, disturbances and pressures
and forming complex multi-layered communities. Trees comprise a
compendium of individual variations, both inter- and intraspecific,
overlaid on the fundamental observable property of tree architecture:
the 3D shape and arrangement of the above- (and less visible but
equally important) below-ground part of the tree.

There have been many ingenious attempts to relate tree form to ecological
function, from Da Vinci’s observations of self-similarity and scaling,
D’Arcy Wentworth Thompson’s explorations of allometry, tropical tree
archetypes, fractal approaches and many more. Alongside these
attempts to describe, classify and generalise observations of structure, there have also been parallel efforts to explain tree function in terms of its relation to structure. This has led to numerous ingenious and varied approaches, from specific to extremely general e.g. via pipe model theory and hydraulics, mechanical constraints, functional-structural relationships, through to very general theories of metabolic function and thermodynamics. A major limitation that has prevented progress in testing these theories relating structure and function, has been the
almost complete lack of precise, detailed measurements of 3D canopy
structure. Measuring canopy structure accurately, repeatably and for anything other than small trees or low stature vegetation, has been very difficult and time-consuming. But this is changing; we are on the cusp of being able to
make routine 3D measurements of canopy structure at the leaf and
branch level, from new ground-based and UAV lidar methods, as well as a resurgence of photogrammetric methods. These 3D measurements in turn, are driving new developments in remote sensing observations of structure, from airborne and space-based lidar and radar. These larger-scale measurements are likely to revolutionise how we understand canopy ecosystem properties, from fine to large scale.

Here, I discuss some of these new observations, particularly terrestrial lidar scanning (TLS) but also UAV lidar and other measurements including Structure from Motion (SfM) and even acoustics. I discuss how these measurements are helping answer questions such as: what are the most important aspects of canopy structure? What do we lose by ignoring or approximating it at different scales, as we do routinely? I also touch on how 3D models are potentially key to exploiting these 3D measurements more effectively, particularly in helping to quantify more general relationships between structure and function and exploit new space observations.

Leaf area index (LAI) and fractional vegetation cover (FCover) are Essential Climate Variables, mechanistically linked to fluxes of water, energy and gases within ecosystems, and between biosphere and atmosphere. Vegetation canopies regulate primary productivity, microclimates and water balances that characterise ecosystem state and function. Canopy structure is thus not only relevant for the mitigation of climate change but also for the maintenance of biodiversity and ecosystem services.

Here, we introduce the Global LAI Project (https://globallai.wordpress.com/), an interdisciplinary collaborative network compiling field measures of tropical forest canopy structure using consistent data protocols (Pfeifer et al. 2012, 2014, 2016). Measured using hemispherical photography, our database currently encompasses 887 plots from protected and unprotected forest and woodland ecosystems across the four major tropical realms (Fig. 1). We tested whether a forest’s protection status affects the response of canopy structure to seasonal water deficits, mitigating or accelerating forest canopy degradation towards reduced canopy cover stages.

We used linear-mixed effects models to predict LAI and FCover (scaled between 0 and 100) as functions of seasonal water deficit (length and severity of the dry season calculated from Worldclim version 2 data, Fick & Hijmans 2017), interacting with IUCN protection status as derived from the World Database on Protected Areas (https://www.protectedplanet.net/c/world-database-on-protected-areas). Acknowledging that forests will differ in traits reflecting adaptations to macroclimate in the four biogeographic regions, we assumed differing random intercepts for each biogeographic region.

The fixed effects of seasonal water deficit and protection status explained 31% and 38% of the variation in LAI and FCover respectively, increasing to 41% and 40% conditional on random intercepts. LAI intercepts are lower in Africa (2.3) compared to Asia (3.0), South America (3.2), and Australia (3.1), while FCover intercepts were similar across regions. Forest protection was associated with canopy resilience to seasonal water deficit: both LAI and FCover respond negatively to prolonged seasonal drought, but the slopes of these relationships are 58% (LAI, Fig. 1) and 48% (FCover) shallower at protected sites compared with unprotected sites.

Research on forest canopies is expanding providing evidence for how canopies respond to climate - disturbance feedbacks thereby fundamentally altering the way forests look, work and function. Our data provide a snapshot of forest canopies along climatic and disturbance gradients. We want to use this session to collaboratively explore how field measurements can integrate with remote sensing and modelling to monitor and map canopy responses to environmental change (sensu Essential Biodiversity Variables) and to predict canopy dynamics as a basis for conservation and restoration of forest functioning under changing climate and land use pressures.

References:
Fick, S.E. and R.J. Hijmans, 2017. Worldclim 2: New 1-km spatial resolution climate surfaces for global land areas. International Journal of Climatology 37: 4302-4315.

Pfeifer, M. et al. 2012. Leaf area index for biomes of the Eastern Arc Mountains: Landsat and SPOT observations along precipitation and altitude gradients. Remote Sensing of Environment 118: 103-115.

Pfeifer, M. et al. 2014. Validating and linking the GIMMS leaf area index (LAI3g) with environmental controls in tropical Africa. Remote Sensing 6: 1973-1990.

Pfeifer, M. et al. 2016 Mapping the structure of Borneo's tropical forests across a degradation gradient. Remote Sensing of Environment 176: 84-97.

Many Essential Biodiversity Variables (EBVs) are linked to ecosystem structure and functioning. LiDAR is an active remote sensing technology that facilitates measuring forest structure in 3D with high detail. In the recent decade, terrestrial laser scanning (TLS) has demonstrated its ability to measure simple structural parameters such as diameter at breast height (DBH) and height, as well as more complex structural parameters like forest aboveground biomass (AGB) or the vertical distribution of plant area volume density. A TLS-driven approach supports the need for accurate, effective and nondestructive methods to assess forest structure at plot scale (typically up to 1 ha). Field inventory plots have been used consistently to extrapolate the understanding of forest structure at plot level to landscape-to-regional scales. This is under the explicit assumption that field plots represent their surrounding environment. The use of airborne laser scanning covering Andes-to-Amazon landscapes has demonstrated significant mean biases in forest canopy structure and AGB and the use of high-resolution remote sensing has been put forward as a vital step towards extrapolating field plot forest dynamics to larger scales.

Recent advancements in the development of unmanned aerial vehicles (UAVs) enables the use of LiDAR sensors from this platform, allowing 3D mapping of larger areas at a fraction of the cost of airborne LiDAR. Fusion of UAV and terrestrial LiDAR data can potentially be an important tool for the 3D mapping of forest structure at plot to landscape levels, which will be key for the reduction of large scale measurement and scaling uncertainties in forest structure and carbon estimates. The combined use of both sensors has not been assessed to date, largely due to the novelty of available UAV LiDAR systems that can be employed in an operational context. We present a unique and novel high-resolution dataset of co-incident TLS and UAV LiDAR data collected in 2018 in four different sites in Australia. Different tropical forest ecosystems were covered, ranging from savanna to tropical rainforest. TLS data were collected with a RIEGL VZ-400 scanner, and UAV LiDAR (UAV-LS) data were collected with the RIEGL VUX-SYS onboard the RiCOPTER platform. Each of the sites had a 1 hectare core-plot that was scanned with TLS. High density UAV-LS (~1000-5000 pts/m2) was collected for an approximate 300 x 300 m area around the core-plot and low density UAV-LS LS (~200-1000 pts/m2) covered an area of approximately 500 x 500 m. A feature matching algorithm (Multi Station Adjustment, available within the RIEGL software RiSCAN PRO) is used to align the different datasets.

We will use the co-incident TLS-UAV LiDAR of the core-plot as calibration/validation data for the landscape scale metrics. Metrics derived from UAV-LS will be benchmarked against those derived from TLS. Following this approach, we will combine the very high spatial detail of the TLS data with the larger spatial scale of the UAV LiDAR. We will focus on the structural forest metrics related to the vertical distribution of canopy elements and tree parameters such as height, DBH, crown dimensions and AGB estimates. Spatial analysis of these 3D structural metrics will allow us to understand the spatial patterns of tree structure and evenness within the 25 ha plots. This information will be crucial for more efficient forest management, but also for better understanding the spatial variation of forest structure, AGB and related EBVs.

Large uncertainties in tree and forest carbon estimates undermine countries’ efforts to accurately estimate aboveground biomass (AGB) for national monitoring, measurement, reporting and verification of emission reductions in forested landscapes. Biomass estimates, although much improved, still rely on destructive sampling; large trees are under-represented in datasets; crown dimensions are typically not considered, and allometric models are often inaccurate when transferred between regions – which all leads to uncertainties and systematic errors in biomass estimations. Moreover, the need for models to address national inventories (following IPCC Tier 3), requires the creation of national allometric models which is traditionally done though destructive sampling. We earlier used terrestrial laser scanning (TLS) to test the accuracy of existing models (Calders et al., 2015; Gonzalez et al., 2018), and now we propose the use of TLS to develop local allometric models without felling trees.
Here we (1) assessed the accuracy of TLS-derived tree metrics (diameter at breast height - DBH, height, crown width, and AGB) and (2) developed local allometric models to estimate tree AGB in Guyana based on tree parameters obtained from TLS point clouds from 72 tropical trees and wood density. We validated our methods and models with data from 26 destructively harvested trees.
We found that TLS-derived DBH was slightly lower, total tree height was higher, and crown width and AGB were not different from field-measured values, even with the presence of hollow and irregularly shaped trees. The assessed pantropical models underestimated AGB by 5 to 13 %. An older pantropical model —Chave et al. (2005) without height— consistently performed best among the pantropical models tested (R² = 0.89). Our best TLS-derived allometric models included crown diameter, and provided more accurate AGB estimates (R² = 0.92–0.93) than traditional pantropical models (R² = 0.85–0.89). On large trees (diameter > 70 cm), our TLS-derived models with crown diameter overestimated AGB by only 1 %, whereas pantropical models underestimated large tree AGB by 6 to 25 %. Our study shows that TLS-derived AGB estimates were not different from destructively harvested samples. Our methods also demonstrate that tree height is difficult to measure, and the inclusion of height in allometric models consistently worsened AGB estimates.
Our study has advanced the use of TLS methods to estimate tree metrics and explored the accuracy of field and TLS-derived methods to develop local allometric models. Moreover, our approach can be used to select the most appropriate published allometric model without destructively harvesting validation data. Interestingly, our study shows that locally developed models are not always better than pantropical models, but this could not be known without destructive or TLS-derived validation data on true AGB. Our findings support our goal of improving tropical forest biomass estimates and can be applied to upcoming remote sensing missions such as GEDI and BIOMASS.

We know that the forests of the tropics are changing fast. We have over the past decades developed the satellite technology and methods methods for monitoring deforestation reasonably well, but deforestation covers only a small fraction of the area of forest change occurring annually in the tropics (Mitchard, 2018). A much larger area is thought likely to be degraded annually, and a still larger area is taking in carbon rapidly while recovering from past disturbance. Finally, the evidence from field plots and atmospheric inversion studies is that intact tropical forest is actively sequestering carbon, increasing in biomass by ~0.4% of total biomass each year (except years with abnormally low precipitation or high temperatures, when this sink reverses).

It is essential that we develop the satellite technology and methods to map these changes, so that we can understand their magnitude and how they very in space and time. Only with such maps can we design and implement policy to for example halt degradation, and only with such data can we correctly parameterise and test couple climate models.

We believe that existing satellites, especially ESA's Sentinel satellite series, collect sufficient daily information to at least produce estimates of the carbon loss and accumulation rates from deforestation and forest degradation, and possibly say something about the growth of intact forest (though for the latter we may have to wait for ESA's BIOMASS mission in 2021). But we believe we do not currently have the correct algorithms to analyse such data, nor the ground data to develop and test these algorithms.

In this talk we will present the first results from our ERC funded project the Tropical Forest Degradation Experiment (FODEX), which started in mid-2018 and collects its first field data in a logging concession in Gabon in January/February 2019. This project aims to collect the field datasets necessary to use these satellite products to their full potential, through measuring tens of thousands of trees in ~50 forest inventory plots in logging concessions in Gabon and Peru. In these we will further collect Terrestrial Laser Scanner (TLS) data, and LiDAR and cm-resolution multispectral data from a project-owned UAV. These plots will then be logged and the measurements repeated.

We will present here our first results about the ability of dense time series of Sentinel-1 C-band SAR data, Sentinel-2 optical data, and other satellite technologies (for example ALOS PALSAR L-band SAR data) to map forest carbon stocks and changes in these stocks. We will further look to the future in terms of analysis methods, data fusion, and the longer-term outputs from the FODEX experiment.








References

Mitchard, E.T.A. 2018. The Tropical Forest Carbon Cycle and Climate Change. Nature, 559, 527-534.

Increasing evidence points towards soil moisture availability as the single most important factor for variations of vegetation productivity globally [1,2] and between years [3,4] with substantial effects on fluctuations in the atmospheric CO 2 growth rate [5].
During droughts in semi-arid regions, typically associated with only sparse tree coverage, photosynthesis is strongly reduced. Forests cover about 30% of the global land and sequester about half of the global annual carbon uptake and still, there is significant understanding missing on how short-term environmental fluctuations affect the metabolism and carbon uptake of forests. Established satellite-based observations of vegetation greenness do not allow to consistently monitor photosynthesis in forests. Current observational capabilities are limited as the trees do not show immediate visible signs of stress or health. Thus, the general understanding of negative impacts of water scarcity on vegetation vigor and carbon uptake appears to be based on the non-forested ecosystems only.
In our study, we re-evaluate how plant productivity varies globally with changing soil moisture along a tree cover gradient. For that, we use three different types of remote sensing proxies for vegetation productivity (16 day temporal and 1° spatial resolution) : 1) traditional vegetation indices sensitive to greenness (MODIS EVI, NDVI, NIRv based on surface reflectances MCD43C4), 2) sun-induced chlorophyll fluorescence as a proxy for photosynthesis (based on GOME2), and 3) data-driven estimates of gross primary productivity based on machine learning upscaling of flux-tower data (FLUXCOM). We find that, when soil water content is reduced, both fluorescence and the modelled gross photosynthesis reflect a consistent “dipole” like pattern of enhanced photosynthesis for high tree cover and reduced photosynthesis in regions with low tree cover. In contrast, the traditional satellite-based greenness indices do barely show this pattern of increased forest productivity during below average soil moisture. Our work therefore represents an important incentive to reconsider the differential responses of ecosystems along a tree cover gradient where higher illumination (and temperature) concurrently to and despite of reduced soil water content actually enhances photosynthesis in forests on average, and vice versa. Our study also illustrates the importance of differentiating indicators of plant greenness from those of photosynthesis for the monitoring and understanding of ecosystems.



[1] Schwalm et al. 2012, Nature Geoscience, doi:10.1038/ngeo1529;
[2] Vincente-Serrano et al. 2013, PNAS, doi:10.1073/pnas.1207068110
[3] Ahlström et al. 2015, Science, doi: 10.1126/science.aaa1668;
[4] Jung et al. 2017, Nature, doi:10.1038/nature20780)
[5] Poulter et al. 2014, Nature, doi:10.1038/nature13376