The NERC Field Spectroscopy Facility (Edinburgh, UK) has provided ground based spectroscopic instrumentation and expertise for UK and international researchers for 16 years, often in support of airborne spectroscopic surveys. A gap, however, has previously existed in the spatial scale of the measurements which the facility could provide. Hyperspectral imagers can show < mm spatial variations in reflectance at ground level, but provision of plot or even landscape scale was limited to support of airborne campaigns. Deploying hyperspectral imagers on a drone based platform would be a step change in data acquisition, allow researchers seeking FSF support too easily scale up hyperspectral reflectance indices over large areas, perform individual-level plant species mapping, monitor plant disease or stress, measure sun induced fluorescence, detect invasive species or perform spatial investigation of different plant physiological traits, all with the potential for satellite data integration.

With recent developments in drone and sensor technology, we have recently been able to combine our expertise in field spectroscopy, as well as our extensive instrument library, with new and emerging UAV technologies, to provide a dedicated Field Spectroscopy Facility (FSF) UAV Suite, which can be loaned to UK and international research groups. To support the varying demands of research questions, the suite provides a number of dedicated drone and sensor platforms, which – due to the modular nation of the UAVs and sensors used – allow for novel, custom solutions. All sensors provided are calibrated and quality assured by FSF staff at the facility’s optical dark lab, and data processing and piloting services are also provided.

In this presentation, we outline select campaigns which the FSF UAV suite has so far been, or is preparing to be, involved in. This includes the use of a UAV mounted hyperspectral imager with LIDAR capability used in support of agricultural drought measurements; the use of the same drone-sensor for the detection of marine plastics from space; the use of a multispectral camera (equipped with the same wavelength intervals used by Sentinel-2) for UAV archaeological surveying; and the planned use of a UAV mounted solar induced fluorescence sensor package as part of the ESA FLEX mission.

The restoration of degraded tropical forests has potential for sequestering large amounts of atmospheric carbon, either through natural regeneration or direct planting. For example, it was estimated that previously logged forests in the Malaysian state of Sabah could gain 362.5 TgC if allowed to fully recover (Asner et al., 2018). A key requirement for funded carbon projects is the establishment of accurate baseline aboveground carbon (AGC) measurements and subsequent monitoring, reporting and verification. At the same time, recent studies emphasise the importance of including Indigenous and local communities in forest decision-making and management, highlighting the significant positive impacts this has on restoration and conservation outcomes (Dawson et al., 2021) and the potentially deleterious impacts that forgoing community involvement can have on tree cover and livelihoods (Höhl et al., 2020). International NGOs advocate that communities should play an integral role in the monitoring and verification of forest carbon stocks (e.g., GOFC-GOLD, 2016).

A common airborne approach to measuring AGC uses LiDAR-derived forest stand metrics for calculating biomass. However, the acquisition of high-resolution LiDAR data is challenging for community-based projects with limited technical or financial resources, and many freely available remote sensing products lack the spatial resolution needed for detailed community-scale mapping (< 5ha). Here, we assess lightweight, off-the-shelf drones as a potential solution. Comparatively inexpensive and straightforward to operate, these drones enable users to quickly generate high-resolution, geo-referenced RGB imagery which can be combined with structure from motion (SfM) photogrammetry—the generation of 3D point clouds from overlapping 2D images—to estimate canopy heights and AGC. However, there are knowledge gaps around both the feasibility of these consumer-grade technologies for generating accurate, community-scale carbon stock estimates and the associated uncertainties.

In this presentation, we assess a methodology for generating community-scale AGC estimates from drone imagery, applying it to two previously logged lowland forest restoration sites in Sabah, Malaysia (< 2ha each). Using an inexpensive, off-the-shelf drone and GPS unit, we gathered high-resolution RGB imagery for both sites, processing them using opensource SfM and GIS packages to generate georeferenced canopy height models. We evaluated several AGC estimation methods that employ regional allometric equations and drone-derived metrics, including one for Southeast Asian tropical forests which relies on a single input metric derived from the 3D point clouds—mean top-of-canopy height. We compare this to field data from botanical plots and freely available satellite-based biomass estimates for the sites. We discuss the overall ability and applicability of drone-based SfM methods to produce realistic baseline AGC values for community-based restoration and carbon monitoring projects, including the uncertainties and the logistical challenges to their successful implementation by Indigenous and local communities.

References

Asner GP, Brodrick PG, Philipson CD, et al. (2018) Mapped aboveground carbon stocks to advance forest conservation and recovery in Malaysian Borneo. Biological Conservation 217: 289–310.
Dawson NM, Coolsaet B, Sterling EJ, et al. (2021) The role of Indigenous peoples and local communities in effective and equitable conservation. Ecology and Society 26(3): art19.
GOFC-GOLD (2016) A sourcebook of methods and procedures for monitoring and reporting anthropogenic greenhouse gas emissions and removals associated with deforestation, gains and losses of carbon stocks in forests remaining forests, and forestation. GOFC-GOLD Report version COP22-1. Wageningen.
Höhl M, Ahimbisibwe V, Stanturf JA, et al. (2020) Forest Landscape Restoration—What Generates Failure and Success? Forests 11(9): 938.

This work studies the deciduous tree degradation in the historical landscape park regions „Park Sanssouci“, „Park Babelsberg“ and „Klein Glienicke“ maintained and preserved by the Prussian Palaces and Gardens Foundation Berlin-Brandenburg (SPSG) in Potsdam and Berlin.
These historical park regions with old deciduous trees planted in the 19th century received the „UNESCO World Cultural Heritage Site“ status in 1990. The park regions encompass a total area of 2064 hectares. The region represents one of the largest UNESCO world heritage sites in Germany. “72% of the contracting states of the UNESCO World Heritage Convention report damages that can be linked to climate change” (Bernecker 2014). The deciduous forest is dominated by old beech trees (Fagus sylvatica) and oak trees (Quercus robur). Under climate change with predicted extreme dry and warm summers it will be a “tremendous challenge” (Schellnhuber & Köhler 2014) to preserve these sites – this is especially true with some of the trees planted in the 19th century on sandy and/or exposed soils with low soil water storage capacity. The very warm and dry summers of 2018 and 2019 in Germany were likely caused by the arctic amplification on the mid-latitude summer circulations that lead to shifted jet-streams and “amplified quasi-stationary” waves with “persistent hot-dry extremes in the mid-latitudes” (Coumou et al. 2018). Resulting low soil water availability was likely the reason for increased defoliation and dying of beech trees in some regions of the historical park areas and other regions in Germany.
This work is performed in cooperation with the park administration (SPSG) and the KERES project (Protecting Cultural Heritage from Extreme Climate Events and Increasing Resilience) and maps the per year degradation of the park regions using spatial very high resolution multispectral UAS (Unmanned Aerial System) data with multiple coverages per year and per park area. We defined three different study sites and captured multispectral data and photogrammetrically derived canopy height data from dense point clouds in mid-summer for all sites. Individual tree crowns were delineated with a nested watershed-algorithm based on a spatial high resolution canopy height model (CHM) and a surface model-based shadow-casting layer – all derived from Phantom 4 Multispectral RTK (Real Time Kinematic) data. We defined four different damage categories and trained a tensor flow convolutional neural network model (CNN) based on 3x3 - 5x5 m training plots using multispectral data and derived spectral channel ratios (red-edge ratio, NDVI and green-red ratio). Training and network model optimization was performed on the 120 ha park region of “Klein-Glienicke” and applied to the park areas of Potsdam Sanssouci and Babelsberg. We implemented a per tree crown classification of percent defoliation on 8 cm pixel resolution and delineated tree crown degradation for all tree crowns in 4 different categories. We found that transferability of the trained CNN model to other park areas was not directly possible and complicated by over-classification likely due to modified shadow area distribution. Training with additional training data however showed better results. Validation was done with reference tree inventory data from SPSG for all park areas.

Calibration of aboveground biomass (AGB) products produced with upcoming missions like BIOMASS and GEDI require accurate AGB estimates, preferably across hectometric reference sites. Terrestrial laser scanning (TLS) based techniques for individual tree AGB estimation have proven to be unbiased predictors, even for large trees. However, data collection is labour- and time-intense, so that upscaling approaches would be desirable. Unoccupied aerial vehicle laser scanning (UAV-LS) can collect high density point clouds across hectares, but past studies have shown limitations in terms of trunk measurements, which are typically involved in allometric model calibration.

In this study, we propose the combination of TLS and UAV-LS for AGB estimation at reference sites. We included data from four sites located in temperate mixed, wet tropical and wet-dry tropical savanna forests. For each site, coinciding TLS and UAV-LS data was collected, and the point clouds were co-registered. Individual tree point clouds were automatically extracted from the TLS and manually quality-controlled with > 170 trees per site. Subsequently, Quantitative Structure Models (QSMs) were built and reference individual tree AGB was determined from tree wood volume estimates derived and wood density databases. For the UAV-LS, a fully automatic tree segmentation routine was applied and the UAV-LS trees that corresponded to the TLS reference trees were identified. A range of individual tree traits like height and crown diameter were estimated based on the UAV-LS trees. Finally, different AGB modelling strategies were tested using published allometric models, and locally calibrating models with parametric and non-parametric regression techniques. All strategies were cross-validated with leave-one-out cross-validation. Individual tree AGB RMSE ranged between 0.30 and 0.69 Mg across the sites. When summing up individual tree AGB to assess bias in estimation of cumulative AGB, as would be performed to estimate plot-scale AGB, the strategies showed diverging patterns that resulted not always in optimal estimation. However, the non-parametric modelling strategy could robustly produce biases < 5% across the sites.

Even though combined TLS and UAV-LS have high requirements in terms of investment in instruments and training of personnel, this study supports their potential for non-destructive AGB estimation. This is relevant for the calibration and validation of space-borne missions targeting AGB estimation at reference sites.

With today’s changing climate comes not only rising temperatures, but also shifts in precipitation patterns which in some regions can result in drought stress for various tree species in particular European Beech (Fagus sylvatica). Recent summer droughts such as in 2018 and 2019 have caused an increased amount of crown defoliation and branch die back which is raising concern especially in regions where minimal rainfall is already an issue. A better understanding of beech drought tolerance and adaptation to climate change could aid in the transition to more resilient forests, which is currently an important research topic at intensive forest monitoring sites. Such sites are equipped with devices designed to make long-term physiological measurements for the purpose of better understanding tree water availability and the influence of meteorology on forest dynamics. Highly accurate sensors such as dendrometers, sap-flow and leaf temperature sensors aid in the quantification of drought related stress factors on an individual tree basis. With recent advancements in Unmanned Aerial Vehicles (UAVs) and mounted sensors, alongside machine learning algorithms and increased computing capacity, a new aspect is added to terrestrial individual tree measurements in terms of the interpretation and classification of spectral information acquired from the upper tree crown. The detection of water status in leaves can enable the early diagnosis of drought stress with the use of multispectral sensors including thermal data. The use of thermal imagery for the purpose of leaf water detection is based on the principle that transpiring leaves will result in cool ambient air temperatures due to open stomata whereas closed stomata will result in an increase in leaf temperature. The acquisition of thermal data of upper canopy leaves however, can prove challenging as thermal data can be influenced considerably from varying solar radiation intensities and other meteorological factors. In this study we explored the possibility to calibrate thermal imagery acquired from the Micasense Altum UAV mounted sensor in order to quantify tree drought stress status from single-shot multispectral imagery. As apposed to the creation of Orthomosaics derived from imagery based on a gridded flight pattern, we implemented single-shot ultra-high-resolution imagery of individual trees crowns corrected through affine transformation and radiometric calibration. Co-registered multi-temporal layer stacks are stored in 4-dimensional data cubes comprised of individual multispectral bands and derivates such as vegetation indices, calibrated thermal data, as well as layers depicting meteorological data based on daily temperature sums, global radiation and precipitation sums recorded at the time of image acquisition. Validation of tree water status during image acquisition was accomplished with dendrometers which capture highly accurate sub-hour stem shrinkage patterns on a multi-seasonal level. We show in this study that UAV-based data cubes, calibrated at intensive forest monitoring sites, can be implemented for the rapid acquisition of sample-based ground truthing data enhanced with local weather station data. UAV data cubes can serve the purpose of assessing drought stress levels in areas outside of intensive monitoring plots as well as training and validation datasets for satellite remote sensing platforms. Furthermore, data cubes offer a simplified data storage system enabling improved access for analysis. The use of UAV-based data cubes in a standardized form could prove decisive in the enhancement of ground-truthing methods when implemented on a national and multi-national level.