Displays

Forests become more important in times of changing climate, increasing demand of renewable energies and natural resources, as well as the high demand of information for economical and management issues. Several previous studies were carried out in the field of forest plantations but there is still a gap in knowledge when it comes to natural mixed forests, which are ecological complex due to varying distributions and interaction of different species. The applicability of Unmanned Aerial Vehicles (UAVs) for forest applications by using image analysis became a common tool because it is cost-efficient, time-saving and usable on a large-scale. Additionally, technologies like Deep Learning (DL) fasten the proceeding of a high number of images. Deep learning is a relatively new tool in forest applications and especially in the case of natural dense mixed forests in Japan. Our approach is to introduce the DL-based ResNet50 network for automatic tree species classification and segmentation, which uses transfer learning to reduce the amount of required data. A comparison between the ResNet50 algorithm and the common UNet algorithm, as well as a quantitative analysis of model setups are presented in this study. Furthermore, the data were analysed regarding difficulties and opportunities. We showed the outperformance of UNet with a DICE coefficient of 0.6667 for deciduous trees and 0.892 for evergreen trees, while ResNet 50 was reaching 0.733 and 0.855. A refinement of the segmentation was performed by the watershed algorithm increasing the DICE coefficient to values of up to 0.777 and 0.873. The results of the transfer learning analysis confirmed the increasing accuracy by adding image classification data basis for the model training. We were able to reduce the number of images required for the application. Therefore, the study showed the applicability and effectiveness of those techniques for classification approaches. Furthermore, we were able to reduce the training time by 16 times for the ResNet 50 performance and by 3.6 times with the watershed approach in comparison to the UNet algorithm. To the best of our knowledge this is the first study using deep learning applications for forestry research in Japan and the first study dealing with images of natural dense mixed forests.

How to cite: Kentsch, S., Lopez Caceres, M. L., and Diez Donoso, Y.: Tree species classification by using computer vision and deep learning techniques for the analysis of drone images of mixed forests in Japan, EGU General Assembly 2020, Online, 4–8 May 2020, EGU2020-197, https://doi.org/10.5194/egusphere-egu2020-197, 2020.

Although various ways of defining forests exist, non of them is eligible on assessing every tree -growing outside forest- on the landscape. In the last decades, forestry and land management sectors have become increasingly aware that Trees Outside Forests (TOF) are critical non-forest tree resources to ensure environmental, economic, social and cultural services and functions. The importance of TOF varies in international, national and local levels. Recently, international programmes have been established to strengthen the services and functions of TOF: sustainable land management, carbon capturing and storage on climate change mitigation and improving local economies. Therefore, in the past years countries have started to take action for assesing their TOF resources on different scales.  

Only little research has been conducted on TOF in Switzerland, yet the explicit spatial distribution of TOF in the landscape is poorly understood and their extent and tree biomass are unknown. Nowadays, remote sensing technologies have opened new opportunities to fill this knowledge gap, and countrywide data sets of TOF have become more feasible. 

The present research aims to introduce a highly automated method to derive extent, spatial distribution and biomass of TOF in different land use classes: Agriculture, Urban, and Non- Agriculture/Urban for the whole of Switzerland. 

The entire process of identifying TOF is done in Python using routinely acquired countrywide remote sensing data, i.e. Vegetation Height Model (Ginzler and Hobi 2015), CORINE Land Cover/Use map and the Forest Mask of Switzerland (Waser et al. 2015) and based on the decision tree algorithm developed by FAO-FRA (Foresta et.al., 2013). The primarily applied criterias are the Presence of Trees on the land, Land Use, and Spatial pattern of Trees. After the application of primary criterias, a set of thresholds were applied as following: the minimum canopy cover threshold: 5% (if trees only), 10% if combined cover is trees and shrubs, minimum area 0.05 ha., tree line lenght 25 m, and tree line width 3 m. 

The present study aims to complement forest data obtained from the Swiss National Forest Inventory and enables to derived relevant TOF parameters such as tree species distribution, biomass and carbon sequestration potential. Moreover, the proposed method is relevant to help other countries to create their own data sets on non-forest tree resources as an input to energy, environment, forest policy making, and wood industry decision making and to contribute to better cope with the challenges of changing climate and environment. Currently, the potential of Sentinel-2 imagery is being tested.

Keywords: Trees Outside Forest, Wall-to-wall, Vegetation Height Model

Reference: Hubert de Foresta, Eduardo Somarriba, August Temu, Désirée Boulanger, Hélène Feuilly and Michelle Gauthier. 2013. Towards the Assessment of Trees Outside Forests. Resources Assessment Working Paper 183. FAO Rome.
Ginzler, C., Hobi, M.L., 2015. Countrywide Stereo-Image Matching for Updating Digital Surface Models in the Framework of the Swiss National Forest Inventory. Remote Sensing, 7, 4343-4370.
Waser, L.T., Fischer, C., Wang, Z., Ginzler, C., 2015. Wall-to-Wall Forest Mapping Based on Digital Surface Models from Image-Based Point Clouds and a NFI Forest Definition. Forests, 6, 4510-4528.

How to cite: Malkoc, E. and Waser, L. T.: New Opportunities for Highly Automated Countrywide Assessment of Trees Outside Forests in Switzerland, EGU General Assembly 2020, Online, 4–8 May 2020, EGU2020-4752, https://doi.org/10.5194/egusphere-egu2020-4752, 2020.

The benchmarking project of image-based point cloud for forest inventory (SFM-Forest-Benchmark) was initiated in 2019 and supported by ISPRS Scientific Initiative 2019. The main goal of the project was the evaluation of the applicability of terrestrial image-based point clouds for forest inventories, the clarification of the potential and limitations of the state-of-the-art techniques, and the exploration of the best practices in practical field inventories. In the project, related tree parameter (i.e. tree position diameter at breast height - DBH) were derived from 14 algorithms and evaluated using field inventory data as a reference. In order to clarify the potential of terrestrial image-based point clouds, the results from the image-based point clouds were also compared to results derived from the best available point clouds obtained by terrestrial laser scanning (TLS).

The project is consisted of two phases. In the first phase, we established two research plots in each country (Austria, China, Czech, Finland and Slovakia), ten plots in total. The stem density ranged from 272 to 875 stems/ha and plot size ranged approximately from 700 to 2500 m². Dominant tree species across research plots were Norway spruce, European beech, bald cypress, Chinese tulip poplar, Scots pine, European silver fir and sessile oak. TLS, images and reference data acquisition were performed on each study site, where TLS data were acquired through multi-scan approach, images were taken in the stop-and-go mode, and tree positions and the DBHs were measured with a tachymeter and a calliper as field references. Images were processed with structure from motion algorithm within Agisoft Metashape software to final point clouds. The TLS data was pre-processed with RiProcess software. And, the co-registration of all three data sources (TLS, SFM, and reference data) was done with OPALS software.

In the benchmarking phase, we distributed point clouds to participants of the benchmark. Altogether 14 different research groups processed the data with own algorithms. The individual results are evaluated through the reference to clarify the applicability of the image-point clouds in deriving tree parameters, were compared to each other to reveal the state-of-the-art of technologies, and were benchmarked to the up-to-data the most accurate data from TLS to explore the strength and weakness of the image-based point cloud. In this presentation the first benchmark results will be presented and discussed.

All images and point clouds collected for this project will be available as open access data for non-commercial uses.

How to cite: Mokros, M., Hollaus, M., Wang, Y., and Liang, X.: SFM-Forest-Benchmark project: The benchmarking of image-based point cloud for forest inventory , EGU General Assembly 2020, Online, 4–8 May 2020, EGU2020-5822, https://doi.org/10.5194/egusphere-egu2020-5822, 2020.

Mangrove forest is considered as one of the pivotal ecosystems to near-shore environment health, adjacent terrestrial ecosystems and even global climate change migration. However, for past two decades, they are declining rapidly. In order to take effective steps to prevent the extinction of mangroves, high spatial resolution information of large-scale mangrove distribution is urgent. Recent study has indicated that a suitable pixel size for extracting mangroves should be at least equal to 10 m. Hence, Sentinel imagery (Sentinel-1 C-band synthetic aperture radar (SAR) and Sentinel-2 Multi-Spectral Instrument (MSI) imagery) whose spatial resolution is 10 m may hold great potentials to achieve this goal, but there are limited researches investigating it. Therefore, in this study, we will explore the potential of Sentinel imagery to extract mangrove forests in China on the Google Earth Engine platform. Specifically, our study was mainly conducted around 3 questions: (1) Which Sentinel imagery provides a higher accuracy for mangrove forest mapping, Sentinel-1 SAR data or Sentinel-2 multi-spectral data? (2) which combination of features from Sentinel imagery provides the most accurate mangrove forest map? (3) Compared to 30-m resolution mangrove products derived from Landsat imagery, how does 10-m resolution map improve our knowledge about the distribution of mangrove forest in China?

Our results show that: (1) The highest producer’s accuracies (the reason why using producer’s accuracy as an accuracy evaluation indicator here is that the omission errors in mangrove forest extent map are much larger than commission errors) of mangrove forest maps derived from Sentinel-1 and Sentinel-2 imagery are 91.76% and 90.39%, respectively, which means that the contributions of Sentinel-1 SAR and Sentinel-2 MSI imagery to mangrove mapping are similar; (2) The highest producer’s accuracy of mangrove forest map at 10-m resolution is 95.4%. The mangrove forest map with the highest accuracy is obtained by combining quantiles of spectral and backscatter bands, spectral index, and texture index derived from time series of Sentinel-1 and Sentinel-2 imagery, indicating that the combination of Sentinel-1 SAR and Sentinel-2 MSI imagery is more useful in mangrove forest mapping than using them separately; (3) In China, the total area of mangrove forest extent at 10-m resolution is similar to that at 30-m resolution (20003 ha vs. 19220 ha). However, compared to 30-m resolution mangrove products, the 10-m resolution mangrove map identifies 1741 ha (occupying 8.7% of total mangrove forest area in China) mangrove forests in size smaller than 1 ha, which are especially important to low-lying coastal zone. This study demonstrates the feasibility of Sentinel imagery in large-scale mangrove forest mapping and gives guidance to map global mangrove forest at 10-m resolution in the future.  

How to cite: Hu, L., Yao, W., Yu, Z., and Wang, L.: National-scale mangrove forest mapping by using Sentinel-1 SAR and Sentinel-2 MSI imagery on the Google Earth Engine Platform, EGU General Assembly 2020, Online, 4–8 May 2020, EGU2020-5305, https://doi.org/10.5194/egusphere-egu2020-5305, 2020.

Unmanned aerial vehicles-borne laser scanning (ULS) allows time-efficient acquisition of high-resolution point clouds on regional extents at moderate costs. The quality of ULS-point clouds facilitates the 3D modelling of individual tree stems, what opens new possibilities in the context of forest monitoring and management. In our study, we developed and tested an algorithm which allows for i) the autonomous detection of potential stem locations within the point clouds, ii) the estimation of the diameter at breast height (DBH) and iii) the reconstruction of the tree stem. In our experiments on point clouds from both, a RIEGL miniVUX-1DL and a VUX-1UAV, respectively, we could detect 91.0 % and 77.6 % of the stems within our study area automatically. The DBH could be modelled with biases of 3.1 cm and 1.1 cm, respectively, from the two point cloud sets with respective detection rates of 80.6 % and 61.2 % of the trees present in the field inventory. The lowest 12 m of the tree stem could be reconstructed with absolute stem diameter differences below 5 cm and 2 cm, respectively, compared to stem diameters from a point cloud from terrestrial laser scanning. The accuracy of larger tree stems thereby was higher in general than the accuracy for smaller trees. Furthermore, we recognized a small influence only of the completeness with which a stem is covered with points, as long as half of the stem circumference was captured. Likewise, the absolute point count did not impact the accuracy, but, in contrast, was critical to the completeness with which a scene could be reconstructed. The precision of the laser scanner, on the other hand, was a key factor for the accuracy of the stem diameter estimation. 
The findings of this study are highly relevant for the flight planning and the sensor selection of future ULS acquisition missions in the context of forest inventories.

How to cite: Bruggisser, M., Otepka, J., Pfeifer, N., and Hollaus, M.: Influence of ULS data acquisition characteristics on the achievable stem reconstruction accuracies, EGU General Assembly 2020, Online, 4–8 May 2020, EGU2020-7378, https://doi.org/10.5194/egusphere-egu2020-7378, 2020.

In semi-arid to arid South-west Morocco, the endemic argan tree (Argania spinosa) forms open woodlands that are the basis of a traditional agroforestry system involving rain-fed agriculture, pasturing of goats, sheep and camels, and oil production. Due to the high grazing pressure, the trees show various morphological traits and growth forms that are strongly related to browsing intensity. The overall appearance of Argania spinosa ranges from trees with a large, round crown and single trunk, over multi-stem, umbrella-shaped and hourglass-shaped trees to heavily condensed cone-shaped cushions.

30 test sites of 1 ha each in argan woodlands of different degradation stages were surveyed with an unmanned aerial vehicle (UAV) and RGB optical camera using a dedicated flight scheme for capturing full 3D tree shape at approx. 1 cm resolution. Structure-from-Motion (SfM)-photogrammetric processing yielded dense 3D point clouds as well as ultra-high resolution (1.5 cm) digital surface models (DSMs), terrain models (DTMs), crown-height models (CHMs) and orthophoto mosaics. Tree height and crown size were extracted from the CHMs, and 3D point-cloud characteristics (point density, profile shape/layer structure) and canopy structures were analysed within a geographical information system (GIS). Using field-based reference data on tree architecture and browsing features of 2494 trees, we were able to assign characteristic combinations of the GIS-derived structural parameters to three browsing-intensity classes and thus classify each argan tree via the architectural shape captured in its UAV-based 3D point cloud. We found that the majority of argan trees at the study sites are characterised by high browsing intensities. The small percentage of trees in the minimum browsing class are mostly inaccessible to grazing livestock. We conclude that UAV-based remote sensing has a high potential for mapping structural indicators of tree degradation by herbivore browsing in open woodland environments.

How to cite: Marzolff, I., Stephan, R., Kirchhoff, M., Seeger, M., Aït Hssaïne, A., and Ries, J. B.: UAV-based classification of tree-browsing intensity in open woodlands, EGU General Assembly 2020, Online, 4–8 May 2020, EGU2020-10301, https://doi.org/10.5194/egusphere-egu2020-10301, 2020.

Spectral imaging of vegetation for phenotyping is a fast-developing field that enables fast, objective and automated assessment of plant traits. Advances in instrumentation allow collection of ever more detailed observations using hyperspectral imaging. This technique captures the reflected light in 100+ wavelengths, compared to multispectral sensors which typically obtain 3 to 5 wavelengths. With machine learning and careful statistical analysis these data can be efficiently transformed into predictions of crop health, yield, etc. However, these instruments are costly to acquire and produce volumes of data which are expensive to handle and archive, and therefore we must ask the question whether/when the investment is worth it. In this case study, we assess the implications of using hyperspectral vs multispectral imaging when monitoring the effects of an invasive fungal pathogen on seedlings of southwestern white pine. We discuss the impacts in terms of the complexity level of the research goals. Firstly, we discuss the impact on the accuracy and timing of infection detection. Pre-symptomatic detection of infection is possible using hyperspectral. To what extent is this possible using multispectral? Next, what is the trade-off between the two spectral methods when predicting for symptom severity? And lastly, the study contains a third level of complexity, a variety in genotypes. Using hyperspectral we can successfully separate the genotypes. However, is there still a significant difference in reflectance between genotypes when using multispectral data? This study shows that the need for hyperspectral depends on the complexity of the research goal, and therefore collecting more data might not always be useful.

How to cite: Haagsma, M., Page, G., Johnson, J., Still, C., Waring, K., Sniezko, R., and Selker, J.: Is more data better? A comparison of multi- and hyperspectral imaging in phenotyping., EGU General Assembly 2020, Online, 4–8 May 2020, EGU2020-10673, https://doi.org/10.5194/egusphere-egu2020-10673, 2020.

Restoring forest ecosystems and predicting forest response to projected climate change have become an increasingly high priority for land managers. Restoration and management goals require accurate, quantitative estimates of vertical and horizontal forest structure, which has relied upon either field-based measurements, manned airborne, or satellite remote sensing datasets. We use unmanned aerial vehicle (UAV) image-derived structure from motion (SfM) models and high resolution multispectral and thermal orthoimagery to: 1) quantify vertical and horizontal forest structure at both fine- (< 4 ha) and mid-scales (4-400 ha) across a forest density gradient, and 2) quantify horizontal structure, health, and survival rates in a genetics experimental garden also with a density gradient. In both cases, we find that UAV multispectral and thermal image-derived SfM model estimates of individual tree height and canopy diameter are most accurate in low-density conditions, with accuracies degrading significantly in high-density conditions. In addition, UAV thermal images demonstrate significant differences in tree health and survival rates among various populations and genotypes within a single species.  Mid-scale estimates of canopy cover and forest density follow a similar pattern across the density gradient, demonstrating the effectiveness of UAV image-derived estimates in low to medium-density conditions as well as the challenges associated with high-density conditions.

How to cite: Sankey, T. and Belmonte, A.: UAV-derived Estimates of Vertical and Horizontal Structure across Forest Density Gradients, EGU General Assembly 2020, Online, 4–8 May 2020, EGU2020-11896, https://doi.org/10.5194/egusphere-egu2020-11896, 2020.

Tropical moist forest (TMF) provide essential ecosystem services_1,2. Fine-scale mapping and characterization of their disturbances are needed to support global conservation policies₃ and to accurately quantify their contribution to global carbon fluxes₄. However, limited information exists on their remaining extent and long-term historical changes.

We produced a wall-to-wall map of TMF cover dynamics at 30-meter resolution from 1990 to 2019. Each individual image of the full Landsat archive (~1 200 000 scenes) has been mapped using an expert system to allow all disturbances in the forest cover - including from selective logging activities and fires that are visible during a short period - to be depicted and characterized in terms of timing (dates and duration), sequential dynamics, intensity, and extent.

The performance of our disturbance classifier has been validated against 12 235 reference sample plots resulting in 9.4% omissions, 8.1% false detections and 91.4% overall accuracy. 

Our dataset depicts the TMF extent and patterns of disturbances through two complementary layers: a transition map and an annual change dataset. The transition map captures the resulting disturbance dynamics over the 30 years by depicting (i) remaining undisturbed forests, (ii) two types of degraded forests (corresponding mostly to either logged or burned forests), (iii) young forest regrowth, and (iv) deforested land that includes four subcategories of converted land cover: (a) water bodies (new dams and river flow changes); (b) tree plantations; and (c) other land cover that includes infrastructure, agriculture, and mining. The annual change dataset is a collection of 30 maps depicting - for each year between 1990 and 2019 - the spatial extents of undisturbed forests and disturbances.

We found that pan-tropical forest disturbances have been underestimated so far. For the first time at this scale, we discriminate deforestation from degradation and we underline the importance of the degradation process in tropical forest ecosystems. Our analysis shows the trends of deforestation and degradation by country, sub-region, and continent. Finally, we extrapolated the recent average rates of disturbances to predict the extent of the undisturbed TMF by 2050.

We will continue to update the TMF dataset with future Landsat data and intend to adapt the methodology to Sentinel 2 data (available since 2015) towards near real-time monitoring of TMF with a higher frequency of observations and finer spatial resolution.

1. Gibson et al. 2011 doi:10.1038/nature10425
2. Watson et al. 2018 Doi:10.1038/s41559-018-0490-x
3. Mackey et al. 2015 doi:10.1111/conl.12120
4. Mitchard E.T.A. 2018 doi

How to cite: vancutsem, C., Achard, F., Pekel, J.-F., Vieilledent, G., Carboni, S., Simonetti, D., and Gallego, J.: High-resolution mapping of tropical moist forest cover dynamics over the last 30 years, EGU General Assembly 2020, Online, 4–8 May 2020, EGU2020-13834, https://doi.org/10.5194/egusphere-egu2020-13834, 2020.

High radiation, low albedo, and limited evaporative cooling greatly affect canopy temperature in many semi-arid ecosystems. This makes the dissipation of excess energy essential to tree survival. Remote sensing has the potential to optimise management and better understanding of tree survival mechanisms in this zone. Stand density is thought to affect canopy and soil temperature through shading, change in overall albedo, evapotranspiration, and its effect on convective cooling through wind penetration into the canopy layer. Our objective was to assess the effect of stand density on the canopy and the mean plot temperature as a basis to optimize energy management of a severely water-limited forest.

We used a drone equipped with RGB, thermal (FLIR) and multispectral cameras (Parrot Sequoia, bands: 550nm, 660nm, 735nm & 790nm) alongside independent Lidar measurements in a set of five replicate plots of three different stand density treatments (100, 200 & 300 trees/ha) alongside ground-based measurements. Drone flights were performed during midday throughout the peak of the summer drought (lasting ~8 months) in our semi-arid Aleppo Pine forest research site in southern Israel (sun near NADIR & midday solar radiation >800 W·m^-2, air temperature >30°C). Finally, a set of techniques were developed to automatically identify and extract data of individual tree canopies from the aerial images.

Initial results highlight the importance of partitioning the forest into exposed and shaded soil and tree canopy: The canopy-to-air and exposed soil-to-air temperature differences reached up to 5°C and 35°C, respectively, while shaded soils were in the same temperature range as canopies. Ground-based measurements of DBH and photosynthetic activity increased with decreasing stand density. This is in spite of up to 30% more longwave radiation reaching the canopies through exposure to the hot soil and lack of shading from neighbouring trees in the lower density plots. Unexpectedly, there was a lack of significant canopy temperature differences among density plots, indicating that trees in all treatments dissipated the excess energy equally efficiently. Therefore, mean plot-scale forest surface (skin) temperatures (including both soil and canopy) were affected by the fraction of canopy cover rather than canopy temperature differences among different stand density plots. The results highlight the limitation of interpreting low-resolution satellite data in open canopy forests. Our results will allow us to assess the stand density effects on the balance between carbon sequestration (biogeochemical effects) and surface energy balance (biogeophysical effects).

How to cite: Rotenberg, E., Müller, J. D., Segev, L., and Yakir, D.: Drone-based remote sensing shows no effect of stand density on canopy temperature in semi-arid pine forest during drought, EGU General Assembly 2020, Online, 4–8 May 2020, EGU2020-13334, https://doi.org/10.5194/egusphere-egu2020-13334, 2020.

The ALtimetry for BIOMass project (ALBIOM) is ESA-funded Permanent Open Call Project that proposes to derive forest biomass using Copernicus Sentinel-3 (S3) SAR altimeter data. The project targets the need to improve our current global observations of biomass as an Essential Climate Variable (ECV) and crucial for bioenergy, risk mitigation activities, and sustainable management of forests. The overall goal is to estimate biomass with sufficient accuracy to be able to increase the existing satellite data for biomass retrieval and to improve the global mapping and monitoring of this fundamental variable.

The project originates from evidence that radar altimetry backscatter over land responds to a variety of land parameters, including vegetation-related parameters, at the different bands used by past and existing altimeters.

To achieve the scientific objectives, the project is structured into six conceptual tasks. After a review of the literature and of the existing user needs, a sensitivity analysis is performed to understand the relationship between SAR altimetry backscatter data and land parameters themselves. This is followed by the development of a Sentinel-3 SAR altimeter backscatter simulator over vegetated areas, and then by the development of a biomass inversion algorithm, testing different retrieval methodologies, both theoretical and empirical. A validation task for both the model and the algorithm is carried out over specific test sites of boreal and tropical forests, to finally generate a prototype of biomass product to be reviewed by potential users.

The sensitivity analysis allows to understand how the S3 Level 1 backscatter power waveforms change with respect to varying biomass, but also how they are affected by other land parameters such as soil moisture, land cover, topography and roughness. This analysis is carried out considering both the single-looked and the multi-looked waveforms, and considering primarily the high-resolution SAR mode, but also the Pseudo Low Resolution Mode (PLRM). The outcome of the sensitivity analysis provides indication of what waveforms, acquisition mode, observables derived from the waveforms and characteristics of the waveforms themselves respond more strongly to biomass variations, and on the degree of influence of the other auxiliary parameters, informing on the best strategies and approaches to adopt for the development of the retrieval algorithm.

Subsequent to the sensitivity analysis, the S3 altimetry backscatter simulator is developed over vegetated areas, reproducing both the coherent scattering component, which represents the echo from the ground, and incoherent scattering component arising from the forest layers between the ground and the top of canopy. The approach followed is similar to that of the SAVERS simulator, developed for GNSS-Reflectometry, with the signal backscatter attenuation introduced by branches, leaves and trunks modelled through the discrete approach of the Tor Vergata Scattering Model.

Results from the sensitivity analysis and the initial stages of the simulation development will be presented and discussed at the conference, together with the foreseen approaches for the development of the biomass retrieval algorithm.

How to cite: Clarizia, M. P., Pierdicca, N., Guerriero, L., Reynolds, J., Taravat, A., De Felice-Proia, G., Vittucci, C., Comite, D., Restano, M., and Benveniste, J.: Estimating biomass using SAR Altimetry data onboard the Copernicus Sentinel-3 Mission: the ALBIOM project, EGU General Assembly 2020, Online, 4–8 May 2020, EGU2020-20303, https://doi.org/10.5194/egusphere-egu2020-20303, 2020.

The utilization of UAVs for the acquisition of ultra-high resolution imagery has heavily increased during the past decade. Once the hardware is purchased, images can be recorded almost at any time and at low cost. The image parameters can be determined in terms of spectral channels, image overlap, and geometric resolution. The overlap between the images enables stereoscopic image processing, the delineation of point clouds, and the generation of seamless image mosaics. UAV image data products have gathered high interest in the forestry community, as structural and spectral features can be delineated. Accordingly, regular forest monitoring and inventory can be supported using UAV data.

In this study, the potential of DJI Phantom 4 Pro RTK imagery based orthomosaics and point clouds to map dead wood on the forest floor is investigated. The test site is located in the center of the Hainich national park. The Hainich national park is an unmanaged forest comprising deciduous tree species such as Fagus sylvatica (beech), Fraxinus excelsior (ash), Acer pseudoplatanus (sycamore maple), and Carpinus betulus (hornbeam). The flight campaigns were controlled from the Hainich flux tower in the central part of the park area. RGB image data was captured in March 2019 during leaf-off conditions. Agisoft Metashape was used for processing the imagery to orthomosaics and point clouds. The living/standing trees were virtually removed from the point clouds as follows: 1.) normalizing the point cloud for topography, 2.) dropping all points above 5 m height. The remaining points were converted to an orthorectified RGB raster file, which solely contains the forest floor including the deadwood (lying stems) and tree stumps of the virtually cut trees. This raster was eventually used for dead wood mapping. The mapping task was accomplished using the OBIA software eCognition using the line extraction function as major method. The detection rate of the automatic mapping was approximately 70%. The dead wood mapping was complicated dead wood of several years of age featuring almost the same color and elevation level as the surrounding forest floor. Due to the latter, no elevation information was used. For regular monitoring considering recent dead wood only elevation information can be implemented and higher detection rates are feasible.

How to cite: Thiel, C., Müller, M., Epple, L., Hese, S., Berger, C., and Voltersen, M.: UAV-based dead wood mapping in a natural deciduous forest in mid-Germany, EGU General Assembly 2020, Online, 4–8 May 2020, EGU2020-18430, https://doi.org/10.5194/egusphere-egu2020-18430, 2020.

Forest canopy reflectance varies with solar and observation geometries and shows distinct anisotropic characteristics. The bidirectional reflectance distribution function (BRDF) of forest canopies is influenced by canopy structure, leaf biochemistry and background reflectance. Multi-angular remote sensing observations of forest canopies provide much more information about canopy structure and background information compared with the nadir observations. The development of unmanned aerial vehicle (UAV) provides great opportunities for multi-angular observations in forests. We developed a solid method to obtained bidirectional reflectance of forest canopies based on a hyperspectral UAV imaging platform in this study. With this multi-angular observation method, we obtained canopy reflectance images with the view zenith angle (VZA) varying from 60° (forward) to 60° (backward) at fixed interval (10°), as well as the hotspot and darkspot images in the principle plane in conifer forests. Since the single pixel with very high spatial resolution (around 10 cm) in the UAV images are not representative for the study of the whole forest canopy, several pixels in the central of each images were selected and averaged to determine the canopy reflectance. Variations of the averaged reflectance with ground distance represented by the selected pixels were analyzed and the optimum ground distance for study the multi-angular forest canopy reflectance was determined. The observed canopy reflectance peaks at the hotspot and clear images of the hotspot are observed. The sensitivities of canopy reflectance to VZAs vary with spectral bands. The reflectance at red bands near 680 nm are most sensitive to VZA. Some common used vegetation indices, such as NDVI, EVI, MTCI, PRI, also vary greatly with VZAs and demonstrate different spatial distribution patterns. The observations fit well with the 4-Scale geometric-optical model simulations. The multi-angular observation methods based on UAV platform have the advantages of efficient and effective in multi-angular observation with higher flexibility in VZA adjustment and lower cost, compared with the airborne or spaceborne sensors. This multi-angular observation method is very useful for study the BRDF and canopy structural and biochemical characteristics of forests and has great potential in forestry and ecological studies.

How to cite: Qiu, F. and Zhang, Q.: Multi-angular observation of forest canopy reflectance based on a hyperspectral UAV imaging platform, EGU General Assembly 2020, Online, 4–8 May 2020, EGU2020-18629, https://doi.org/10.5194/egusphere-egu2020-18629, 2020.

Ground-based point clouds (from laser scanning or photogrammetry, and from static or mobile devices) give very detailed 3D information of forest plots. Also, if this information is complemented with data gathered from aerial vehicles, some parts of the forest structure that are not visible from the terrain can be represented (e.g. treetops). However, the heterogeneity of the point clouds, the complexity of some forest plots and the limitations of some data gathering/processing techniques lead to some occlusions and misrepresentations of the features in the plot. Therefore, complete automation of very detailed characterizations of all the items/features/structures in a forest plot is, most of the times, not possible yet.

On one hand, single trees (or small groups of them) can be modelled in detail from dense point clouds (e.g. using quantitative structure models), but this processes usually require  complete absence of leaves and  intense and/or active operator labouring. On the other hand, many methods automate the location of the trees in a plot and the estimation of basic parameters, like the diameters and, sometimes, the total tree height.

We are developing a fully automatic method that lies in between some very accurate but labour-intensive single-tree models, and the mere location and diameter calculation of the trees in a plot. Our method is able to automatically detect and locate the trees in a plot and calculate diameters, but it is also able to characterize the 3D tree structure: stem model, inclination and curvature; inclination and location of the main branches (in some cases); and tree crown individualization and diameter estimation. In addition, our method also classifies the points on understory vegetation.

Our method relies on the integration of algorithms that have been developed by our team, and includes the development of new modules. The first step consists in an initial classification of the point cloud using a multiscale approach based on local shapes. As a result, the point cloud is preliminarily classified into three classes: stems, branches and leaves, and ground. After that, a series of geometric operations lead to the final 3D characterization of the plot structure: (i) stem axes and section modelling (from the pre-classified points on the stems), (ii) distance points-closest stem axis and tree individualization, (iii) extraction and characterization of the main branches, and (iv) final classification of the points laying on stems, main branches, rest of the canopy, understory and ground.

We are testing the algorithm in several forest plots with coniferous and broadleaf trees. Initial results show values of completeness and correctness for tree detection and point classification over 90%.

Currently, there are already several cross-cutting projects using our method´s results as inputs: (i) Automatic calculation of taper functions (use: diameters along the stem and tree height), (ii) wood quality based on shape (use: diameters along the stem and insertion of main branches), and (iii) wildfire behaviour models (use: fuel classification and 3D structure to adapt the data to the format of the existing 3D fuel standard models).

How to cite: Cabo, C., Ordoñez, C., Prendes, C., Doerr, S., Roces-Diaz, J. V., and Santin, C.: Towards the automatic 3D characterization of forest plots , EGU General Assembly 2020, Online, 4–8 May 2020, EGU2020-20688, https://doi.org/10.5194/egusphere-egu2020-20688, 2020.

As climate change progresses, the form of forests has been changing. Developmental studies of remote sensing methods are needed to accurately estimate the changing form of forests. Recently, studies for estimating the forest vegetation height of forest area using Landsat data have been actively conducted. Therefore, this study calculated the SLAVI index composed of 4 (red), 5 (NIR), 7 (SWIR2) band combinations of Landsat 8. The relationship between the height of trees was estimated by linear regression analysis. Based on the result, a comparison in the height of the forest stands measured by the National Forest Inventory (NFI) shows a very high accuracy by the height of trees over 9 meters. The applicability of the study was investigated with the results, and the accuracy of the study will be compared through field surveys. The estimated accuracy of the height of trees in this study is not as high as 0.5-0.6 (R²), but it has an advantage of low cost and less effort to estimate the ​​height of trees in a large area and to acquire image data easily. Information about the height of trees is an important parameter for estimating forest biomass and carbon stocks, which is significant in studies of forest under climate change.

How to cite: Hong, M., Ham, B., Lee, S. J., Lee, H., and Lee, W.-K.: Estimation of forest vegetation height using Landsat data, EGU General Assembly 2020, Online, 4–8 May 2020, EGU2020-21257, https://doi.org/10.5194/egusphere-egu2020-21257, 2020.

Protection from landslides is one of the most important regulating services provided by forest ecosystems. Tree roots provide an increase in tensile strength, compression and shear resistance, compared to that uniquely due to the soil properties. This additional effect is known as root reinforcement. The degree of soil reinforcement given by roots have been modeled using laboratory and field data. The great spatial and temporal variability of root distribution is one of the main sources of uncertainty for the development of accurate and reliable models to quantify root reinforcement. The relative importance of stand structure remains poorly known. Here, we analyze the relationships between observed stand structure from a sample of spruce, beech, chestnut and mixed stands of the Southeastern Alps, and a spatially explicit model of root reinforcement. Data were collected in 20-m radius sampling units inclined 15-40° and covered by a low-resolution airborne LiDAR-derived canopy height model. Tree size and position were used to calculate root reinforcement through commonly used and calibrated models. Then, we studied the relationships between root reinforcement, stand structural indexes and area-based stand metrics from canopy height model. In specific conditions, the three groups of variables were correlated. Therefore, root reinforcement values might be spatially extrapolated through available canopy height models. Final step is to integrate the extrapolated values into a landslide susceptibility model, which combines other data available from forest plans, digital elevation models, geological and meteorological data. This study provides managers with a tool to periodically update maps of the service given by forest trees to protection of humans from landslides.

How to cite: Alterio, E., Rizzi, A., Fogliata, P., Marchi, N., Cislaghi, A., Lingua, E., Sitzia, T., and Bischetti, G. B.: Extrapolating a spatially explicit tree root reinforcement model with field and LiDAR-derived stand data, EGU General Assembly 2020, Online, 4–8 May 2020, EGU2020-21466, https://doi.org/10.5194/egusphere-egu2020-21466, 2020.

Forests play an important role in global carbon, hydrological and atmospheric cycles. Current environmental issues have a strong impact on forest health. Satellite remote sensing is widely used for forest state monitoring due to increasing availability of satellite data and high temporal resolution. However, a spatial resolution of satellite data is often insufficient to detect small areas of forest drying. For a clearer detection of affected forest areas, spectral unmixing is required.

The results of spectral unmixing of Belarusian spacecraft data (4 bands: blue, green, red, NIR; spatial resolution 10 meters) are performed. To detect affected forest areas that need to be specified, the vegetation index NDVI is calculated. Then, spectral mixture analysis is running for these areas. The library of endmembers (pure spectral signatures) was created by ground measurements using spectral instruments that were developed in the department of aerospace researches of Belarusian state university. Comparison of spectral unmixing results and airborne measurements shows high agreement. Airborne measurements of study forest area was carried out using Leica airborne digital sensor. Spatial resolution of airborne data is around 40 centimeters. The developed spectral unmixing approach can be used for other tasks, such as burned area mapping, crop monitoring, etc.

How to cite: Siliuk, V., Katkovsky, L., and Beliaev, B.: Monitoring of forest health using spectral unmixing of multispectral satellite data, EGU General Assembly 2020, Online, 4–8 May 2020, EGU2020-679, https://doi.org/10.5194/egusphere-egu2020-679, 2020.

As the societal need to mitigate anthropogenic CO₂ emissions aggravates, scientists are challenged to improve climate projections, which in turn calls for better estimates of terrestrial carbon (C) stocks and fluxes. In order to meet this growing demand, we are developing a novel methodology for the production of precise annually-resolved C estimates in forest ecosystems, by integrating Terrestrial Laser Scanning (TLS), flux-tower data, forest inventories, and tree-ring measurements. By coupling C estimates in the sampling year with radial growth and wood density data from tree cores, we are able to precisely reconstruct forest biomass in mature forest stands across Europe and create new insight into historical C dynamics. Here, we present our first results of biomass estimates in a Fagus sylvatica dominated tree stand in Hainich National Park, Thuringia, Germany. We provide an overview of the methodology that was developed for the extraction of biomass information from TLS point clouds. Furthermore, we discuss the challenges introduced at different processing steps and highlight the opportunities that the TLS provides for C cycle research. Ultimately, we aim at reducing uncertainties in the scaling of annual C stock changes and at advancing our understanding of C cycling in temperate forests. We expect that this information will create a refined empirical baseline for vegetation (and by extent climate) model parameterization across multiple spatiotemporal domains and thus improve our understanding of carbon sink trajectories and carbon allocation dynamics and drivers in temperate forests.

How to cite: Karamihalaki, M., Wei, J., Marty, M., and Babst, F.: Improving assessments of forest carbon cycling by integrating terrestrial LiDAR with dendrochronological and flux tower data, EGU General Assembly 2020, Online, 4–8 May 2020, EGU2020-1176, https://doi.org/10.5194/egusphere-egu2020-1176, 2020.

Land Surface Phenology (LSP) is the study of the phenology through satellite sensors. It integrates phenological patterns (mainly spatial) and processes (mainly temporal) within heterogeneous biophysical environments across multiple scales. It is a very useful tool for the characterization and monitoring of forests. Tropical montane cloud forest is the most diverse type of vegetation per unit area, since it occupies less than 1% of Mexico but harbours 10% of the country’s plant biodiversity. It is a critical priority for biodiversity conservation, its permanence in the medium and long term is threatened by habitat destruction and climate change. A regional conservation approach, which values all fragments of this type of forest as contributing to regional biodiversity, will be required to conserve plant biodiversity in central Veracruz. This area is one of the Rare forest ecoregions within biodiversity hotspots. Our primary aim was to identify priority zones for stablishing a Tropical montane cloud forest monitoring network in Central Veracruz based on its phenological responses at multiples scales. Our methodology can be applied in other tropical biodiversity zones, even in the absence of adequate cartography. We start from homogeneous and reliable pixels and automatically calculate the number of pheno-regions that exist within this type of vegetation in the study area, based on different LSP pheno-metrics extracted from different MODIS vegetation index time-series (NDVI & EVI) with Timesat and BFAST algorithm. We extract Fraction cover subpixels homogeneus from MODIS and Sentinel 2 LC map with Random Forest classification and success rate analysis curve ensures the reliability of the LC map. We identify 4 statistically different representative pheno-regions through cluster analysis in this type of forest within the study area and we obtained 351 priority areas where a phenological monitoring network could be located.

How to cite: Aragones, D., Rodriguez-Galiano, V. F., Caparros-Santiago, J. A., and Espinoza-Guzman, M. A.: Delineating monitoring Network in Biodiversity Hotspot based on Land Surface Phenology. The case of Tropical Montane Cloud Forest, EGU General Assembly 2020, Online, 4–8 May 2020, EGU2020-4194, https://doi.org/10.5194/egusphere-egu2020-4194, 2020.

The high forests in southwest Ethiopia, some of the last remaining Afromontane forests in the country, are home to significant forest coffee production. While considered as beneficial in maintaining forests, there have been growing concerns about the degradation caused by intensive coffee production in the forests. However, yet no suitable methods have been developed to map the intensively managed coffee forests. In this study, we explore the feasibility of monitoring the extent of the degradation within the intensively managed coffee forests by using satellite imagery (Landsat-8 and Sentinel-2). For this, we conducted in-situ field canopy photo and tree surveys, and the results were analysed with satellite-derived vegetation indices such as NDVI and NBR. This feasibility study informed us that the detection of the intensively managed forest coffee areas (disturbances caused by this practice) using satellite imagery can be possible, as the dry-season forest structure (canopy, undergrowth) and vegetation indices in the intensively managed coffee forests are significantly distinctive from those in natural forests. This study will contribute to the long-term sustainable management of the coffee forest.

How to cite: Hwang, B., Hundera, K., Mekuria, B., Wood, A., and Asfaw, A.: Identifying Intensively Managed Coffee Forests in Southwest Ethiopia using Satellite Imagery, EGU General Assembly 2020, Online, 4–8 May 2020, EGU2020-5808, https://doi.org/10.5194/egusphere-egu2020-5808, 2020.

Forest aboveground biomass (AGB) plays an important role in measuring forest carbon reserves. Accurate mapping AGB is important for monitoring carbon stocks and will contribute to achieve the goal of sustainable development. In this study, we explored the potential of mapping AGB in north China using a three-year monthly time series of Senitinel-1 (S1) and Sentinel-2 (S2) data. The backscattering and indices of SAR S1 combined with spectral reflectance, vegetation indices and biophysical parameters from multispectral S2 imagery were evaluated for AGB prediction in a Random Forest regression. Three scenarios were conducted with different datasets to determine: (1) the potential of using S1 and S2 to estimate AGB, (2) optimal variables selection for AGB mapping, (3) contribution of time series datasets to improving the accuracy of AGB mapping. Random forest regression was used to develop forest AGB estimation models, which was divided into three types of modeling using only S1, only S2, and a combination of S1 and S2. Compared to S1 (RMSE = 65.7 Mg/ha), S2 achieved better prediction accuracy (RMSE = 58.4 Mg/ha), although the combination of S1 and S2 time series datasets estimated the best AGB results (RMSE = 42.3 Mg/ha). The research implied that incorporation of SAR and multispetral data considerably improved AGB mapping performance when compared with the use of SAR or multispectral data alone. This proposed approach provides a new insight in improving the estimation accuracy of forest AGB in north China.

How to cite: Wang, Y. and Zhang, X.: Estimation of aboveground biomass in North China using Sentinel-1 and 2 datasets, EGU General Assembly 2020, Online, 4–8 May 2020, EGU2020-6548, https://doi.org/10.5194/egusphere-egu2020-6548, 2020.

Natura 2000 network, the world's largest network of protected areas, is considered a success for habitat and biodiversity protection, in the last decades. Our objective is to develop an algorithm for satellite data temporal analysis of protected areas, and to apply subsequently this algorithm for analysis of all Natura 2000 sites in Europe. We have developed an algorithm for satellite data temporal analysis of protected areas using JavaScript in Google Earth Engine, which is a web interface for the massive analysis of geospatial data, providing access to huge amount of data and facilitating development of complex workflows. This work focused on analysis of Global Forest Change dataset representing forest change, at 30 meters resolution, globally, between 2000 and 2018. Our results show that at least regarding forest protection, the network is not very successful, the 25350 sites losing 35246.8 km² of forest cover between 2000 and 2018, gaining only 9862.1 km². All 28 countries recorded a negative forest net change, with a mean value of -906.6 km², the largest forest area change recording Spain (-5106.4 km² in 1631 sites), Poland (-4529 km² in 962 sites), Portugal (-2781.9 km² in 120 sites), Romania (-1601.4 km² in 569 sites), Germany (-1365.7 km² in 5049 sites) and France (-1270.9 km² in 1520 sites). Among countries with the lowest values in net forest change is Ireland (-17.4 km² in 447 sites), Estonia (-104.1 km² in 518 sites), Netherlands (-132.3 km² in 152 sites), Finland (-268.6 km² in 1722 sites) and Sweden (-341.6 km² in 3786 sites).

How to cite: Chețan, M.-A. and Dornik, A.: 20 years of forest change in Natura 2000 protected areas network , EGU General Assembly 2020, Online, 4–8 May 2020, EGU2020-11227, https://doi.org/10.5194/egusphere-egu2020-11227, 2020.

The north-East of Europe is affected by the ash (Fraxinus excelsior) dieback caused by the fungal pathogen Hymenoscyphus pseudoalbidus. A great variety of studies utilize remote sensing data and subsequently derived spectral indices to estimate the magnitude and spatial distribution of the damage for different tree types. 

Often, structural indices, such as the NDVI are applied to detect already affected tree (sometimes even for early detection). However, there are differences in the suitability of an index. While a structural index, might have advantages when the canopy is not closed, pigment-based indices can show more variation within a full crown coverage forest. Therefore, the season of data acquisition might define the preferred index-selection. The same accounts not just for seasonal but for inter-annual changes, too. Here, the pigment indices show a higher sensitivity towards changes due to damages than structural indices.

To show these differences, the presented study is evaluating a variety of indices derived by hyperspectral imagery for affected ash trees in north-east Germany. This includes images from different phenological stages within one year (2015) and a comparison between 2011, 2015, and 2019 because the decline increased severely within this timespan for the observed trees. The indices were compared with tree damage estimations from the regional forest administration. 

Preliminary results show a better relation for structural indices in autumn, but higher relation for pigment-based indices in spring and summer, once the crown is closed. A higher sensitivity to changes between 2011 and 2019 can be shown for pigment-based indices.

How to cite: Förster, M., Clasen, A., Jütte, K., Döpper, V., and Kleinschmit, B.: Comparison of remote sensing-based indices for ash vitality detection in North-East Germany, EGU General Assembly 2020, Online, 4–8 May 2020, EGU2020-13046, https://doi.org/10.5194/egusphere-egu2020-13046, 2020.

Central European forests face challenges with climate changing much faster than they can adapt. Extremely hot and dry summers like in 2018 deprive forests of soil moisture, leaving them with low ground water levels. While individuals with deep and well-established root systems survive, young individuals and shallow-rooted species perish.

In southern Germany, die-off of single trees or small groups got noticeable recently. Such effects of harsher conditions rarely occur over large areas, but more in a spotted, irregular manner. This makes the phenomenon difficult to detect and to estimate its extent. The share of trees lately deteriorated may be larger than expected and represent a considerable portion of forests. Therefore, we see the great need for monitoring. Remote sensing data is suitable to examine inaccessible areas at a large scale. To quantify mortality of individual trees among a majority of vital ones, sensor platforms and respective data have to fulfill certain criteria regarding spatial, temporal and spectral resolution. Dead trees can be distinguished from others due to discoloration and defoliation. This change in appearance affects the spectral response, even in pixels larger than the tree’s extent.

This study aims at recommending a suitable spatial scale for space-borne multispectral imagery products to achieve this task. We evaluate commercial and free remote sensing data products and their ability to estimate fractional cover of dead vegetation. Satellite data employed in this study comes from Landsat 8 (30 m), Sentinel-2 (10 m), RapidEye (6.5 m) and PlanetScope (3 m). Classification performance is tested against high-resolution multispectral aerial imagery (17 cm) acquired with a Micasense RedEdge-M camera.

High-resolution Micasense images are capable of detecting single dead trees, even after downgrading the resolution from 17 cm to 3 m. For all data products tested, fraction of dead trees per pixel did not differ significantly among land cover types (dead vegetation, vital vegetation, pavement, open soil). This indicates that individual dead trees may not be detectable in vital forest stands. The finding even seems to be valid for a resolution of 3 m (PlanetScope), which is identical to the downgraded Micasense data. In the near future the detection of this phenomenon might profit from technical developments towards even higher spatial detail of space-borne sensors. Alternatively, high resolution images from aerial campaigns, manned or unmanned, could bridge this gap when flight time and spatial coverage are increased significantly and facilitating policies are in place.

How to cite: Heisig, J. and Samimi, C.: Detecting drought effects on tree mortality in forests of Franconia (Germany), EGU General Assembly 2020, Online, 4–8 May 2020, EGU2020-15579, https://doi.org/10.5194/egusphere-egu2020-15579, 2020.

Importance of biodiversity is increasingly highlighted as an essential part of sustainable forest management. As direct monitoring of biodiversity is not possible, proxy variables have been used to indicate site’s species richness and quality. In boreal forests, European aspen (Populus tremula L.) is one of the most significant proxies for biodiversity. Aspen is a keystone species, hosting a range of endangered species, hence having a high importance in maintaining forest biodiversity. Still, reliable and fine-scale spatial data on aspen occurrence remains scarce and incomprehensive. Although remote sensing-based species classification has been used for decades for the needs of forestry, commercially less significant species (e.g., aspen) have typically been excluded from the studies. This creates a need for developing general methods for tree species classification covering also ecologically significant species.

Our study area, located in Evo, Southern Finland, covers approximately 83km², and contains both managed and protected southern boreal forests. The main tree species in the area are Scots pine (Pinus sylvestris L.), Norway spruce (Picea abies (L.) Karst), and birch (Betula pendula and pubescens L.), with relatively sparse and scattered occurrence of aspen. Along with a thorough field data, airborne hyperspectral and LiDAR data have been acquired from the study area. We also collected ultra high resolution unmanned aerial vehicle (UAV) data with RGB and multispectral sensors.

Our aim is to gather fundamental data on hyperspectral and multispectral species classification, that can be utilized to produce detailed aspen data at large scale. For this, we first analyze species detection at tree-level. We test and compare different machine learning methods (Support Vector Machines, Random Forest, Gradient Boosting Machine) and deep learning methods (3D convolutional neural networks), with specific emphasis on accurate and feasible aspen detection. The results will show, how accurately aspen can be detected from the forest canopy, and which bandwidths have the largest importance for aspen. This information can be utilized for aspen detection from satellite images at large scale.

How to cite: Kumpula, T., Viinikka, A., Mäyrä, J., Kuzmin, A., Hurskainen, P., Tanhuanpää, T., Kivinen, S., Kullberg, P., Poikolainen, L., Korpelainen, P., Stranden, M., Ritakallio, A., and Vihervaara, P.: Aspen detection in boreal forests: Capturing a key component of biodiversity using airborne hyperspectral, lidar, and UAV data, EGU General Assembly 2020, Online, 4–8 May 2020, EGU2020-21268, https://doi.org/10.5194/egusphere-egu2020-21268, 2020.

Identifying factors that influence the land surface temperature (LST) of urban forests can help improve simulations and predictions of spatial patterns of urban cool islands. This requires a quantitative analytical method that combines spatial statistical analysis with multi-source observational data. The purpose of this study was to reveal how human activities and ecological factors jointly influence LST in clustering regions (hot or cool spots) of urban forests. Using Xiamen City, China from 1996 to 2006 as a case study, we explored the interactions between human activities and ecological factors, as well as their influences on urban forest LST. Population density was selected as a proxy for human activity. We integrated multi-source data (forest inventory, digital elevation models (DEM), population, and remote sensing imagery) to develop a database on a unified urban scale. The driving mechanism of urban forest LST was revealed through a combination of multi-source spatial data and spatial statistical analysis of clustering regions. The results showed that the main factors contributing to urban forest LST were dominant tree species and elevation. The interactions between human activity and specific ecological factors linearly or nonlinearly increased LST in urban forests. Strong interactions between elevation and dominant species were generally observed and were prevalent in either hot or cold spots areas in different years. In conclusion, quantitative studies based on spatial statistics and GeogDetector models should be conducted in urban areas to reveal interactions between human activities, ecological factors, and LST.

How to cite: Ren, Y., Deng, L., Zuo, S., Song, X., Liao, Y., Xu, C., Chen, Q., Hua, L., and Li, Z.: Quantifying the Influences of Various Ecological Factors on Land Surface Temperature of Urban Forests, EGU General Assembly 2020, Online, 4–8 May 2020, EGU2020-2173, https://doi.org/10.5194/egusphere-egu2020-2173, 2020.

The aim of this study was to evaluate the utility of deep learning (DL) approaches to estimate forest growing stock volume from L-band SAR data over areas characterized by diverse species composition. For comparison, parametric models were also used. When using one independent variable (i.e. HV backscatter coefficient) the lowest estimation errors were observed for the empirical model followed by Random Forests (RF). Increasing the number of independent variables resulted in marginally more accurate results for the machine learning approaches. However, for the studied area, DL approaches did not improve GSV retrieval when compared to RF or empirical modelling suggesting that L-band data sensitivity to GSV values is the main limiting factor.

How to cite: Tanase, M. A., Belenguer-Plomer, M. A., Marin, G., and Badea, O.: Deep Neural Networks FOR forest Growing stock volume retrieval: a compartive analysis for L-band SAR data, EGU General Assembly 2020, Online, 4–8 May 2020, EGU2020-3854, https://doi.org/10.5194/egusphere-egu2020-3854, 2020.

Boreal forests in Siberia store huge amounts of aboveground carbon. Global warming potentially threatens this carbon storage due to more frequent droughts or other disturbances such as fires. These disturbances can change recruitment patterns, and thus may have long-lasting impacts on population dynamics. Assessing high-resolution forest stand structures and forecasting their response for the upcoming decades with detailed models is needed to understand the involved key processes and consequences of global change.

We present forest stand inventories derived from UAV imagery and a developed processing chain including Individual Tree Detection (ITD) and species determination for 56 sites on a bioclimatic gradient at the Tundra-Taiga-Ecotone in Northeastern Siberia. We will use these and further 58 traditional count and measurement data as starting points for the detailed individual-based spatially explicit forest model LAVESI to predict future forest dynamics covering multiple sites across the Siberian treeline.

In our analyses, we will focus on assessing future structural changes of the forests and their aboveground biomass dynamics. For our discussion, we will evaluate the reliability of UAV-derived forest inventories by measuring the impact strength of error sources introduced in the methodology on the forecasts.

How to cite: Kruse, S., Shevtsova, I., Brieger, F., Wieczorek, M., Pestryakova, L. A., and Herzschuh, U.: Forecasting forest dynamics with the individual-based model LAVESI across the Siberian treeline: from UAV surveys to simulations, EGU General Assembly 2020, Online, 4–8 May 2020, EGU2020-13453, https://doi.org/10.5194/egusphere-egu2020-13453, 2020.

As part of a project aiming to support FSC certified logging concessions in their tasks of forest inventory and management, we collected aerial imagery over 9000 ha of tropical forests in Northern Congo using long range Unmanned Aerial Vehicles (UAVs). Once processed into orthomosaics, the aerial imagery is used in combination with reference training samples to train a deep learning object detection model (FasterRCNN) capable of detecting and predicting tree species. The remoteness and diversity of these forests make both data acquisition and generation of a training dataset challenging. Unlike natural images containing common objects like cars, bicycles, cats and dogs, there is no easy way to create a training dataset of tree species from overhead imagery of tropical forests. The first reason is that a human operator cannot as easily recognize and label objects. The second reason is that the polymorphism of tree species, phenological variations and uncertainty associated with visual recognition makes the exhaustive labeling of all instances of each class very difficult. Such exhaustive labeling is required to successfully train any object detection model. To overcome these challenges we built an interactive and ergonomic interface that allows a human operator to work in a spatial context, being guided by the approximate geographic location of already inventoried trees. We solved the issue of non-exhaustive instance labeling by building synthetic images, hence allowing full control of the training data. In addition to these specific developments related to training data generation, we will present details of the UAV missions, modelling results on synthetic images, and finally preliminary results of model transfer to aerial imagery.

How to cite: Dutrieux, L., Azzabi, R., Bauwens, S., Bouka Dipelet, U. G., Chenoz, O., Couturier, A., Dérian, P., Doumenge, C., Dubois, H., Gond, V., Laverdunt, A., Olé, J., Prosperi, J., Rivière, L., and van Loon, T.: Tree species detection and identification from UAV imagery to support tropical forest monitoring, EGU General Assembly 2020, Online, 4–8 May 2020, EGU2020-17759, https://doi.org/10.5194/egusphere-egu2020-17759, 2020.