BG11

Over the last 25 years, the world’s forests have undergone substantial changes. Deforestation and forest degradation in particular contribute greatly to biodiversity loss through habitat destruction, soil erosion, terrestrial water cycle disturbances and anthropogenic CO2 emissions. In certain regions and countries, the changes have been more rapid, which is the case in the Greater Mekong sub-region recognized as deforestation hotspot (FAO, 2020). In this region, illegal and unsustainable logging and conversion of forests for agriculture, construction of dams and infrastructure are the direct causes of deforestation. Effective tools are therefore urgently needed to survey illegal logging operations which cause widespread concern in the region.

Monitoring systems based on optical data, such as the UMD/GLAD Deforestation alerts implemented on the Global Forest Watch platform, are limited by the important cloud cover which causes delays in the detections. However, it has been demonstrated in the last few years that forest losses can be timely monitored using dense time series of (synthetic aperture) radar data acquired by Sentinel-1 satellites, developed in the frame of the European Union’s Earth observation Copernicus programme. Ballère et al. (2021) showed for example that 80% of the forest losses due to gold mining in French Guiana are detected first by Sentinel-1-based forest loss detection methods compared with optical-based methods, sometimes by several months. Methods based on Sentinel-1 have been successfully applied at the local scale (Bouvet et al., 2018, Reiche et al., 2018) and can be adapted and tested at the national scale (Ballère et al., 2020).

We show here the main results of the SOFT project funded by ESA in the frame of the EO Science for Society open calls. The overall SOFT project goal is to provide validated forest loss maps every month over Vietnam, Cambodia and Laos with a minimum mapping unit of 0.04 ha, using Sentinel-1 data. The results confirm the analysis of the deforestation fronts published recently by the WWF (Pacheco et al., 2021), showing that Eastern Cambodia, and Southern and Northern Laos are currently forest disturbances hotspots.

References:

Ballère et al., (2021). SAR data for tropical forest disturbance alerts in French Guiana: Benefit over optical imagery. Remote Sensing of Environment, 252, 112159.

Bouvet et al., (2018). Use of the SAR shadowing effect for deforestation detection with Sentinel-1 time series. Remote Sensing, 10(8), 1250.

FAO. Global Forest Resources Assessment; Technical Report; Food and Agriculture Association of the United-States: Rome, Italy, 2020.

Pacheco et al., 2021. Deforestation fronts: Drivers and responses in a changing world. WWF, Gland, Switzerland

How to cite: Mermoz, S., Bouvet, A., Ballère, M., Koleck, T., and Le Toan, T.: Forest disturbances detection in Vietnam, Cambodia and Laos using Sentinel-1 data, EGU General Assembly 2021, online, 19–30 Apr 2021, EGU21-16177, https://doi.org/10.5194/egusphere-egu21-16177, 2021.

The assessment of forest fire risk has recently gained interest in countries of Central Europe and the alpine region since the occurrence of forest fires is expected to increase with a changing climate. Information on forest fuel structure, which is related to forest structure, is a key component in such assessments. Forest structure information can be derived from airborne laser scanning (ALS) data, whose value for the derivation of respective metrics at a high accuracy level has been demonstrated in numerous studies over the last years.

Yet, the temporal resolution of ALS data is low as flight missions are typically carried out in time intervals of five to ten years in Central Europe. ALS-derived forest structure descriptors for fire risk assessments, therefore, are often outdated. Open access earth observation data offer the potential to fill these information gaps. Data provided by synthetic aperture radar (SAR) sensors, in particular, are of interest in this context since this technology has a known sensitivity to the vegetation structure and acquires data independent of weather or daylight conditions.

In our study, we investigate the potential to derive forest structure descriptors from time series of Sentinel-1 (S-1) SAR data for a deciduous forest site in the Eastern part of Austria. We focus on forest stand height and fractional cover, which is a measure for forest density, as both of these components impact forest fire propagation and ignition. The two structure metrics are estimated using a random forest (RF) model, which takes a total of 36 predictors as input, which we compute from the S-1 time series. The model is trained using ALS-derived structure metrics acquired during the same year as the S-1 data.

We estimated stand height with a root mean square error (RMSE) of 4.76 m and a bias of 0.09 m at 100 m resolution, while the RMSE for the fractional cover estimation is 0.08 with a bias of zero at the same resolution. The spatial comparison of the structure predictions with the ALS reference further shows that the general structure is well reproduced. Yet, fine scale variations cannot be completely reproduced by the S1-derived structure products, and the height of tall stands and very dense canopy parts are underestimated. Due to the high correlation of the predicted values to the reference (Pearson’s R of 0.88 and 0.94 for the stand height and the fractional cover, respectively), we consider S-1 time series in combination with ALS data with low temporal resolution and machine learning techniques to be a reliable data source and workflow for regularly (e.g. < yearly) updating ALS structure information in an operational way.

How to cite: Bruggisser, M., Dorigo, W., Dostálová, A., Hollaus, M., Navacchi, C., Schlaffer, S., and Pfeifer, N.: Derivating forest structure from Sentinel-1 time series to assist forest fire risk assessments, EGU General Assembly 2021, online, 19–30 Apr 2021, EGU21-9033, https://doi.org/10.5194/egusphere-egu21-9033, 2021.

Deforestation and degradation are two major threats to the global forest that jeopardize their functions to store carbon and mitigate climate change. Forest degradation undermines the health and functions of the forest to perform ecosystem services and is a stepping stone to deforestation. However, forest degradation has not been sufficiently monitored and quantified due to the varying intensity of disturbance and usually inconsistent spectral signals reflected in optical remote sensing. Drivers of forest degradation can be natural and/or human-related, and charcoal production is a key driver of forest degradation in sub-Saharan Africa due to the high demands for charcoal for energy consumption and the increasing rate of population growth and urbanization. In this study, we focus on charcoal production-driven forest degradation that occurred at the Mabalane district in Southern Mozambique from 2008 to 2018. We intend to demonstrate the potential of combining Global Ecosystem Dynamics Investigation (GEDI) data and Landsat time stacks for inspecting the changes in forest structure and aboveground biomass (AGB). To do so, we categorize the degraded forest by the year of disturbance based on a disturbance map produced for the study area for 2008-2018 by Sedano et al. (2019) and analyze the first year of publicly-released GEDI data to characterize forest structure and AGB at different disturbance classes. We also compare the GEDI L4A biomass with three other global and continental AGB products to understand the pre-disturbance biomass storage and the degradation patterns. Lastly, we build an empirical model between GEDI biomass and Landsat spectral bands and vegetation indices to quantify the biomass removal and regrowth from 10-year charcoal production. Uncertainties from the GEDI-Landsat models are estimated using Monte Carlo Simulations to propagate errors. The study improves the current understanding of forest degradation and carbon dynamics associated with it in tropical dry forests of sub-Saharan Africa. It also demonstrates the potential of combining spaceborne lidar missions and Landsat archives to facilitate accurate mapping of forest structural and AGB change in the degraded forest at a local scale. 

How to cite: Liang, M., Duncanson, L., and Sedano, F.: Quantifying Local-scale Forest Degradation Intensity from Charcoal Production Using a Fusion of GEDI and Landsat Data, EGU General Assembly 2021, online, 19–30 Apr 2021, EGU21-3147, https://doi.org/10.5194/egusphere-egu21-3147, 2021.

Multispectral imaging using unmanned aerial systems (UAS) enables rapid and accurate detection of pest insect infestations, which are an increasing threat to midlatitude natural forests. Pest detection at the level of an individual tree is of particular importance in mixed forests, where it enables a sensible forest management approach. Moreover, urban forests may be affected more seriously because an urban environment produces additional stressors. The stressors include changes in forest soil properties, tree species diversity, higher temperatures, and carbon dioxide content. The stressed trees are then optimal material for a bark beetle feeding. Therefore, it is necessary to use an appropriate method for the detection of individual infested trees.

In this contribution, we present a novel method for individual tree crown delineation (ITCD) followed by feature extraction to detect a bark beetle disturbance in a mixed urban forest using a photogrammetric point cloud (PPC) and a multispectral orthomosaic. An excess green index (ExG) threshold mask was applied before the ITCD to separate targeted coniferous trees from deciduous trees and backgrounds. The individual crowns of conifer trees were automatically delineated as (i) a full tree crown using marker-controlled watershed segmentation (MCWS), Dalponte2016, and Li 2012 region growing algorithms or (ii) a buffer around a treetop from the masked PPC.

We statistically compared selected spectral and elevation features extracted from automatically delineated crowns of each method to reference tree crowns to distinguish between the forest disturbance classes and two tree species. Moreover, the effect of PPC density on the ITCD accuracy and feature extraction was investigated. The ExG threshold mask application resulted in the excellent separability of targeted conifer trees and the increasing shape similarity of automatically delineated crowns compared to reference tree crowns. The results revealed a strong effect of PPC density on treetop detection and ITCD. If the PPC density is sufficient (> 10 points/m²), the automatically delineated crowns produced by Dalponte2016, MCWS, and Li 2012 methods are comparable, and the extracted feature statistics insignificantly differ from reference tree crowns. The buffer method is less suitable for detecting a bark beetle disturbance in the mixed forest because of the simplicity of crown delineation. It caused significant differences in extracted feature statistics compared to reference tree crowns. Therefore, the point density was found to be more significant than the algorithm used.

We conclude that the automatic methods may constitute a reliable substitute for the time-consuming manual tree crown delineation in tree-based bark beetle disturbance detection and sanitation of individual infested trees using the suggested methodology and high-density (>20 points/m², 10 points/m² minimum) PPC.

How to cite: Minařík, R., Langhammer, J., and Lendzioch, T.: Automatic Tree Crown Feature Extraction from UAS Multispectral Imagery for the Detection of Bark Beetle Disturbance in an Urban Forest, EGU General Assembly 2021, online, 19–30 Apr 2021, EGU21-8297, https://doi.org/10.5194/egusphere-egu21-8297, 2021.

In the past few decades, the occurrence of shrub forest dominated by the two species Green alder (Alnus viridis) and Dwarf mountain pine (Pinus mugo) has increased in the Swiss Alps. Up-to-date and area-wide information on its distribution is required for countrywide forest reporting (5 % of Swiss forest consists of shrub forest) and of great interest to the forestry sector. Such information helps to better understand forest succession and supports the evaluation and management of protection forests. Until now, this information has been based on estimates from the Swiss National Forest Inventory (NFI). Due to their sampling scheme that uses a regular grid, these data are not area-wide maps. However, new developments in remote sensing techniques in combination with high spatial and temporal resolution data have facilitated the production of maps over large areas, e.g. the whole of Switzerland (41’285 km²).

To map the shrub forest areas, we developed an approach that uses a Random Forest (RF) model, active learning techniques and data from multiple remote sensing sources. The training data was produced via aerial image interpretation of areas covered by shrub forest. We used predictor data from different sensors and technologies, complementing each other by their diverse sensitivity to properties of shrub forests. These data included airborne Digital Terrain (DTM) and Vegetation Height Models (VHM), and spaceborne Synthetic Aperture Radar (SAR) backscatter from the Sentinel-1 constellation and multispectral imagery from Sentinel-2. To improve mapping quality, an iterative and semi-automatic active learning technique was used to generate further training data.

The above outlined workflow enabled the production of a shrub forest map for the whole of Switzerland with a spatial resolution of 10 m. An accuracy assessment was performed using independent validation data of a total of 7’640 regularly distributed NFI plots. Mean shrub forest cover per plot (50 m x 50 m) was slightly underestimated by 1.5 % with a root mean square error of 10 %. The influence of the active learning was observed and revealed higher accuracies after each additional iteration of training data production. The proposed approach underscores the potential of multi-sensor data combined with active learning techniques to provide cost-effective and area-wide information on the occurrence of shrub forest in a manner complementary to the NFI measurements.

How to cite: Rüetschi, M., Weber, D., Koch, T. L., Small, D., and Waser, L. T.: Wide-area shrub forest map based on multi-sensor data and active learning, EGU General Assembly 2021, online, 19–30 Apr 2021, EGU21-11209, https://doi.org/10.5194/egusphere-egu21-11209, 2021.

Although forests cover about one third of global land surface, forests act as important biophysical, biogeochemical, hydrological, economic and cultural roles in the Earth systems. Forests contribute up to 75% of terrestrial gross primary production and store more carbon in forest biomass and soil compared to the atmosphere. Forest aboveground biomass (AGB) plays a crucial role in regional and global ecological balance. However, due to the difficulties in measuring forest biomass in the field at regional scales, a quantitative estimation with high accuracy of forest AGB by linking remote sensing is still a challenge, particularly in mountainous region. Thus, we combined the Landsat 8 OLI and Sentinel-2B data to estimate subalpine forest AGB using linear regression (LR), and two machine learning approaches - random forest (RF) and extreme gradient boosting (XGBoost), with the linkage of field observations in Jiuzhaigou National Nature Reserve, Eastern Tibet Plateau. A 10-fold cross validation (CV) method was used to evaluate the model accuracy, and then the proximity between the predicted value and the actual value was compared. The model efficiency (pseudo R²) and root mean square error (RMSE) were used as the accuracy evaluation criteria. Based on 54 field observations, results showed that mean forest AGB was 180.6 Mg ha^-1with a strong spatial variability from 61.7 to 475.1 Mg ha^-1. AGB varied significantly among forest types that AGB in coniferous forests was significantly higher than coniferous mixed forests and broad-leaved forests. Landsat 8 OLI and Sentinel-2B imagery were successfully applied to estimate AGB separately or combined. Integrating the Landsat 8 OLI and Sentinel-2B imagery significantly improved model efficiency for different modelling approaches. For the regression algorithms, machine learning method outperformed the linear regression. Among LR, RF and XGBoost approaches, XGBoost performed best with a model efficiency (R²) of 0.71 and root mean square error values of 46 Mg ha^-1 and subsequently used for spatial modelling. Modelled results indicated a strong spatial variability in AGB, with a total 6.6×10⁶ Mg across the study area. AGB distribution in the study area had obvious spatial characteristics, which was closely related to the elevation. It was mainly concentrated in the north and central areas, while in the southern region the AGB was relatively low, which was contrary to the trend of the elevation variation in the study area where the terrain was high in the south and low in the north. Our study highlighted a potential way to improve the estimate accuracy of forest AGB in mountainous region by integrating the Landsat 8 OLI and Sentinel-2B data using machine learning algorithms.

How to cite: Luo, K., Tang, X., Liu, L., Luo, X., and Li, J.: Machine learning based estimation of aboveground biomass in subalpine forests using Landsat 8 OLI and Sentinel-2B images in Jiuzhaigou National Nature Reserve, Eastern Tibet Plateau, EGU General Assembly 2021, online, 19–30 Apr 2021, EGU21-4616, https://doi.org/10.5194/egusphere-egu21-4616, 2021.

Aboveground plant water storage (APWS), the total of water storing in aboveground parts of plant, has the function of sustaining the balance between water loss by transpiration and water gain of root uptake. APWS is also essential for plants and hydrological cycle, particularly for semi-arid areas, where water availability is limited. However, APWS varies spatially due to the heterogeneity of natural areas that are composed of a large variety of vegetation types, and studies on the spatial variability of APWS are quite limited in semi-arid areas. To fill this knowledge gap, we established 55 inventory plots with 36 plots in forests and 19 plots in shrubs to detect the spatial variability of APWS using a Random Forest (RF) algorithm and Sentinel-2 images in Mao County, China. Field observations indicated that APWS varied significantly with ecosystems, with the highest APWS in forests. Regardless of ecosystem type, mean APWS in Mao Country was 117.63 Mg ha-1. 10-fold cross-validation suggested that the RF model could reasonably predict APWS (model efficiency = 0.68, root mean square error = 54 Mg ha-1), enabling to capture the spatial variability of APWS. A robust spatial variability of APWS was observed with the highest APWS in forests located high altitude areas, while the lowest APWS was found in shrubs located in low altitude areas. Total APWS was 3.39×107 Mg across the whole study area, which could be used as a valuable natural resource for the semi-arid area. Our study successfully explored the spatial variability of APWS, suggesting the capability of detecting APWS using Sentinel-2 and providing essential data evidence for environmental protection for semi-arid areas.

How to cite: Liu, L., Li, S., Luo, X., Yang, W., Zhang, Y., Lei, J., Chen, G., and Tang, X.: Estimation of aboveground plant water storage using Sentinel-2 images in a semi-arid area, EGU General Assembly 2021, online, 19–30 Apr 2021, EGU21-5111, https://doi.org/10.5194/egusphere-egu21-5111, 2021.

Remote sensing (RS) techniques have great potentials for earth surface monitoring. Nevertheless, for most low to moderate resolution satellites, the problem of mixed pixels with information from the vegetation of interest and the background surfaces can cause large biases in signals and also in their interpretations. This is especially so in low-density forests and semi-arid ecosystems. Ground-level multispectral instruments reduce these effects by measuring at close range to the canopy. However, little work has been published on partitioning the contributions from vegetation and the background elements for both approaches.

This work was motivated by the observed mismatch between data for the same ecosystem from Landsat 8 satellite and Skye radiometer installed on a flux tower in a low-density semi-arid pine forest from 2013-2019. Data from both sources showed similar seasonal patterns in NDVI, but large differences in the reflectance bands. This was most prominent in the NIR reflectance, which showed an opposite seasonal cycle in the two sensors. Thus, similar changes in NDVI were produced by different signals. We hypothesized that the different contributions of the surface components (canopy, shaded areas, and exposed soil) in the footprint areas of the two sensors can explain, and can help correct, these differences.  

Multispectral images with a spatial resolution of 5 cm were captured monthly using an Unmanned Aerial Vehicle (UAV) from April 2018 to November 2019. Reflectance-based algorithms were developed to identify and estimate the fraction and reflectance from the canopy, shaded areas, and open soil. This information was, in turn, applied in the equivalent nadir-viewing satellite pixel. For the tower-based Skye footprint, the same quantities were calculated from its 90° angle of view and the 3D canopy data.

The results showed a canopy fraction of 45% and 95% in the Landsat 8 and Skye footprints. The remaining soil fraction showed a similar seasonal cycle in NDVI as the canopy, but different in the NIR reflectance. The partition between exposed and shaded soil was related to the sun angle, with the exposed soil having a NIR seasonal cycle opposite to that of the vegetation (correlating with soil moisture), and shaded soil having a weak NIR signal variably diluting the overall pixel NIR signal. Differences in the red reflectance were smaller with less effects on the seasonal NDVI cycles.

The results demonstrated firstly, that accounting for the fractional contributions of the surface components can reconcile differences between satellite and ground-based RS. Secondly, vegetation indices such as NDVI obtained by satellite RS in low-density forests can provide misleading information, despite its apparent correlation with certain vegetation variables.

How to cite: Wang, H., Muller, J., Tatarinov, F., Rotenberg, E., and Yakir, D.: Using high-resolution UAV spectral images to disentangle soil, shade, and tree contributions to satellite vegetation indices in sparse dry forests, EGU General Assembly 2021, online, 19–30 Apr 2021, EGU21-9189, https://doi.org/10.5194/egusphere-egu21-9189, 2021.

Shifts in climate driven by anthropogenic land use and land cover change are expected to alter various land–atmosphere interactions. Evapotranspiration (ET) is one of these processes and plays a fundamental role in the hydrologic cycle. Using gridded reanalysis and remote sensing data, we investigated the spatiotemporal trends of precipitation, temperature, and ET for croplands and forest areas in the Baltic states where these land cover type had not changed from 2000 to 2018. We focused on ET but investigated the spatiotemporal trends for the three variables at monthly, seasonal, and annual time scales during this period to quantify trade-offs among months and seasons. We used the Mann-Kendall test and Sen’s slope to calculate the trends and rate of change for the three variables. Although precipitation showed fewer statistically significant increasing and decreasing trends due to its high variability, temperature showed only increasing trends in all time scales. The increasing trends were concentrated in late spring (May, +0.14ºC per year), summer (June and August, +0.10ºC), and early autumn (September, +0.13ºC). For unchanged forest and cropland areas, we found no statistically significant ET trends. However, Sen’s slope indicated increasing ET in April, May, June, and September for forest areas and in May and June for cropland. Our results indicate that during the study period, the temperature changes may have lengthened the growing season, which affected the ET patterns of forest and cropland areas. The results also provide important insights into the regional water balance, specially for critical periods where the ET rates increase while precipitation decrease (May, June. and July). Moreover, our study also complements the findings of other studies over the Baltic states.

How to cite: Montibeller, B., Jaagus, J., Mander, Ü., and Uuemaa, E.: Evapotranspiration intensification over unchanged temperate vegetation in the Baltic states is being driven by climate shifts, EGU General Assembly 2021, online, 19–30 Apr 2021, EGU21-11098, https://doi.org/10.5194/egusphere-egu21-11098, 2021.

At the end of the 1980s the Municipal Department for Environmental Protection of Vienna - MA 22 initiated a detailed biotope mapping on the basis of the Viennese nature conservation law. Approximately 40 % of Vienna’s city area were covered, however only 2 % of the densely populated areas. This biotope mapping was the basis for the biotope types mapping (2005-2011) and of the green areas monitoring (2005). An update of these surveys has been planned in order to meet the various requirements of urban nature conservation and the national and international, respectively, legal monitoring and reporting obligations.

Since the 1970s the municipality of Vienna has built up a comprehensive database and uses state-of-the-art methods for collecting geodata carrying out services for surveying, airborne imaging and laser-scanning. Currently systems for mobile mapping, oblique aerial photos and a surveying flight with a single photon LiDAR system are being implemented or prepared. Because of the numerous high-resolution data available within the municipality and limitations mainly in spatial resolution of satellite data, the City of Vienna saw no need or benefit in integrating satellite images until now.

However, satellite data are now available within the European Copernicus program, which have considerable potential for monitoring green spaces and biotope types due to their high temporal resolution and the large number of spectral channels and SAR data. For the first time, the Sentinel-1 mission offers a combination of high spatial resolution in Interferometric Wide Swath (IW) recording mode and high temporal coverage of up to four shots every 12 days in cross-polarization in the C-band. The Sentinel-2 satellites deliver multispectral data in 10 channels every 5 days with spatial resolutions of 10 or 20 m.

Within the SeMoNa22 project, various indicators are derived for the Vienna urban area (2015-2020) and used for object-oriented mapping and classification of biotope types and characterization of the green space:

• Sentinel-1 data (→ time series on the annual cycles in the backscattering properties of the vegetation, phenology),

• Sentinel-2 data (→ multispectral time series via parameters for habitat classification / vegetation indices),

• High-resolution earth observation data (airborne laser scanning (ALS), image matching, orthophoto → various parameter describing the horizontal and vertical vegetation structure).

The main goals of SeMoNa22 is to explore efficient and effective ways of knowing if, how and to what extent the data collected can form the basis and become an integrative part of urban conservation monitoring. For this purpose, combinations of different earth observation data (satellite- and aircraft- supported or terrestrial sensors) and existing structured fieldwork data collections (species mapping, soil parameters, meteorology) are examined by means of pixel- and object-oriented methods of remote sensing and image processing. The study is done for several test sites in Vienna covering different ecosystems. In this contribution the ongoing SeMoNa22 project will be presented and first results will be shown and discussed.

How to cite: Iglseder, A., Immitzer, M., Bauerhansl, C., Hoffert-Hösl, H., Kramer, K., Kasper, A., Schnetz, M. E., Lehner, H., and Hollaus, M.: Potential of Sentinel and high resolution EO data for monitoring nature protection in cities – the SeMoNa22 project, EGU General Assembly 2021, online, 19–30 Apr 2021, EGU21-14211, https://doi.org/10.5194/egusphere-egu21-14211, 2021.

To gain a better understanding of global carbon storage and albedo feedback mechanisms it is important to have insights into high latitude vegetation change. Boreal forest compositions are changing in response to changes in climate, which in turn can lead to feedbacks in regional and global climate through altered carbon cycles and albedo dynamics. Circumpolar boreal forests represent close to 30% of all forested area on the planet, between 900 and 1,200 million ha. These forests are located primarily in Alaska, Canada, and Russia. Due to the remote location of these forests and the short seasons without snow, data collected on the boreal vegetation is limited. 

The proposed dataset is an attempt to remedy data scarcity whilst providing adjusted data for machine learning practices.We present a dataset containing diverse formats of forest structure information that covers two important vegetation transition zones in Siberia: the Evergreen - Summergreen transition zone in Central Yakutia and the northern treeline in Chukotka (NE Siberia).

This dataset contains data from the locations covered by fieldwork was performed by the Alfred Wegener Institute for Polar and Marine research, (AWI) and the North-Eastern Federal University of Yakutsk​ (NEFU). The fieldwork upscaled through the addition of Red Green Blue(RGB) UAV (Unmanned Aerial Vehicle) camera data and Sentinel-2 satellite data cropped to a 5 km radius around the fieldwork sites. The dataset is created with the aim of providing ground truth validation and training data to be used in various vegetation related machine learning tasks .

The dataset contains:

1.Labelled individual trees per 30x30 m plot assigned in field work with additional data on species, height, crown width, and biomass.

2.Structure from Motion (SfM)point clouds that provide 3D information about the forest structure, included generated Canopy Height Model (CHM), Digital Elevation Model (DEM) and a Digital Surface Model (DSM) per 50x50 m.

3.Multispectral Sentinel-2 satellite data (10 m ) cropped to a 5km radius with generated a NDVI(normalized difference vegetation index), available in three seasons: Early Summer, Peak Summer and Late Summer.

4.Extracted tree crowns with species information and a synthetically generated large (10.000 samples) dataset for training machine leaning algorithms.

The dataset will be made publicly available on the data repository PANGAEA.

How to cite: van Geffen, F., Heim, B., Herzschuh, U., Pestryakova, L., Zakharov, E., Hänsch, R., Demir, B., Kleinschmit, B., Förster, M., and Kruse, S.: SiDroForest  Siberian Drone-mapped Forest inventory, EGU General Assembly 2021, online, 19–30 Apr 2021, EGU21-15106, https://doi.org/10.5194/egusphere-egu21-15106, 2021.

Sustainable forest management increasingly highlights the maintenance of biological diversity and requires up-to-date information on the occurrence and distribution of key ecological features in forest environments. Different proxy variables indicating species richness and quality of the sites are essential for efficient detecting and monitoring forest biodiversity. European aspen (Populus tremula L.) is a minor deciduous tree species with a high importance in maintaining biodiversity in boreal forests. Large aspen trees host hundreds of species, many of them classified as threatened. However, accurate fine-scale spatial data on aspen occurrence remains scarce and incomprehensive.

We studied detection of aspen using different remote sensing techniques in Evo, southern Finland. Our study area of 83 km² contains both managed and protected southern boreal forests characterized by Scots pine (Pinus sylvestris L.), Norway spruce (Picea abies (L.) Karst), and birch (Betula pendula and pubescens L.), whereas European aspen has a relatively sparse and scattered occurrence in the area. We collected high-resolution airborne hyperspectral and airborne laser scanning data covering the whole study area and ultra-high resolution unmanned aerial vehicle (UAV) data with RGB and multispectral sensors from selected parts of the area. We tested the discrimination of aspen from other species at tree level using different machine learning methods (Support Vector Machines, Random Forest, Gradient Boosting Machine) and deep learning methods (3D convolutional neural networks).

Airborne hyperspectral and lidar data gave excellent results with machine learning and deep learning classification methods The highest classification accuracies for aspen varied between 91-92% (F1-score). The most important wavelengths for discriminating aspen from other species included reflectance bands of red edge range (724–727 nm) and shortwave infrared (1520–1564 nm and 1684–1706 nm) (Viinikka et al. 2020; Mäyrä et al 2021). Aspen detection using RGB and multispectral data also gave good results (highest F1-score of aspen = 87%) (Kuzmin et al 2021). Different remote sensing data enabled production of a spatially explicit map of aspen occurrence in the study area. Information on aspen occurrence and abundance can significantly contribute to biodiversity management and conservation efforts in boreal forests. Our results can be further utilized in upscaling efforts aiming at aspen detection over larger geographical areas using satellite images.

How to cite: Kumpula, T., Mäyrä, J., Kuzmin, A., Viinikka, A., Kivinen, S., Tanhuanpää, T., Hurskainen, P., Keski-Saari, S., Kullberg, P., Poikolainen, L., Korpelainen, P., Ritakallio, A., Tuominen, S., and Vihervaara, P.: Detecting a keystone species European aspen in boreal forests with airborne hyperspectral, LiDAR and UAV data with machine learning methods, EGU General Assembly 2021, online, 19–30 Apr 2021, EGU21-16273, https://doi.org/10.5194/egusphere-egu21-16273, 2021.

The relative concentrations of photosynthetic and photoprotective pigments provide important information about the physiological state of the plant and are determined, among other things, by the lighting regime and the presence of nutrients. Relative composition of the pigments is depending on the physiological response of the plant to external influences. In most cases, when an on-line in-situ analysis is required, only the main pigments are measured: Chla, Chlb and a rough estimate of the "total carotenoids" in higher plants, but such an estimate may not always be reliable. Differential Optical Absorption Spectroscopy (DOAS) is known for its applications for the trace gases measurements in the atmosphere sciences; however, no application has been found for the determination of color pigments for plant extracts. For the correct application of the DOAS method, it is necessary to determine the appropriate optical thickness of the sample under study, the fitting intervals for analysis, as well as a set of absorption cross sections for the target pigments.

Purpose of the work is to determine the appropriate settings for the retrieval of concentrations of colored pigments employing the DOAS method by investigating the sample of pine and spruce needles extraction. The relevance of the work consists in the development of a new method for analyzing transmission spectra, which does not require the creation of specialized software, since programs for analyzing spectra by the DOAS method are available.

For the spectra registration, Solar M150 spectrometer with Hamamatsu S7031-1006S detector has been used, the transmission spectra recorded in the 330 - 750 nm range, and pure acetone employed as a solvent. The paper presents the results of DOAS-analysis of extracts of various coniferous samples, from which it was possible to retrieve the contents of Cha, Chb, B,b-carotene, B,e-carotene, and small amounts of Phaeophytin-a, Neoxanthin. Optimal settings for the DOAS-analysis and experimental setup details for photosynthetic and photoprotective pigments retrieval are discussed.

How to cite: Bruchkouski, I., Siliuk, V., Guliaeva, S., and Litvinovich, H.: Quantitative measurements of pigments in coniferous by the method of differential optical absorption spectroscopy, EGU General Assembly 2021, online, 19–30 Apr 2021, EGU21-12980, https://doi.org/10.5194/egusphere-egu21-12980, 2021.

Leaf phenology, the timing of leaf life cycle events, is a vital indicator of terrestrial biosphere function. The influence of global change upon leafing phenology in mid to high latitude regions is uncertain due to a complex interaction of drivers and lack of temporally and spatially resolved baseline data.  Leaf phenology has been observed manually for millennia, and through satellite platforms for decades. A novel technique of monitoring leaf phenology known as near remote sensing employing time-lapse photography at the canopy level (or phenocams) allows for objective observations with high temporal and spatial resolution. We deployed 13 solar-powered time-lapse camera stations across a climate gradient in Nova Scotia, Canada to observe leaf phenology of locally abundant species including more than 300 individuals over the 2019 and 2020 growing seasons. To examine the influence of thermal, photoperiodic, and genetic drivers, our remote phenology monitoring stations were situated in comparative edaphic and topographic contexts and complemented with relative humidity and ambient temperature sensors. We observed variability in the timing of leaf budburst, peak of season greenness, redness, senescence, and abscission between and within species, despite similar degrees of environmental forcing. Moving forward, we will apply our insights to develop species specific process based models of leaf phenology, and test the wider application of our techniques to observational records from other regions. This work demonstrates the complexity of environmental influence upon leaf phenology, as well as the utility of phenocams in monitoring leafing phenology in remote regions of Maritime Canada.

How to cite: Spafford, L. and MacDougall, A. H.: Heterogenous Patterns in Leaf Phenology Across a Climate Gradient in Maritime Canada Observed through Phenocams, EGU General Assembly 2021, online, 19–30 Apr 2021, EGU21-83, https://doi.org/10.5194/egusphere-egu21-83, 2021.

Sentinel-2 time series provide large amounts of data and information which can be easily used to classify tree species with machine learning algorithms. In addition to the original Sentinel-2 bands, further data such as indices, phenological metrics or even synthetical images can be derived. While tree species classifications highly benefit from such additional data resulting in improved prediction accuracy, severe drawbacks have to be considered - For large data sets, large storage is needed, the computation time expands and a linkage to ecological or phenological reasons behind the usage of these variables can hardly be drawn. Therefore, the implemented variables should be limited to the ones, which are meaningful and on the same time providing the best prediction accuracy. To identify meaningful variables from original Sentinel-2 images and the additionally calculated data first we used basic correlation analyses and subsequently feature selection methods in combination with the commonly used Random Forest algorithm. We classified the most common forest tree species in the Swiss canton of Grisons, which is mountainous and characterized by diverse landscapes. The presented approach will lead to higher efficiency for classifying tree species and additionally provides potential conclusions regarding ecological patterns beyond the distinction of tree species by remote sensing data. Moreover, the proposed approach can also be used to improve classifications or predictions of other outcome variables for vegetated areas with Sentinel-2.

How to cite: Koch, T. L., Rüetschi, M., and Waser, L. T.: The identification of meaningful variables from Sentinel-2 time series data for effective tree species classification, EGU General Assembly 2021, online, 19–30 Apr 2021, EGU21-14341, https://doi.org/10.5194/egusphere-egu21-14341, 2021.

With consumer-grade unmanned aerial vehicles (UAVs) on the rise, which enable easy, time-flexible, and cost-effective acquisition of very high-resolution RGB data, the mapping of forest tree species using solely RGB imagery is of high interest, as it does not rely on sophisticated sensors, does not require extensive calibration and preprocessing and, therefore, enables the application by a wide audience. In combination with convolutional neural networks (CNNs), which particularly exploit spatial patterns and, therefore, highly benfit from very high-resolution remote sensing, this offers great potential for accurately mapping forest tree species.

Here, we present the findings of our recent study, in which we used very high-resolution RGB imagery from UAVs in combination with CNNs for the mapping of forest tree species. In this study, we used multicopter UAVs to obtain very high-resolution (<2 cm) RGB imagery over 51 ha of temperate forests in the Southern Black Forest region, and the Hainich National Park in Germany. To fully harness the end-to-end learning capabilities of CNNs, we used a semantic segmentation approach (U-net) that concurrently segments and classifies tree species from imagery. With a diverse dataset in terms of study areas, site conditions, illumination properties, and phenology, we accurately mapped nine tree species, three genus-level classes, deadwood, and forest floor (mean F1-score 0.73). We found that a coarser spatial resolution substantially reduced the model accuracy (mean F1-score of 0.26 at 32 cm resolution) and that larger tile sizes during CNN training negatively affected the model accuracies for underrepresented classes. Additional height information from normalized digital surface models slightly increased the model accuracy but simultaneously increased computational complexity and data requirements. Our results highlight the key role that UAVs can play in the mapping of forest tree species, given that air- and spaceborne remote sensing currently does not provide comparable spatial resolutions. Given the end-to-end learning capabilites of CNNs extensive preprocessing becomes partly obsolete, whereas the use of a large and diverse dataset facilitates a high degree of generalization of the CNN, thus fostering transferability. The synergy of high-resolution UAV imagery and CNN provide a fast and flexible yet accurate means of mapping forest tree species.

In this contribution, we will give an outlook on how the combination of UAV imagery and CNNs can be integrated with multitemporal satellite imagery (Sentinel-1 and Sentinel-2) in order to extrapolate the UAV-based tree species maps to larger areas.

How to cite: Schiefer, F., Kattenborn, T., Frick, A., Frey, J., Schall, P., Koch, B., and Schmidtlein, S.: Mapping forest tree species in high resolution UAV-based RGB-imagery by means of convolutional neural networks, EGU General Assembly 2021, online, 19–30 Apr 2021, EGU21-12957, https://doi.org/10.5194/egusphere-egu21-12957, 2021.

Drying out of coniferous trees (Picea abies) due to bark beetle infestation and other diseases leads to a high rate of conifers mortality. The coniferous forests in Belarus are largely exposed to damage by the bark beetle, the early symptoms of which are the changes in the color and loss of shine of the needles.  

Purpose of the work is to identify drying out stages combining the TripleSat multispectral satellite data (spatial resolution 3.2 m MS, 0.8 m PAN, bands R, G, B, NIR) for the test coniferous forest area in Belarus (53.65419º N, 27.640213º E) with quasi-synchronous airborne photo-spectral measurements which have been used as a reference data. Airborne measurements of reflectance coefficient function of underlying coniferous trees have been carried out by employing two spectrometers (wavelength range 400-900 nm, spectral resolution 4.3 nm) and photo-camera (visible range, FOV 50º) mounted on board of Diamond DA40NG aircraft in nadir geometry.  

Airborne RGB-images have been used for visual identification of the type of underlying surface and for subsequent training data set formation. Training data consist of several sets (10 – 20) of vegetation indexes for each type of underlying surface. The linear discriminant analysis (LDA) classification algorithm has been applied in this study for distinguishing the conifers drying out stages. A set of vegetation indices evaluated for each reflectance coefficient function has been applied as input data for LDA classification algorithm.

LDA classification algorithm has been employed to the TripleSat image for identification drying out stages of coniferous trees. The reference data for LDA classification algorithm of the TripleSat image included the combination of coordinates and corresponding types of underlying surface obtained from the results of the airborne experiment classification. A set of vegetation indices has been derived for each pixel of the image and used as input data for LDA algorithm; also vegetation indices calculated for the reference pixels have been applied for training data set formation.

The classification accuracy of three conifers drying out stages based on the airborne experiment is estimated to be in a range of 27 - 74%. The verification of TripleSat classification results has been performed by visual comparison with high resolution aerial images.

How to cite: Litvinovich, H., Guliaeva, S., Bruchkouski, I., Siliuk, V., and Katkouski, L.: Drying out conifers classification employing TripleSat satellite data, EGU General Assembly 2021, online, 19–30 Apr 2021, EGU21-15269, https://doi.org/10.5194/egusphere-egu21-15269, 2021.

LiDAR-based forest inventories focusing on estimating and mapping structure-related forest inventory variables across large areas have reached operationality. In the commonly applied area-based approach, a set of field-measured inventory plots is combined with spatially co-located airborne laserscanning data to train empirical models that can then be used to predict the target metric over the entire area covered by LiDAR data.

The area-based approach was found to produce reliable estimates for structure-related forest inventory metrics such as wood volume and biomass across many forest types. However, the current workflows still leave space for improvement that may result in cost-reduction with respect to data acquisition or improved accuracies. This is particularly relevant as the area-based approach is increasingly used in operational forestry settings. To further optimize existing workflows, experiments are required that need large amounts of forest inventory data (e.g., to examine the effect of sample size or the field inventory design on the model performances) or multiple LiDAR acquisitions (e.g., to identify optimal/cost-efficient acquisition settings). The acquisition of these types of data is cost-intensive and is hence often limited to small extents within scientific experiments.

Here, we present the ”GeForse - Generating Synthetic Forest Remote Sensing Data” approach to create synthetic LiDAR datasets suitable for such optimization studies. GeForse combines a database of single-tree models consisting of point clouds extracted from real LiDAR data with the outputs of a spatially explicit, single tree-based forest growth simulator (in this case SILVA). For each simulated tree, we insert a real point-cloud tree with properties (species, crown diameter, height) matching the properties of the simulated tree. This results in a synthetic 3D forest with a realistic 3D-structure where the inventory metrics of each tree are known. This 3D forest then serves as input to the “Heidelberg LiDAR Operations Simulator” (HELIOS++, https://github.com/3dgeo-heidelberg/helios) and thereby enables the simulation of LiDAR acquisition flights with varying acquisition settings and flight trajectories. In combination with the “full inventory” of all trees in the simulated forest, this enables a wide variety of sensitivity analyses.

In this contribution, we give an overview of the complete GeForse approach from extracting the tree models, to generating the 3D forest and simulating LiDAR flights over the 3D forest using HELIOS++. Further, we present a brief case-study where this approach was applied to optimize certain aspects of area-based forest inventory approaches using LiDAR data from a forest area in central Europe. Finally, we provide an outlook on future application fields of the GeForse approach.

How to cite: Fassnacht, F. E., Schäfer, J., Weiser, H., Winiwarter, L., Krašovec, N., Latifi, H., and Höfle, B.: Presenting the GeForse approach to create synthetic LiDAR data from simulated forest stands to optimize forest inventories, EGU General Assembly 2021, online, 19–30 Apr 2021, EGU21-9197, https://doi.org/10.5194/egusphere-egu21-9197, 2021.

Virtual laser scanning (VLS) is a valuable method to complement expensive laser scanning data acquisition in the field. VLS refers to the simulation of LiDAR to create 3D point clouds from models of scenes, platforms and sensors mimicking real world acquisitions. In forestry, this can be used to generate training and testing data with complete ground truth for algorithms performing essential tasks such as tree detection or tree species classification. Furthermore, VLS allows for the in-depth investigation of the influence of different acquisition parameters on the point clouds and thus also the behaviour of algorithms, which is important when relating point cloud metrics to forest inventory variables. Finally, VLS can be used for acquisition planning and optimisation, as different configurations can be tested regarding their ability to create data of the required quality with minimal effort. For these purposes, we developed the open source Heidelberg LiDAR Operations Simulator HELIOS++ (written in C++) which is available on GitHub (https://github.com/3dgeo-heidelberg/helios), as a precompiled command line tool, and as Python package (pyhelios). HELIOS++ provides a high-fidelity framework for full 3D laser scanning simulations with multiple platforms and a flexible system to represent the scene. HELIOS++ models the beam divergence and supports the recording of the full waveform.

One important premise for the usefulness of VLS data is the use of an adequate 3D scene in the simulation. In this context, we conducted a study investigating point clouds simulated based on opaque voxel-based forest models computed from terrestrial laser scanning data using different voxel sizes. Coupling the LiDAR simulation with a database containing point clouds of single trees from terrestrial, UAV-borne and airborne acquisitions, allowed us to compare metrics derived from real and simulated data. Furthermore, by including the tree neighbourhood in the scene, we were able to consider occlusion effects between the trees.

We found that the voxel size is an important parameter, where values of e.g. 0.25 m lead to unrealistic occlusion effects of the mid- and understory, as only few gaps remain in the forest models through which the laser beam can pass. This results in fewer multiple returns, the vertical point distribution is shifted upwards, and tree metrics such as crown projection area and crown base height are estimated poorly. Smaller voxel sizes are therefore preferable, though the appropriate voxel size depends on the resolution of the input point cloud. With very small voxels, the voxel model may become too transparent. To achieve realistic simulations without the need for a high number of voxels we suggest variable downscaling of voxel cubes based on appropriate local metrics such as the plant area density. This approach decreases the computational requirements for the simulation, as fewer primitives are present in the scene. In our study, the use of such scaled voxels derived for a grid size of 0.25 m achieves equally and partly more reliable estimates of point cloud and tree metrics than regular voxels at fixed side lengths of 0.05 and 0.02 m.

How to cite: Weiser, H., Winiwarter, L., Schäfer, J., Fassnacht, F. E., Anders, K., Esmorís Pena, A. M., and Höfle, B.: Virtual laser scanning (VLS) in forestry – Investigating appropriate 3D forest representations for LiDAR simulations with HELIOS++, EGU General Assembly 2021, online, 19–30 Apr 2021, EGU21-9178, https://doi.org/10.5194/egusphere-egu21-9178, 2021.

Trees supply a multitude of ecosystem services (e.g. carbon storage, suppression of air pollution, oxygen, shade, recreation etc.) not only in forested areas but also in urban landscapes. Many of these services are positively correlated with tree size and structure. The assessment of carbon storage potential via the quantification of above ground biomass (AGB) is of special importance. However, quantification of AGB is difficult and applied allometries are often based on forest trees, which are subject to very different growing conditions, competition and form compared to urban trees. In this contribution, we highlight the potential of terrestrial laser scanning (TLS) techniques to extract high detailed information on tree structure and AGB with a focus on urban trees.

A total of 55 urban trees distributed over eight cities in Switzerland were measured using TLS and traditional forest inventory techniques before they were felled and weighted. Tree structure, volumes and AGB from the TLS point clouds were extracted using Quantitative Structure Modelling (QSM). TLS derived AGB estimates were compared to allometric estimates dependent on diameter at breast height only. The allometric models were established within the Swiss National Forest Inventory and are therefore optimised for forest trees.

TLS derived AGB estimates showed good performance when compared to destructively harvested references with an R² of 0.954 (RMSE = 556 kg), compared to an R² of 0.837 (RMSE = 1159 kg) for allometrically derived AGB estimates. A correlation analysis showed that different TLS derived wood volume estimates as well as trunk diameters and tree crown metrics show high correlation in describing total wood AGB.

The presented results show that TLS based wood volume estimates show high potential to estimate tree AGB independent of tree species, size and form. This allows us to retrieve highly accurate, non-destructive AGB estimates that could be used to establish new allometric equations without the need of extensive destructive harvest.

How to cite: Kükenbrink, D., Gardi, O., Morsdorf, F., Thürig, E., Schellenberger, A., and Mathys, L.: Above Ground Biomass References for Urban Trees from Terrestrial Laser Scanning Data, EGU General Assembly 2021, online, 19–30 Apr 2021, EGU21-5399, https://doi.org/10.5194/egusphere-egu21-5399, 2021.

Fuel management is a crucial action to maintain wildland fires under the threshold of manageability; hence, in order to allocate resources in the best way, wildland fuel mapping is regarded as a necessary tool by land managers. Several studies have used Aerial Laser Scanner (ALS) data to estimate forest fuels characteristics at plot level, but few have extended such estimates at a zonal level.

In the context of the EU Interreg Project CROSSIT SAFER, a test of the possibilities of ALS data to predict fuels attributes has been performed in three different areas: an alpine basin, a coastal wildland-urban interface and a karstic highland. Eighteen sampling plots have been laid out over 6 forest categories, with a special focus on Pinus nigra J. F. Arnold artificial forests. Low density (average 4 points/m²) discrete return LiDAR data has been analysed with FUSION, a free point cloud analysis software tailored to forestry purposes; field and remote sensing data have been connected with simple statistical modelling and results have been spatialised over the case study areas to provide wall-to-wall inputs for FLAMMAP fire behaviour simulation software.

Resulting maps can be of relevance for land managers to better highlight the most vulnerable or fire prone areas at a mesoscale administrative level. Limitations and room for improvement are pointed out, in the view that land management should keep updated with the latest technology available.

How to cite: Taccaliti, F., Venturini, L., Marchi, N., and Lingua, E.: Forest fuel assessment by LiDAR data. A case study in NE Italy, EGU General Assembly 2021, online, 19–30 Apr 2021, EGU21-12755, https://doi.org/10.5194/egusphere-egu21-12755, 2021.

Forest ecosystems represent an important source of income for landowners and at the same time an important source of ecosystem services for the society. Quantitative and qualitative information about timber assortments are particularly important to support sustainable forest management, representing a crucial prerequisite for active forest management. To date, the most accurate methods for assessing the timber assortments availability within forest stands are destructives, and the development of an effective method for deriving these estimates on standing trees is highly needed. This is particularly more evident for mixed forests, which are often subject of the conflict between conservation and productive functions.

This study aims to introduce a stepwise approach for timber assortment estimation and classification using TLS data. The approach consists of four steps: a) timber-leave discrimination, b) tree detection, c) stem reconstruction, and d) timber assortment estimation and classification. The study was carried out in a mixed tree-species and multi-layered Mediterranean forests, observing 178 trees of twelve different species, wherein 66 out of 178 were large trees, with a diameter at breast height higher than 20 cm.

Results indicate that the discrimination between timber and leaves reached 0.98 for accuracy using Random Forest algorithm. All trees with a diameter at breast height higher than 30 cm were correctly identified. The overall detection accuracy was 84.40 % (SD± = 4.7%). Best detection accuracy was found for A. lobelii, S. torminalis, F. excelsior, Q. cerris, A. campestre and F. sylvatica (higher than 84.3%) tree species. 47 out of 66 detected large stems were correctly reconstructed. The stepwise approach allows to classify 168 logs (134 merchantable logs and 34 non-merchantable) extracted from 47 stems through the automatic functions (i.e. cylinder-fitting approach), with an accuracy ranging between 75% (134 out of 179 reference merchantable logs) and 85% (34 out of 40 reference merchantable logs). The overall reconstruction accuracy was 71.40 % (SD± = 17.1%). Best reconstruction accuracy was found for Q. cerris, A. opalus and F. excelsior (higher than 43.5%). Concerning the timber assortment 134 out of 179 merchantable logs were classified in one of the 15 assortment types (i.e. A+, A0, A-). The whole predicted logs were classified in 11 assortment types, so eleven out of 15 assortment types were correctly matched between predicted and reference data. The classification of merchantable logs was more accurate for eight assortment types (A-, B-, B0, B+, C-, D-, D+ and Fuelwood-), which was ±2 merchantable logs. The abovementioned results support the feasibility of this stepwise approach for calculating the timber assortment of standing trees, ensuring the valorisation of the productivity of forest characterized by tree species richness and heterogeneous stand structure.

How to cite: Alvites, C., Santopuoli, G., Hollaus, M., Pfeifer, N., Maesano, M., Moresi, F. V., Marchetti, M., and Lasserre, B.: A stepwise approach for deriving timber assortment of trees from Terrestrial Laser Scanning data, EGU General Assembly 2021, online, 19–30 Apr 2021, EGU21-13509, https://doi.org/10.5194/egusphere-egu21-13509, 2021.

The use of three-dimensional point clouds in forestry is steadily increasing. Numerous algorithms to detect individual trees from point clouds and derive some fundamental inventory parameters have been proposed so far, but they usually provide higher accuracy in coniferous stands than in deciduous one. In the latter kind of stands, indeed, the tree identification is hampered by the geometrical round shape of the crowns, the interlacing branches of adjacent trees and the usual presence of understory vegetation.

In an attempt to overcome these limitations, we developed an algorithm that is innovatively based on the areal point density of the three-dimensional cloud and that provides the height and coordinates of all the trees within a region of interest.

In this work, we apply the algorithm to different situations, ranging from the regularly-arranged plantations to the very interlaced crowns of the naturally established stands, demonstrating how it is able to correctly detect most of the trees and recreate a map of their spatial distribution. We also test its capability to deal with relatively low point density and explore the possibility to use it to recreate time series of vegetation biomass. Finally, we discuss the algorithm’s limitations and potentialities, particularly focusing on its coupling to other existing tools to deal with a wider range of applications in forestry and land management.

How to cite: Camporeale, C., Latella, M., and Sola, F.: Density-Based Individual Tree Detection from Three-Dimensional Point Clouds, EGU General Assembly 2021, online, 19–30 Apr 2021, EGU21-15628, https://doi.org/10.5194/egusphere-egu21-15628, 2021.

Drainage ditches are common in boreal forests and provide better soil aeration and tree growth by draining the wet soils. Peatland drainage in the last century has created about 1 million km of artificial streams in Sweden, making it one of the most widespread human-induced environmental disturbances. This extensive use of ditches over a long period resulted in a major shift in forest hydrology and impacted the ecosystem functions. Therefore, there is a pressing need for an accurate database of forest ditches so that sustainable management can be practiced to improve the overall functioning of the forest ecosystem. Comparisons with national field datasets show that existing maps only show a small fraction (<10%) of the ditches across the country. To address this knowledge gap, we applied AI methods on high-resolution (1 m) LIDAR data to map drainage ditches in a subset of the Swedish forest. We developed a suite of topographic indices and analyzed those with machine learning and deep learning algorithms to perform automatic ditch detection. Both models produced reasonably accurate results and a substantial improvement over the existing maps in terms of ditch detection. The impoundment index and high pass medium filter from the digital elevation model were among the top predictors of drainage ditches. The study introduced a new avenue for accurate detection of forest ditches across the whole country. Our AI-generated maps of ditches provide effective tools for the restoration of degraded land and support ditch cleaning operations to increase forest growth.

How to cite: Paul, S. S., Ågren, A. M., and Lidberg, W.: Detection of drainage ditches using high-resolution LIDAR data in the Swedish forest, EGU General Assembly 2021, online, 19–30 Apr 2021, EGU21-2226, https://doi.org/10.5194/egusphere-egu21-2226, 2021.

Survey-grade drone laser scanners suitable for unmanned aerial vehicles (UAV-LS) allow the efficient collection of finely detailed three-dimensional information of tree structures. This data type allows forests to be resolved into discrete individual trees and has shown promising results in providing accurate in-situ observations of key forestry variables. New and improved approaches for analyzing UAV-LS point clouds have to be developed to transform the vast amounts of data from UAV-LS into actionable insights and decision support. Many different studies have explored various methods for automating single tree detection, segmentation, parsing into different tree components, and measurement of biophysical variables (e.g., diameter at breast height). Despite the considerable efforts dedicated to developing automated ways to process UAV-LS data into useful data, current methods tend to be tailored to small datasets, and it remains challenging to evaluate the performance of different algorithms based on a consistent validation dataset. To fill this knowledge gap and to further advance our ability to measure forests from UAV-LS data, we present a new benchmarking dataset. This data is composed of manually labelled UAV-LS data acquired a number of continents and biomes which span tropical to boreal forests. The UAV-LS data was collected exclusively used survey-grade sensors such as the Riegl VUX and mini-VUX series which are characterized by a point density in the range 1 – 10 k points m². Currently, such data represent the state-of-the-art in aerial laser scanning data. The benchmark data consists of a library of single-tree point clouds, aggregated to sample plots, with each point classified as either stem, branch, or leaves. With the objective of releasing such a benchmark dataset as a public asset, in the future, researchers will be able to leverage such pre-existing labelled trees for developing new methods to measure forests from UAV-LS data. The availability of benchmarking datasets represents an important driver for enabling the development of robust and accurate methods. Such a benchmarking dataset will also be important for a consistent comparison of existing or future algorithms which will guide future method development.

How to cite: Stefano, P., Pears, G. D., Watt, M. S., Mitchard, E., McNicol, I., Bremer, M., Rutzinger, M., Surovy, P., Wallace, L., Hollaus, M., and Astrup, R.: A new drone laser scanning benchmark dataset for characterization of single-tree and forest biophysical properties, EGU General Assembly 2021, online, 19–30 Apr 2021, EGU21-10677, https://doi.org/10.5194/egusphere-egu21-10677, 2021.