The total quantity of carbon fixed by photosynthesis per unit time in an ecosystem is referred to as gross primary productivity (GPP). This is an important activity in the Earth’s carbon cycle. In near-equilibrium conditions, GPP is calculated as the sum of net carbon exchange during the day plus ecosystem respiration. In our study, we compared the GPP from satellite-based model estimates with the actual GPP calculated by the eddy covariance method available from the FLUXNET database in Borneo, Southeast Asia. We found that the GPP models were unable to capture the actual daily fluctuations of GPP in tropical vegetation in Borneo, although there were moderate correlations when comparing GPP from two different remote sensing models (e.g. the GPP derived from the Vegetation Photosynthesis Model (VPM) has a moderate correlation with GPP products from Moderate Resolution Imaging Spectroradiometer (MODIS) Terra/Aqua for Maludam tropical peat swamp forest vegetation). Parameterization was required to improve the GPP models, which included reanalyzing each model parameter. These parameter include LUI (light use efficiency), which is a challenging model parameter to measure but is critical in determining GPP, and cloud cover on MODIS satellite data, which determines the quality of remote sensing indices such as LSWI (land surface water index), due to the importance of this index as a proxy for W_scalar to estimate VPM GPP.

How to cite: Ginting, Y. R. S. and Esters, L.: How accurately does gross primary productivity derived from remote sensing-based models represent the products from field measurements? Case studies of tropical vegetation in Borneo, Southeast Asia, EGU General Assembly 2024, Vienna, Austria, 14–19 Apr 2024, EGU24-405, https://doi.org/10.5194/egusphere-egu24-405, 2024.

Quantifying Gross Primary Production (GPP) is fundamental for understanding terrestrial carbon dynamics, particularly in forests. The overarching question we address here is whether integrating remote sensing (RS) with deep learning (DL) methodologies can enhance the estimation of daily forest GPP on a European scale.

The Eddy Covariance (EC) method, although widely used to infer ecosystem-scale estimates of GPP from in situ CO2 exchange measurements, suffers from limited global coverage. When EC data are not available, RS data are often employed to estimate GPP by establishing statistical relationships with in situ observations. Recently, Machine Learning (ML) strategies, particularly involving RS and meteorological inputs, have been used for estimating GPP continuously in space and time.

However, the potential of DL techniques, particularly those exploiting the sequential characteristics of time series data (i.e. recurrent neural network architectures) in estimating daily forest GPP has not been comprehensively explored. This includes their performance during photosynthetic downregulation (e.g. during climate extremes such as droughts) and an in-depth examination of the importance of the utilised features.

This study presents a comparative analysis of three recurrent neural network architectures—Recurrent Neural Networks (RNNs), Gated Recurrent Units (GRUs), and Long-Short Term Memory (LSTMs)—for daily forest GPP estimation across ICOS ecosystem stations. We assess their performance throughout the entire year, during the growing season, and under photosynthetic downregulation events. This assessment utilises RS inputs —optical data from Sentinel-2, Land Surface Temperature (LST) from MODIS, and radar data from Sentinel-1—, combined with potential radiation data. Furthermore, we analysed the importance of these features to provide insights into the complex interactions and dependencies when estimating GPP.

Our results indicate that all three architectures yield similar accuracy for full period and growing season GPP estimations, with mean NRMSE values of 0.136 and 0.170, respectively. All models exhibit increased errors while estimating GPP during photosynthetic downregulation events and their performance varies notably, with LSTMs showing the best results (NRMSE=0.202), followed by RNNs (NRMSE=0.214), and GRUs exhibiting the least efficacy (NRMSE=0.276). Additionally, our study underscores the importance of potential radiation as a critical feature influencing the GPP response, with LST data proving particularly valuable when estimating GPP during photosynthetic downregulation events, more so than optical or radar data.

These findings suggest that recurrent neural network architectures, especially LSTMs, are effective in daily GPP estimations, with slight performance decreases during climate extreme conditions. The study also reveals the efficacy of combining RS data with potential radiation for accurate forest GPP estimation, with LST data being especially crucial when estimating GPP during photosynthetic downregulation events.

Our research paves the way for further exploration into recurrent models and innovative architectures, with an emphasis on overcoming the challenges in modelling climate-induced GPP extremes.

How to cite: Montero, D., Mahecha, M. D., Martinuzzi, F., Aybar, C., Klosterhalfen, A., Knohl, A., Koebsch, F., Anaya, J., and Wieneke, S.: Estimating Gross Primary Production via Recurrent Neural Networks: A comparative analysis, EGU General Assembly 2024, Vienna, Austria, 14–19 Apr 2024, EGU24-6773, https://doi.org/10.5194/egusphere-egu24-6773, 2024.

Forest net primary productivity (NPP) constitutes a key flux within the terrestrial ecosystem carbon cycle and serves as a significant indicator of the forests carbon sequestration capacity, which is closely related to forest age. Despite its significance, the impact of forest age on NPP is often ignored in future NPP projections. Here, we mapped forest age in Hunan Province at a 30 m resolution utilizing a combination of Landsat time series stack (LTSS), national forest inventory (NFI) data, and the relationships between height and age. Subsequently, NPP was derived from NFI data and the relationships between NPP and age was built for various forest types. Then forest NPP was predicted based on the NPP-age relationships under three future scenarios, assessing the impact of forest age on NPP. Our findings reveal substantial variations in forest NPP in Hunan Province under three future scenarios: under the age-only scenario, NPP peaks in 2041 (133.56 Gg C yr^-1), while NPP peaks three years later in 2044 (141.14 Gg C yr^-1) under the natural development scenario. The maximum afforestation scenario exhibits the most rapid increase in NPP, with peaking in 2049 (197.95 Gg C yr^-1). However, with the aging of the forest, NPP is projected to then decrease by 7.54%, 6.07%, and 7.47% in 2060, and 20.05%, 19.74%, and 28.38% in 2100, respectively, compared to their peaks under the three scenarios. This indicates that forest NPP will continue to decline soon. Optimizing the age structure of forests through selective logging, afforestation and reforestation could mitigate this declining trend in forest NPP. Insights from the future multi-scenarios are expected to provide data to support sustainable forest management and national policy development, which will inform the achievement of carbon neutrality goals by 2060.

How to cite: Tian, L., Tao, Y., Li, M., and Simms, J.: How forest age impacts on net primary productivity: insights from future multi-scenarios, EGU General Assembly 2024, Vienna, Austria, 14–19 Apr 2024, EGU24-15159, https://doi.org/10.5194/egusphere-egu24-15159, 2024.

Tropical moist forests are one of the richest ecosystems on earth and provide multiple ecosystem services, but are also supporting many human activities and thus undergo continuous threats from logging, fire, shifting cultivation, road development or urban expansion. From this perspective, continuous monitoring and mapping those ecosystems keeps emphasize the importance of their preservation for the 21^st century.

Satellite remote sensing has proven to be essential to assess the deforestation in wide and remote areas at regional scale. Nevertheless, annual-based assessments using optical images tend to show inconsistency in tropical regions where the semi-persistent cloud coverage prevents steady periodic cloud-free acquisitions. The 8-year-old+ C-band SAR archive from Sentinel-1 provides now robust material to produce consistent yearly-based forest loss assessment.

Previous studies showed that a sudden decrease of the backscattered signal in an intact tropical forest canopy indicates a forest loss. However, SAR-based forest loss detection is sensitive to commission errors due to the speckle, showing their limitations as the thresholding of significant land cover change may include stable targets. Moreover, recent near-real-time detection systems focus on mapping forest loss alerts from a temporal early-warning perspective but are less reliable in providing accurate quantification of the degraded surfaces.

In this study, we introduce an annual index called MinB-Q10, computed pixel-based with a statistical probabilistic approach that combines VV and VH features. The result is represented as a chi-squared distribution based on Euclidean distance. The index is designed to highlight statistical deviation from stable forest distribution. The study has been developed in Democratic Republic of the Congo. It was calibrated in study area surrounding Yangambi (2.000 km²) and validated in a disconnected study area located around Kisangani (12.000 km²). The probabilistic approach and statistical considerations are developed to design an early-warning system as a second step, where the annual index would consolidate spatial assessment.

The preliminary results indicate a marked reduction of the commission errors compared with standard thresholding methods. Object-based accuracy assessment from optical independent imagery (in progress) enables to identify and distinguish the proportion of lost surface from the geometric accuracy of the detections. Moreover, combining pixel-counting with statistical estimates of the false detection rate generates unbiased prediction of the regional forest loss.

How to cite: Delhez, B., Radoux, J., Toussaint, F., Collet, T., and Defourny, P.: Consolidated Sentinel-1 SAR-based index provides robust annual forest loss assessment in tropical forests with semi-permanent cloud cover , EGU General Assembly 2024, Vienna, Austria, 14–19 Apr 2024, EGU24-1029, https://doi.org/10.5194/egusphere-egu24-1029, 2024.

Deforestation rates have been increasing in the Congo Basin in recent years, especially in Cameroon. To support actions to slow deforestation, Earth Observation (EO) has been used extensively to detect forest loss, but approaches to automatically identify specific drivers of deforestation in a level of detail that allows for intervention prioritisation have been rare. In this paper, we use deep learning to classify direct deforestation drivers in Cameroon and create a country-specific dataset for this task. We also compare the effectiveness of two types of freely available optical satellite imagery: Landsat-8 (pan-sharpened to a 15 m spatial resolution) and NCIFI PlanetScope (4.77 m spatial resolution). Our detailed classification strategy includes 15 direct deforestation drivers for forest loss events taking place between 2015 and 2020. We obtain an overall accuracy of 82% (F1-score: 0.82) with Landsat-8 data and an overall accuracy of 76% (F1-score: 0.76) for NICFI PlanetScope. Despite a coarser spatial resolution, Landsat-8 performs better than NICFI PlanetScope overall, including for small-scale drivers, although results vary by class. With Landsat-8, using only a single-image approach, we achieve an accuracy of at least 70% for all classes except for ‘Hunting’, ‘Oil palm plantation’, and ‘Fruit plantation’. These results show the potential of using this approach to monitor or analyse land-use changes leading to deforestation with more refined classes than before. In addition, our study demonstrates the potential of leveraging existing available datasets and straightforwardly adapting a generalised framework for other tropical locations with a relatively small amount of location-specific data.

How to cite: Debus, A., Beauchamp, E., and Lines, E. R.: Identifying direct deforestation drivers in Cameroon using deep learning and optical satellite data, EGU General Assembly 2024, Vienna, Austria, 14–19 Apr 2024, EGU24-2092, https://doi.org/10.5194/egusphere-egu24-2092, 2024.

Natural disturbances and post-disturbance recovery are principal drivers of forest ecosystem dynamics. While disturbances and their causes and consequences have received considerable attention from the scientific community in recent years, there is – however – a substantial lack of knowledge on post-disturbance recovery, despite its importance for forest resilience, carbon storage and developing effective conservation and management strategies. This is particularly pertinent in mountain landscapes, such as the Alps, where steep topography and frequent climate extremes could hamper natural tree regeneration but closed canopy forests are needed for protecting infrastructure from natural hazards. In our study, we aim to close this knowledge gap by the means of Earth Observation. Specifically, we mapped land cover fractions (treed vegetation, non-treed vegetation and bare soil) annually at 30 m spatial grain and over the period 1990-2021 across the Alps. To do so, we employed a temporally generalized regression-based spectral unmixing approach to dense time series of Landsat and Sentinel-2 data, including more than 73,000 individual scenes. From this dataset, we characterized post-disturbance recovery intervals, that is the time it takes to reach a similar canopy closure than pre-disturbance, across 1.76*10⁶ disturbance patches, including both natural and human disturbances. Results show that disturbed sites close their canopy on average after 10.6 years, with 60% of the disturbances reaching closed canopy after 10 years. We then compared recovery intervals derived from spectral unmixing to existing recovery indicators based on simple vegetation indices (NDVI, NBR), showing that those recovery indicators underestimate post-disturbance canopy closure time by a factor of 1.5 – 2. Finally, we tested whether post-disturbance bare soil fractions and disturbance characteristics (i.e., pre-disturbance tree cover and relative severity) can be used to predict long-term recovery success.  Results show that long-term recovery success (defined as canopy closure at 10 years post-disturbance) could be predicted with > 80% accuracy. From our results we conclude that (i) recovery indicators based on spectral indices are not well suited to characterize post-disturbance recovery in complex landscapes such as mountain forests and (ii) that disturbance characteristics and post-disturbance bare soil fractions are largely sufficient to predict whether a pixel will recover in the future or not. Our approach thus overcomes a major limitation of past remote sensing-based recovery assessments, which required long time series (>10 years) to assess recovery and thus were limited in understanding changes in post-disturbance forest recovery over time.

How to cite: Mandl, L., Viana-Soto, A., Stritih, A., Seidl, R., and Senf, C.: Benchmarking remote sensing-based forest recovery indicators for predicting long-term recovery success, EGU General Assembly 2024, Vienna, Austria, 14–19 Apr 2024, EGU24-2456, https://doi.org/10.5194/egusphere-egu24-2456, 2024.

Forests are one of the most active terrestrial ecosystems on Earth, especially intact forests, which possess high integrity and biodiversity of flora and fauna. Climate change and human activities are considered crucial environmental factors affecting forests. However, there remains significant uncertainty regarding the universal impact of these environmental factors on the multidimensional structure of global forests and how they will shape future forest structure and functionality. This study utilized a substantial amount of satellite data from the GEDI lidar satellite and multispectral MODIS satellite to quantify forest canopy structure density. It successfully delineated the spatial distribution patterns of the multidimensional canopy structure of global forests, focusing on analyzing the influence of human activity on the structural density of global forests, protected forests, and intact forests. Additionally, it analyzed the relative importance concerning human activities, climate, and other environmental factors. Based on this analysis, the study further investigated the differences in the spatial distribution patterns of multidimensional canopy structure in naturally regrowing forests and plantation forests under various human management types. It also implicated the impact of establishing protected areas and excluding human disturbances on forest ecosystem functioning like carbon storage. This research, from a satellite remote sensing perspective, notably revealed that human pressures extend even into forests traditionally believed to be protected and of higher integrity. It contributes to significant corrections regarding prevailing notions about the driving factors behind forest degradation. Moreover, it underscores the critical importance of better management and sustainable maintenance of protected forests, intact forests, and naturally regrowing forests for restoring and safeguarding forest ecosystem functioning.

How to cite: Li, W.: Remote sensing of multidimensional canopy structure of global forests in human-dominated landscapes, EGU General Assembly 2024, Vienna, Austria, 14–19 Apr 2024, EGU24-13970, https://doi.org/10.5194/egusphere-egu24-13970, 2024.

Forests are the most complex ecosystem on the planet and and play a crucial role in the exchange of mass and energy between the land surface and atmosphere. India's tropical region encompasses a diverse range of ecosystems, including deserts, mangroves, shrublands, deciduous, and evergreen forests, contributing to its ecological diversity. The varied canopy structural attributes of these vegetation types impact the exchange of carbon and water fluxes between the land surface and the atmosphere. We use high-resolution spaceborne data from the Global Ecosystem Dynamics Investigation (GEDI) to map the variability of vegetation canopy attributes at a synoptic scale. The total canopy height (RH100), foliage density (PAI) and foliage height diversity (FHD) demonstrated wide variability across the country and clearly differentiated the forested regions from the other land covers. The distribution of the canopy structural attributes in the forested regions and biomes also exhibited significant change across the years 2019-2021 with p-value < 0.05 (from the Kolmogorov–Smirnov test). Further, we fitted α and β (shape and scale) parameters of the Beta distribution function to the PAVD data from GEDI to compactly represent the spatial variation of the vertical variability of canopy. The high coefficient of determination (R² = 0.80) between the fitted beta distribution function and GEDI-PAVD suggests an effective representation of canopy vertical variability. Based upon the fitted parameters α and β, k-means clustering was performed which resulted in six distinct canopy structure classes. Differing canopy structures led to significant variations in radiation regimes throughout the day. Our observations suggest that incorporating Beta distribution-fitted shape and scale parameters into multi-layer canopy models enhances the estimation of flux exchanges over terrestrial ecosystems by capturing vertical variations in canopy structure.

Keywords: GEDI, Canopy structure, Plant area volume density, Beta distribution function

How to cite: Satapathy, T. and Dutta, D.: Characterising the Vertical Canopy Structure of Vegetation and its Impact on Radiation Regimes using Spaceborne LiDAR, EGU General Assembly 2024, Vienna, Austria, 14–19 Apr 2024, EGU24-5472, https://doi.org/10.5194/egusphere-egu24-5472, 2024.

Selected as European Space Agency’s seventh Earth Explorer in May 2013, the BIOMASS mission will provide crucial information about the state of our forests and how they are changing. This mission is being designed to provide, for the first time from space, P-band Synthetic Aperture Radar measurements to determine the amount of biomass and carbon stored in forests. The data will be used to further our knowledge of the role forests play in the carbon cycle.

In this context of an innovative sensor, the concept of Mission Algorithm and Analysis Platform dedicated to the BIOMASS, to the NASA-ISRO SAR (NISAR) mission  and to the NASA Global Ecosystem Dynamics Investigation (GEDI) mission mission is proposed. Developed in a collaborative way between ESA and NASA, this Mission Algorithm and Analysis Platform will implement, as part of the payload data ground segment, a virtual open and collaborative environment. The goal is to bring together data centre (Earth Observation and non- Earth Observation data), computing resources and hosted processing, collaborative tools (processing tools, data mining tools, user tools, …), concurrent design and test bench functions, accounting tools to manage resource utilisation, communication tools (social network) and documentation. This platform will give the opportunity, for the first time, to manage the community of users of the BIOMASS mission thanks to this innovative concept.

To best ensure that users can collaborate across the platform and to access needed resources, the MAAP requires all data, algorithms, and software to conform to open access and open-source policies. As an example of best collaborative and open-source practices, most of the BIOMASS Processing Suite (BPS) will be made openly available within the MAAP. This Processing Suite contains all elements to generate the BIOMASS upper-level data products and is currently in development under the umbrella of the open-source project called BioPAL. BioPAL is developed in a coherent manner, putting a modular architecture and reproducible software design in place. BioPAL aims to factorize the development and testing of common elements across different BIOMASS processors. The architecture of this scientific software makes lower-level bricks and functionalities available through a well-documented Application Programming Interface (API) to foster the reuse and continuous development of processing algorithms from the BIOMASS user community. This API will greatly simplify the use of the BIOMASS Processing Suite (BPS) on the MAAP.

In addition to open satellite data and open-source algorithms, open reference data is needed for Calibration and Validation. GEOTREES is composed of Biomass Reference Measurement sites that are in situ forest measurement sites with a common standard for high-quality data acquisition, transparent measurement protocols, long-term monitoring, and measurements traceable to SI units. GEO-TREES will be established through collaboration with existing international networks of high-quality forest plots that use standard forest monitoring protocols.

How to cite: Queune, T., Albinet, C., Dion, I., Lopes, C., Pinheiro, M., Rommen, B., and Scipal, K.: The BIOMASS Mission Algorithm & Analysis Platform (MAAP) and Related Open-Source Developments (BioPAL and GEOTREES), EGU General Assembly 2024, Vienna, Austria, 14–19 Apr 2024, EGU24-20778, https://doi.org/10.5194/egusphere-egu24-20778, 2024.

Building on the positive experiences with open forest map data in Scandinavia, it is evident that extending a similar solution globally has the potential to revolutionize forest management and business on a worldwide scale. While forest management in the Nordic countries can certainly be enhanced, the most rapid solution for climate change mitigation involves providing other nations with opportunities akin to those that have benefited the forestry sector in Sweden during the initial stages of digitalization.

In the proposed project, we aim to create a novel hierarchical decision-making system for efficient forest mapping, leveraging a diverse range of remote sensing data sources with varying resolutions. This hierarchical system will be developed using state-of-the-art AI methods, complemented by results from traditional computer vision techniques such as texture analysis, saliency, and probabilistic object representation. A significant strength of the project lies in using the forest data and maps of Sweden and Finland as test beds to benchmark the methodology developed.

We are confident that this project will make substantial contributions to climate change mitigation, biodiversity enhancement, and other societal values. Moreover, it aims to foster the creation of new business models by developing an innovative methodology for the next generation of forest maps. Our vision is to adapt the success story of open forest map data from the Nordic region globally, harnessing the power of advanced AI technology and integrated use of remote sensing and field data.

How to cite: Fransson, J. E. S., Soomro, S., Holmström, A., Nilsson, M., Salo, J., Santoro, M., Sertel, E., Wallerman, J., Ünsalan, C., and Zariņš, J.: ForestMap: The next generation of forest maps - adapting a Nordic success story, EGU General Assembly 2024, Vienna, Austria, 14–19 Apr 2024, EGU24-22372, https://doi.org/10.5194/egusphere-egu24-22372, 2024.

Vegetation phenology is closely linked to the functioning of multiple aspects of forest ecosystems and is regulated by a complex interaction between climatic and environmental factors. In particular, the end of the growing season has proven to be very sensitive to extreme weather events, leading to alterations in the regular physiological behaviour of forests. Autumn phenology represents a little-explored season due to the highly variable response of forests to environmental factors. This work aims to investigate late-season dynamics by comparing ground-based and satellite observations in European beech forests. The objectives of this research are: (i) quantify the temporal discrepancy between phenology obtained from ground-based observations (PEP725 stations) and satellite-derived data (MODIS EVI time series); (ii) assess the influence of the main biophysical factors, i.e. latitude, elevation, total annual precipitation and mean annual temperature, on the mismatch. The results identified key end-of-season metrics, distinguishing different stages during the season that were affected differently by biophysical factors, such as temperature and precipitations. This study highlights the complexity of late-season phenology, emphasizing the crucial role of remotely sensed phenometric analysis compared to ground-based observations, revealing a fundamental contribution to understanding of late-season phenology in the context of climate change.

How to cite: Cesaretti, L., Ferrara, C., Corona, P., and Bajocco, S.: Exploring the autumn phenology of European beech forests: a comparative analysis of ground-based and satellite observations, EGU General Assembly 2024, Vienna, Austria, 14–19 Apr 2024, EGU24-10217, https://doi.org/10.5194/egusphere-egu24-10217, 2024.

Accurate mapping and monitoring of forest tree species are crucial for understanding ecosystem dynamics [1], assessing biodiversity [2], and enabling sustainable forest management [3]. Tree species adapt their morphology and phenology to the environment [4], leading to variability in spectral signatures across geographic regions. Furthermore, the spectral reflectance of a given tree species varies significantly with growth stages and seasons [5], making the classification based solely on RGB data extremely challenging. At the local level, spectral variability also closely correlates with stand structure factors such as crown size, stand density, and gap sizes. This results in varying signal reflectance from different parts of the same crown, further complicating tree species classification [6]. Thus, we proposed a spectral-spatial-temporal constrained deep learning method, an end-to-end multi-head attention-based network, to automatically extract deep features for tree species mapping. Employing this model on multi-temporal hyperspectral imagery from the DLR Earth Sensing Imaging Spectrometer (DESIS), we produced a 30 m resolution forest species distribution map of the Harz Forest in Germany. DESIS, a VNIR sensor aboard the International Space Station, captures detailed Earth images upon request, offering extensive spectral data across 235 bands ranging from 400 to 1000 nm [7]. Our methodology leverages the comprehensive spectral information provided by DESIS, enhancing the tree species mapping accuracy. Utilizing the reference data from TreeSatAI Benchmark Archive [8], we prepared 134,886 hyperspectral data patches, each labelled with tree species information. The evaluation involved assessing the F1-score, Jaccard index, Hamming loss, and accuracy for various tree species using National Forest Inventory (NFI) data plots. The results reveal the potential of deep learning using hyperspectral data in the precise and automated mapping of forest tree species distribution, thereby supporting evidence-based decision-making in sustainable forest management.

[1] Welle, Torsten, et al. "Mapping dominant tree species of German forests." Remote Sensing 14.14 (2022): 3330.

[2] Grabska, Ewa, David Frantz, and Katarzyna Ostapowicz. "Evaluation of machine learning algorithms for forest stand species mapping using Sentinel-2 imagery and environmental data in the Polish Carpathians." Remote Sensing of Environment 251 (2020): 112103.

[3] Xie, Bo, et al. "Analysis of regional distribution of tree species using multi-seasonal sentinel-1&2 imagery within google earth engine." Forests 12.5 (2021): 565.

[4] Chuine, Isabelle. "Why does phenology drive species distribution?." Philosophical Transactions of the Royal Society B: Biological Sciences 365.1555 (2010): 3149-3160.

[5] Hesketh, Michael, and G. Arturo Sánchez-Azofeifa. "The effect of seasonal spectral variation on species classification in the Panamanian tropical forest." Remote Sensing of Environment 118 (2012): 73-82.

[6] Ferreira, Matheus Pinheiro, et al. "Tree species classification in tropical forests using visible to shortwave infrared WorldView-3 images and texture analysis." ISPRS journal of photogrammetry and remote sensing 149 (2019): 119-131.

[7] de los Reyes, Raquel, et al. "The Desis L2a Processor And Validation Of L2a Products Using Aeronet And Radcalnet Data." The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences 46 (2022): 9-12.

[8] Ahlswede, Steve, et al. "TreeSatAI Benchmark Archive: A multi-sensor, multi-label dataset for tree species classification in remote sensing." Earth System Science Data Discussions 2022 (2022): 1-22.

How to cite: Mu, Y., Shahzad, M., and Zhu, X. X.: A spectral-spatial-temporal attention network for tree species mapping using DESIS hyperspectral imagery, EGU General Assembly 2024, Vienna, Austria, 14–19 Apr 2024, EGU24-10854, https://doi.org/10.5194/egusphere-egu24-10854, 2024.

Forest ecosystem conservation is of crucial importance for preserving biodiversity, regulating climate patterns, and providing ecosystem services essential for human well-being. The Ticino Park temperate mixed forest represents the last remaining natural ecosystem of the Po Valley region. Recognized as a UNESCO-MAB Biosphere Reserve, this precious ecosystem has been increasingly affected by natural and human-induced disturbances, including severe drought, exacerbated by climate change.

Remote sensing has proven to be a cost-effective tool for the indirect estimation and mapping of forest characteristics and conditions at different spatial and temporal scales. This study aims to develop a better understanding of the relationship between drought-induced forest stress, spectral changes observed from Sentinel-2 satellite data, and how these relate to functional traits and species composition. We believe this will help to identify spectral indicators and metrics for the early detection of drought-induced forest mortality.

In summer 2022, an intensive field campaign was carried out in the Ticino Park Forest. First, in June, data on functional traits, specifically Leaf Area Index (LAI), Leaf Chlorophyll Content (LCC) and Leaf water content (LWC), were collected within 31 homogeneous 30x30 mforest stands. Secondly, in September, a subset of 19 of the 31 stands initially sampled were revisited (for a total of 52 sampling stations). In addition, vascular plant species composition was analysed in 64 selected stands to define the different vegetation associations and calculate the corresponding Ellenberg indexes in order to ecologically characterise the sites. Meanwhile, the standardized precipitation-evapotranspiration index (SPEI) from 2017 to 2023 was calculated to assess the severity and duration of drought events in the Ticino Park area.

Concerning the remote sensing analysis, the time series of cloud-free Sentinel-2 images collected over the Ticino Park from 2017 to 2023 were processed to compute LAI, Canopy Chlorophyll content (CCC = LCC x LAI) and Canopy water content (CWC = LWC x LAI) maps of each image through the Sentinel Application Platform (SNAP) biophysical processor tool. LAI, CCC and CWC maps were validated using LAI, CCC and CWC field measurements. These plant functional trait time series were used to quantify the deviation of LAI, CCC and CWC at a precise location and time from the 2017-2023 multi-year daily averages, thus obtaining the standard anomalies. Generalized additive models (GAMs) were then applied to examine the correlation between functional trait anomalies and a series of factors expected to influence the response of plant traits to water stress, such as SPEI value, vegetation association, and other environmental characteristics.

This study is a first attempt to analyse by remote sensing-based approaches the vegetation response to extreme weather conditions, accounting for differences in local climatic conditions, species ecology, and environmental variables.

How to cite: Savinelli, B., Panigada, C., Tagliabue, G., Vignali, L., Gentili, R., Padoa-Schioppa, E., Fassnacht, F. E., and Rossini, M.: Plant trait time series from Sentinel-2 to detect drought stress in a mid-latitude forest ecosystem, EGU General Assembly 2024, Vienna, Austria, 14–19 Apr 2024, EGU24-17857, https://doi.org/10.5194/egusphere-egu24-17857, 2024.

Climate change disrupts ecosystems and increases extreme weather events. Modeling ecosystem functioning is crucial for effective adaptation strategies. This study focuses on quantifiable vegetation properties, including leaf area index (LAI), chlorophyll content (CHL), leaf mass per area (LMA), and equivalent water thickness (EWT). One of the most effective ways to retrieve plant biophysical and structural properties at a large scale is by using multispectral satellite images. However, accurately quantifying plant biophysical variables from such datasets presents several challenges, including understanding the influence of the vertical distribution of such traits within the canopy on the corresponding signal; the impact of sun-view geometry during image acquisition, the use of genotype-specific relationship or instead the use of genotype biophysical traits in the model. This study estimates biophysical variables from a large measurement dataset obtained on forest plantations in Sao Paulo, Brazil, and analyzes the effect of vertical heterogeneity, solar and viewing geometry, and associated biophysical properties.

The dataset comprises in-situ and remote sensing data. In-situ measurements of LAI, CHL, LMA, and EWT were collected from 2019 to 2021 on 25 eucalypt genotypes. Biomass measurements were conducted in 1323 trees, and CHL, LMA, and EWT were measured per vertical third of the canopy. To evaluate the influence of the vertical heterogeneity, we defined a weighted expression of the top and middle thirds of the canopy to average these biophysical variables and relate them to the remote sensing data, which includes Sentinel-2 images acquired from 2019 to 2021, at dates close to the field measurements.  First, 22 vegetation indices (VIs) were used to build regression models, each regressing the weighted target variable to a VI. After determining the optimal weight, we tested the accuracy of the linear models by accounting for sun-sensor geometry and vegetation traits. Both 10-fold cross-validation and an independent test dataset were used to assess model performance together with root mean square (RMSE) and coefficient of determination (R²).

Results show that most models using 70% top and 30% medium canopy produced the best performances in estimating CHL. For both LMA and EWT, the optimal percentage was 50%-50%. These outcomes indicate that shaded parts of the canopy play a significant role in the above-canopy reflectance, especially for LMA and EWT, which are particularly sensitive to spectral domains ranging from 1700 to 2400 nm, which has a higher transmittance rate towards the canopy. Concerning the inclusion of sun-sensor geometry, most models, generated with different VIs, benefitted from these variables to predict the target variable, resulting in lower RMSEs. The use of several canopy traits in the model reduced the error but would require to have previous knowledge of them. These empirical results underline the influence of sensor geometry and other biophysical properties on the prediction of LAI, CHL, EWT, and LMA. We believe that these results advocate for further investigation using radiative transfer model inversion.

How to cite: Barbosa Ferreira, V., le Maire, G., and Féret, J.-B.: Biophysical variables estimation from Sentinel 2 images: emphasizing the importance of vertical heterogeneity, solar and viewing geometry, and associated biophysical properties, EGU General Assembly 2024, Vienna, Austria, 14–19 Apr 2024, EGU24-10167, https://doi.org/10.5194/egusphere-egu24-10167, 2024.

Many LiDAR remote sensing studies over the past decades declared data fusion as a potential avenue to increase accuracy, spatial and temporal resolution of the final datasets. LiDAR data fused with other datasets such as multispectral, hyperspectral and radar has proven beneficial for various applications, including the segmentation processes, the above ground biomass (AGB) assessments, the tree height estimation and the tree species identification. Despite progress in data fusion techniques and opportunities, the proliferation of scientific papers has given rise to questions within the scientific community. What is ‘data fusion’ and how should this term be used in our community? What opportunities does it provide and are these approaches as good as promised? What are the main challenges in LiDAR data fusion for forest observations? In this paper, we performed a structured literature review to analyse relevant studies on these topics published in the last decade (2014-2023). We used a specific query in the Web of Science database, selecting only papers published in English language with a publication status of “article” or “review article”. These limitations in the query resulted in 407 papers. The abstract were screened by two independent reviewers, following these criteria: (1) The paper must assess some aspect of trees/forests relevant to forestry applications, with the exclusion of those solely focusing on crops or human-made structures (such as infrastructure or buildings). (2) The fusion process must include LiDAR data. A significant portion of excluded papers did not actively engage in data fusion; instead, they merely discussed it as a potential solution to identified limitations in their analyses. Alternatively, these papers did not use data from a LiDAR sensor in their application. The screening process resulted in 153 papers. From our findings, there is a slight general upward publication trend over the last 10 years, with an increasing trend in the use of spaceborne LiDAR sensors. The predominant form of fusion observed in this study was airborne LiDAR with other airborne data types, accounting for 45.4% of the total papers. Following closely was the fusion of airborne LiDAR data and spaceborne devices, constituting 29.8%. Equally represented, each with 11.3%, were spaceborne LiDAR-data with other spaceborne sensors and airborne-terrestrial fusion. The least commonly encountered method was the fusion of terrestrial LiDAR with other data from terrestrial platforms, representing only 2.1%. 27.2% of the papers are dealing with an individual tree detection approach, 49.7% with an area-based approach and 17.2% with both. 6% are reviews with no defined study area/application. Our review indicated that, generally, all common applications are improved using data fusion. The benefits include improved accuracy, particularly noticeable in tree species composition classifications, and advancements in spatial or temporal resolution, especially for canopy height assessments. However, a critical consideration arises regarding whether the incremental improvements, at times marginal, justify the additional economical and computational investment.

This abstract is based upon work from COST Action 3DForEcoTech, CA20118, supported by COST (European Cooperation in Science and Technology).

How to cite: Balestra, M., Marselis, S., and Mokroš, M.: Unveiling the Canopy: Insights, Definitions, and Outcomes from a LiDAR Data Fusion Review for Forest Observation, EGU General Assembly 2024, Vienna, Austria, 14–19 Apr 2024, EGU24-19325, https://doi.org/10.5194/egusphere-egu24-19325, 2024.

An animal’s survival depends on its ability to successfully navigate a dynamic resource landscape that varies in space and time. To study animal cognition in ecologically-relevant scales and settings, there is a need for reliable and efficient measures of nutritional resource distribution and quality. Hyperspectral imagery leverages the differential surface reflectance to estimate the relative chemical composition of a pixel, and may therefore enable remote sensing the distribution of nutrients at the landscape-scale. To explore the potential of this method in wild settings, we used airborne hyperspectral imagery with ground-based field spectroscopy and high-throughput wet chemistry data to predict nutrients present across an apple orchard landscape in Ravensburg, Germany. In this pilot study, we collected data on 24 apple trees over a four week period preceding harvest. We used spectral samples taken on the ground with a field spectrometer to create a spectral library of leaf and fruit samples. Simultaneously, we flew a hyperspectral drone (Headwall CoAlign) to collect hyperspectral voxels that were then spectrally unmixed to determine the endmember abundance in each pixel. After predicting the presence of fruit in a pixel, we then used the relationship between fruit reflectance and the sugar content to predict the amount of sugar available within a pixel. Our results indicate that apple sugar content is correlated with lower reflectance of the fruit in the near infrared. We are able to predict fruit presence on a pixel basis with 85% accuracy, and to predict sugar content using individual fruit reflectance with 80% percent accuracy. From this information, we can then extrapolate to create a prediction of nutritional elements on a landscape. Our approach demonstrates strong potential for use as a means of remotely sampling the nutritional landscapes in which wild animals live. This will open exciting opportunities for ecological studies at the landscape scale, including animal behavior researchers and movement ecologists to test detailed hypotheses related to animal movement and decision-making.

How to cite: Tiedeman, K., Núñez, C. L., Alavi, S., Schuerkmann, A., and Crofoot, M.: Remotely sensing fruit presence and nutritional content in complex landscapes, EGU General Assembly 2024, Vienna, Austria, 14–19 Apr 2024, EGU24-21242, https://doi.org/10.5194/egusphere-egu24-21242, 2024.

In the realm of environmental monitoring and urban planning, the accurate assessment of tree parameters from 3D point clouds is essential for effective resource management and decision-making. This paper introduces a versatile and user-friendly approach designed to streamline the extraction of key tree parameters from 3D point clouds.

The related software employs advanced point cloud processing algorithms to identify and analyze individual trees within a point cloud dataset, acquired through terrestrial laser scanners (TLS), facilitating the extraction of crucial parameters such as tree height, crown diameter, and trunk diameter at breast height (DBH). Leveraging state-of-the-art computer vision techniques, this approach ensures high precision and efficiency in tree parameter extraction, even in complex and densely vegetated environments.

Key features include an intuitive graphical user interface, allowing users to interactively visualize and validate the extracted tree parameters. The ability to adjust the various tree extraction and segmentation settings at any time gives the user complete freedom to modify their analysis to their needs.

The results showcase the software's ability to accurately and efficiently extract tree parameters, making it a valuable tool for researchers, urban planners, and environmental professionals engaged in forestry management, green infrastructure planning, and ecological monitoring.

A reliable database and a procedure suitable for everyday use are enormously important to ensure highly accurate monitoring and to cope with the ever faster changing conditions and their effects on vegetation. Therefore an additional approach to use a terrestrial laser scanner in kinematic mode is presented, which allows the generation of a comprehensive point cloud in a short time.

Overall, the software LIS TreeAnalyzer, a plugin to RIEGL’s RiSCAN PRO, contributes to the advancement of 3D point cloud analysis by providing a robust solution for the extraction of tree parameters, ultimately supporting sustainable urban development and informed decision-making in the field of environmental science and resource management.

How to cite: Groiss, B. and Handl, M.: Efficient Extraction of Tree Parameters from 3D Point Clouds, EGU General Assembly 2024, Vienna, Austria, 14–19 Apr 2024, EGU24-9834, https://doi.org/10.5194/egusphere-egu24-9834, 2024.

Clumping describes the heterogeneity of forest structure - the spatial arrangement of foliage elements, such as leaves or needles, within a vegetation canopy. Clumping information is essential for assessing radiation transfer through canopies, photosynthesis, and hydrological processes. The challenge in conifer stands arises from the difficulty of measuring gaps between needles within a shoot using traditional optical instruments, or LiDAR. Previous methods for estimating needle-to-shoot-area ratio were often destructive and labor-intensive. In this study, we introduce a highly efficient technique—blue light 3D photogrammetry scanning—to comprehensively characterize the structure of conifer shoots and determine shoot-level clumping. This approach significantly reduces the labor intensity associated with previous methods. To validate our technique, we compared it to the established photographic/volume displacement method for quantifying shoot-level clumping. Here, we present 3D shoot models, shoot-level clumping values, and their seasonal variations for a wide range of European native conifer species.

The demonstrated effectiveness and performance of the blue light 3D photogrammetry scanning method offer the potential for more frequent and accurate measurements of 3D shoot structures. This advancement opens doors to further improvements in measuring and upscaling optical properties for coniferous canopies. In future research, the enhanced understanding of needle shoots, a fundamental yet often overlooked aspect of foliage clumping in canopies, will significantly improve 3D radiative transfer modeling for coniferous forests.

How to cite: Pisek, J., Borysenko, O., and Kuusk, A.: Estimation of coniferous shoot structures by high precision blue light 3D photogrammetry scanning, EGU General Assembly 2024, Vienna, Austria, 14–19 Apr 2024, EGU24-4548, https://doi.org/10.5194/egusphere-egu24-4548, 2024.

By simulating laser scanning of dynamic tree scenes, we investigate how tree movement during point cloud acquisition affects the accuracy of a range of tree metrics.

Terrestrial laser scanning (TLS) has proven to be an effective surveying method for forestry and ecology, producing highly detailed 3D point clouds of trees. From these point clouds, a variety of metrics can be derived, such as tree and crown dimensions, stem diameter and taper, foliage parameters, and woody volume. In this way, TLS supports traditional forest inventory and monitoring, and provides valuable in-situ data for the calibration of remote sensing approaches.

Typically, TLS point clouds are acquired from multiple scan positions to increase coverage and minimise occlusion. Scans from these positions are then co-registered and merged into a single point cloud. If wind is blowing during data acquisition and branches and leaves are moving, the merged point clouds may show multiple or blurred representations of branches and leaves. This is likely to affect the quality of the tree information derived from the point clouds. Although this problem is well known, few studies have systematically investigated the effect of vegetation movement during the scanning process on the derived tree metrics.

The aim of this work is to quantify the errors induced by vegetation movement during TLS acquisition on a variety of metrics. We also investigate the extent to which point cloud filtering methods and the omission of 'problematic' scan positions can improve metric accuracies.

To enable a systematic and controlled investigation, we use virtual laser scanning (VLS) with the open-source laser scanning simulator HELIOS++ [1, 2]. We first generate synthetic 3D tree models using procedural modelling [3, 4]. These tree models are then animated in different scenarios by simulating different wind conditions. For each wind scenario, the trees are virtually scanned from multiple positions, each scan being performed at a randomly sampled frame of the animation. From the simulated multi-scan TLS point clouds, we estimate several point cloud metrics, both with and without prior point cloud filtering. We compare the metrics with metrics derived from the reference meshes or point clouds.

Performing such an analysis in a simulation environment has several major strengths: a) we can isolate the wind effects from other errors such as co-registration errors, b) we can define arbitrary custom wind scenarios and do not need to carry out real wind measurements, and c) reference data is available in the form of the input 3D tree models and base simulations without wind.

We demonstrate how VLS can be used to investigate wind effects, which are a common source of error and uncertainty in TLS of vegetation. This not only allows us to develop strategies to account for these effects, but also informs us of the importance of modelling these effects when using VLS in other contexts, such as algorithm development or machine learning.

REFERENCES

[1] HELIOS++: https://github.com/3dgeo-heidelberg/helios

[2] Winiwarter, L., et al. (2022): DOI: https://doi.org/10.1016/j.rse.2021.112772

[3] Sapling Tree Gen: https://docs.blender.org/manual/en/latest/addons/add_curve/sapling.html

[4] Weber, J. & Penn, J. (1995): DOI: https://doi.org/10.1145/218380.218427

How to cite: Weiser, H., Esmorís Pena, A. M., and Höfle, B.: How Tree Movement Influences Tree Metrics Derived from Laser Scanning Point Clouds, EGU General Assembly 2024, Vienna, Austria, 14–19 Apr 2024, EGU24-1633, https://doi.org/10.5194/egusphere-egu24-1633, 2024.