Satellite remote sensing with the third-generation geostationary satellites has recently gained significant attention. These satellites provide hyper-temporal datasets, enabling the mitigation of cloud contamination effects and the observation of diurnal changes in targets, in contrast to polar-orbiting satellites. Currently, several third-generation geostationary satellites, such as the Japanese Himawari series, the US GOES series, the Korean GK2A, the Chinese FY4 series, and the European MTG1, have been under operation, and their datasets are available. In contrast with polar-orbiting satellites, geostationary satellites cover the limited areas of the Earth due to orbital constraints. Consequently, collaboration among multiple geostationary satellites is required to cover the Earth comprehensively. For this collaboration involving satellites with different sensors, spectral band adjustments among sensors and subsequent data fusion based on these adjustments are indispensable. In this study, spectral band adjustments and data fusion were performed using band simulations with hyperspectral data from satellites, in situ observations, and a 3-D radiative transfer model. The spectral band adjustments and data fusion focused on the visible and near-infrared regions, which are critical for terrestrial ecosystem monitoring, including vegetation monitoring. Our simulations revealed linear relationships in the visible and near-infrared regions among the bands of each sensor after specific mathematical processes. Additionally, experimental data fusion using actual geostationary satellite datasets demonstrated the success of our spectral band adjustment approach. These results suggest that the proposed method can significantly contribute to environmental monitoring with third-generation geostationary satellites, mainly observation of the terrestrial ecosystem. Further research, such as applications for terrestrial vegetation monitoring, is anticipated.

How to cite: Sasagawa, T., Ichii, K., Yamamoto, Y., Miura, T., Yang, W., Matsuoka, M., Yoshioka, H., Wang, W., Hashimoto, H., and Nasahara, K.: Spectral Band Adjustment and Data Fusion of Multiple Third-Generation Geostationary Satellites: Toward Hyper-Temporal Monitoring of the Biosphere, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-14924, https://doi.org/10.5194/egusphere-egu25-14924, 2025.

Hot deserts cover between 19 and 25 million square kilometers of the Earth land surface. Climate change together with impacts from anthropogenic activities which intensify with the growing world population are claimed to cause desertification. One interpretation of desertification processes concern shifting of the deserts’ boundaries into semi-arid zones. However, “ the delimitation of desert areas is difficult, particularly the location of the outer boundaries.... desert boundaries are often considered as shifting zones of transition rather than lines clearly demarcated by climate or by abrupt changes in species  or associations. Transitional boundaries may result from human impact or from decadal climatic fluctuations.” (Laity, 2008). In broad terms, desert boundary bounds  terrain which may not sustain natural vegetation growth. When considering herbaceous plants, such bound shifts locally in response to small amounts of rainfall. Shrubs’ extents are bounded by sub-surface water accumulation which is affected by hydrology, topography and yearly rainfall fluctuations. Drought tolerant  dwarf-shrubs may spread quickly into bare soils during rainy years and sustain there during long draughts. These growth forms’ and their  mixed patterns responses to periodic, seasonal and annual precipitation fluctuations are thus complex and highly vary in space and time. Adding to this complexity anthropogenic impacts, such as from grazing, wood cutting and fire, make the search for desert boundaries a challenging  task. Remote sensing offers  data and tools for  the search for such boundaries across wide regions. Landsat TM is instrumental for this purpose with its continuous coverage since 1985 at moderate resolution.  Seasonal / phenological vegetation cover fractions or NDVI allow  for differentiating between growth-form patterns and their transitions. The following four conceptual methods were developed utilizing multi-date Landsat TM imagery for discovering such transition zones across  desert fringes:

• Extreme rainfall conditions:
  • Total (all growth forms) green cover at extremely high winter rainfall year: reveals the boundary between the maximal extents of vegetation and the areas which has very low rainfall  or cannot support vegetation growth.
  • Shrubs cover at years of extremely low rainfall: reveals the boundary of sustainable and resilient shrubs.
• Vegetation drying rates at the beginning of the dry season (spring) indicate differences in soil chemical and physical properties . Transitions from clay soils and those of high organic matter to lithosols and rocky surfaces can be clearly detected.
• Winter trends of green vegetation change as a function of the  progression of rainfall accumulation may reveal transition from phrygana to Mediterranean shrublands. 
• Imagery spatial erosion and dilation of patchy vegetation patterns may allow differentiation  between  areas according to their draught  recovery potential: high for dense and large patches and low for small and sparse shrubs.

The conference presentation will demonstrate the results of applying these methods for years of average, high and low rainfall across  a Mediterranean to arid gradient  in the south-eastern side of the Mediterranean basin

How to cite: Shoshany, M.: Remote sensing of desert boundaries: Challenges, Concepts and Methods, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-1884, https://doi.org/10.5194/egusphere-egu25-1884, 2025.

Water-use strategies in arid regions operate across multiple scales, from individual plants to entire ecosystems, shaped by responses to drought. This Perspective contrasts isohydric and anisohydric strategies and introduces the ecological concept for the "long-distance interactions" and "structural overshoot," where ecosystems exceed their water-use capacity during prolonged droughts. We propose scalable vegetation cover indicators, such as fractional vegetation and biocrust cover, to monitor these dynamics. We also discuss the potential applications of plant water sources, biomass allocation, and functional traits in understanding arid ecosystems. By integrating remote sensing technologies with these indicators, we emphasize the need for advanced drought monitoring tools to enhance plant resilience, optimize water resource management, and improve our understanding of adaptive strategies in arid landscapes.

How to cite: xu, M.: Remote sensing for monitoring plant Water-Use Strategies across scales in Arid Ecosystems, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-19692, https://doi.org/10.5194/egusphere-egu25-19692, 2025.

The Arctic tundra vegetation is going through major changes as global warming alters atmospheric functions and weather patterns. These changes have been shown to affect for instance phenological patterns, plant community structures, herbivory patterns as well as carbon storage in biomass. Extensive remote sensing research with multispectral sensors has revealed significant greening trends and events as well as shrub expansion, also known as shrubification, across the Arctic. These trends have been hypothesized to counteract increases in carbon content in the atmosphere. However, the magnitude of this effect as well as the shrub expansion rates are still unanswered due to low data availability as well as topographic and phenological differences across the region. This research was conducted on the Yamal Peninsula in Russian Arctic, where, in addition to climate change, vegetation is strongly influenced by the reindeer grazing practiced by the indigenous Nenets reindeer herders, as well as the expanding gas and oil drilling activities, which are accompanied by extensive infrastructure development. In this study our aim is to assess the opportunities of multispectral remote sensing data with varying spatial and temporal resolutions to examine shrubification in ecologically complex Arctic landscapes. Our research questions are the following: 1) Do Landsat-derived vegetation indices from a 30-year timespan show significant amount of greening in Arctic Russia; 2) How does image availability and phenology affect the way greening trends are analyzed; 3) Has shrub height and area increased during the study period and what implications does reindeer grazing have for shrub expansion and plant community structures; 4) Are greening trends associated with increased shrub height and area.

Methodologically, we first extracted several vegetation indices from Landsat-satellite collections to evaluate greening trends. After satellite sensor cross-calibration with Random Forests, we examined how phenology and imaging frequency affects these trends and the analysis. We then compared the results with high-resolution QuickBird-2 and WorldView-2/3 imagery from 2004, 2013, 2017 and 2023. Secondly, we utilized drone imagery and VHR images to upscale vegetation height field data collected in 2017, and to delineate shrub areas with GeoSAM AI algorithm. In the last part, we created a classification with machine learning to estimate shrub expansion and height as well as change in community structure. Our preliminary results suggest that Landsat maximum vegetation indices have increased slightly across the entire study area. However, we also found a connection between image availability and the amount of greening detected. In addition, we found that shrub area and height has increased during the study period which could potentially benefit herbivore grazing activity. We therefore suggest coupling plant community changes with herbivore dynamics in the future studies on shrubification in the Arctic tundra.

How to cite: Koivisto, E., Kuzmin, A., Berner, L., Forbes, B. C., Kerby, J., Kolari, T., Korpelainen, P., Skarin, A., Tahvanainen, T., Verdonen, M., Villoslada, M., and Kumpula, T.: Understanding Arctic Greening Trends: A Multispectral Approach to Shrubification and Ecological Shifts, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-9963, https://doi.org/10.5194/egusphere-egu25-9963, 2025.

Due to accelerating global warming [1], the polar regions, and in particular the Arctic, are subject to many changes and cascading effects [2]. Northern lands are facing a generalized rise in soil temperature causing changes in the surface cover [3], the hydrological and mechanical state of the subsoil including permafrost thaw [4], [5] and potentially triggering massive release of greenhouse gases [6]. As land cover is a key control parameter for permafrost state, the survey of surface changes is of great importance. Consequently, monitoring the evolution of surface boreal ecosystems over large time scales, satellite imagery combined with reliable and proven methodologies is crucial for understanding the impact of climate change on polar continental regions. In this study, we use a Random Forest algorithm to analyze satellite images from the Copernicus (ESA) Sentinel-1 and Sentinel-2 programs in combination with ground truth data collected in July 2024, to monitor changes in the surface ecosystem over a 480 km² area in the Abisko region (Arctic Sweden). Random Forest method applied to features derived from satellite images allows the production of reliable land cover maps (>87% accuracy). Our results demonstrate that radar imagery is a vital source of information for overcoming the inherent limitations of optical imagery caused by frequent and dense cloud cover, particularly in summer, when average monthly cloud cover can reach up to 85% [7]. Additionally, they highlight that combining optical and radar imageries with a robust machine learning approach enables the production of high-quality land cover maps, providing significant added value for long term and high temporal resolution monitoring of land cover changes in northern continental regions.

[1]           P. M. Forster et al, ‘Indicators of Global Climate Change 2023: annual update of key indicators of the state of the climate system and human influence’, 2024

[2]           Intergovernmental Panel on Climate Change (IPCC), The Ocean and Cryosphere in a Changing Climate: Special Report of the Intergovernmental Panel on Climate Change, 2022

[3]           M. Wenzl et al, ‘Vegetation Changes in the Arctic: A Review of Earth Observation Applications’, 2024

[4]           E. J. Burke et al, ‘Evaluating permafrost physics in the Coupled Model Intercomparison Project 6 (CMIP6) models and their sensitivity to climate change’, 2020

[5]           T. Xavier et al, ‘Future permafrost degradation under climate change in a headwater catchment of central Siberia: quantitative assessment with a mechanistic modelling approach’, 2024

[6]           R. M. Varney et al, ‘Evaluation of soil carbon simulation in CMIP6 Earth system models’, 2022

[7]           J. E. Kay et al, ‘Recent Advances in Arctic Cloud and Climate Research’, 2016

How to cite: Carry, R., Auda, Y., Remy, D., Gustafsson, J., Pokrovski, O., Lundin, E., Bouvet, A., and Orgogozo, L.: Mapping the land cover of a Northern Sweden watershed using Sentinel-1 & 2 data and an optimized Random Forest, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-6343, https://doi.org/10.5194/egusphere-egu25-6343, 2025.

Forest gains and losses may have unequal effects on forest resilience, particularly given their distinct temporal dynamics. Here, we quantify the sensitivities of boreal forest resilience to forest cover gain and loss using a resilience indicator derived from the temporal autocorrelation (TAC) of the kernel Normalized Difference Vegetation Index (kNDVI) from 2000 to 2020. Our findings unveil pronounced asymmetric sensitivities, with stronger sensitivity to forest loss (-4.26 ± 0.14 × 10^-3; TAC increase per 1% forest cover loss) than to forest gain (-1.65 ± 0.12 × 10^-3; TAC decrease per 1% forest cover gain). Locally, approximately 73% of the boreal forest exhibits negative sensitivity, indicating enhanced resilience with forest cover gain and vice versa, especially in intact forests compared to managed ones. This sensitivity is affected by various trajectories in forest cover change, stemming primarily from temporal asynchrony in the recovery rates of various ecosystem functions. The observed asymmetry underscores the importance of prioritizing forest conservation over reactive management strategies following losses, ultimately contributing to more sustainable forest management practices.

How to cite: Liu, X.: Asymmetric sensitivity of boreal forest resilience to forest gain and loss, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-4801, https://doi.org/10.5194/egusphere-egu25-4801, 2025.

The Observing System Simulation Experiments are simulation tools that support the development, benchmarking, and evaluation of remote sensing missions and methods. The study of biodiversity, from remote sensing, and particularly that of plant diversity,  is an emerging topic with potentially beneficial outcomes for global-scale monitoring and conservation. However, advances in this field are limited by insufficient systematic, standardized, and global field datasets that enable a comprehensive linkage of the plant diversity aspects with the variability of the spectral signals captured by remote sensing instruments. This typically makes the findings site- and method-dependent, preventing the validation of new methods in a sufficiently wide range of ecosystems, seasons, and vegetation types. Pioneering modelling works have proven as valuable tools to answer methodological questions, screen potentially reliable methods, detect spuriousness, and identify strengths and limitations of remote sensing to infer plant diversity. However, these simulations have been develop using simplistic assumptions so far.

Here, we present the Biodiversity Observing System Simulation Experiment (BOSSE) v1.0, a simulation tool able to represent vegetation taxonomy and functional traits and a wide range of physically linked spectral signals and ecosystem functions in space and time. BOSSE simulates maps of vegetation species and their (functional) traits that evolve in time as a function of biometeorological drivers. From these dynamic scenes, BOSSE can simulate hyperspectral reflectance factors, sun-induced chlorophyll fluorescence, land surface temperature, and provide estimates of plant functional traits based on these signals. BOSSE can simulate observations of varying spectral configurations and spatial and temporal resolutions, mimicking current and future remote sensing missions. Remote sensing imagery can be generated with an hourly temporal resolution, and the spatial resolution can be degraded to assess the impact of this feature in the estimation of plant diversity. Ecosystem functions (mainly related to carbon, water, and energy fluxes) can be generated at hourly steps to develop robust methods that allow for testing the variability of biodiversity-ecosystem function (BEF) relationships, which is still an open question in ecological research.

BOSSE is an open-source Python model that we make available to the community to support the development of new remote sensing-based biodiversity products and assess the role of biodiversity in ecosystem functions. Moreover, it could also be useful to address other methodological questions regarding the study of vegetation. BOSSE does not aim to substitute the still-necessary observational data and studies but to support their design and interpretation.

How to cite: Pacheco-Labrador, J., Gomarasca, U., Duveiller, G., Pabon-Moreno, D. E., Li, W., Weber, U., Jung, M., and Migliavacca, M.: The Biodiversity Observing System Simulation Experiment (BOSSE) v1.0, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-13639, https://doi.org/10.5194/egusphere-egu25-13639, 2025.

Creating an ideal environment to enrich biodiversity can be managed in two ways, i.e. by protecting rare and common species (i), and by improving the environmental conditions of the specific ecosystem (ii). The first measure results in an excellent balanced food pyramid with a sustainable cyclic biotic and mineral energy transition. The second measure, however, is connected to minimal entropy production according to Penrose´s claim that biologic life is lowering the entropy production rate of Earth. Minimum entropy implies maximum order so that any ecosystem tends to maximum biodiversity, at least to its local boundary conditions. Entropy production of an ecosystem, is linked to the Shannon entropy of the statistical species distribution of the respective ecosystem according to Stephan Hubbel´s Unified Neutral Theory of Biodiversity (2001). Hubbel put on the statistical ensemble of species situated in a mature vegetation area, a lognormal distribution of species according to the McArthur´s Island theory and confirmed by Fisher. The standard deviation decreases with entropy production increase and vice versa. This distribution contains one deficiency since it underestimated the contributions of rare species.

We derive Earth´s entropy production from the Stefan-Boltzmann law. Monthly land surface temperature (LST) are obtained from remotely sensed MODIS and SENTINEL data over the period 2003-2020 and monthly latent heat data from the FLUXCOM-X global fluxes collection for a one by one kilometer pixels. We analyse 11 ecosystems worldwide (mean of 3 x 3 pixels). Eight of them are National Parks where minimal anthropogenic stress can be assumed. Three control areas subjected to human economic activity nearby National Parks are added for comparison.

A decline in entropy production down to -3.7% per decade is observed in areas around the equator (Foz do Iguaço in Brazil, the Ngorongoro in Tanzania, and Gal Oya in Sri Lanka). The entropy production over the Mediterranean area (Spain) and northern Europe (Finland) is stable, while the entropy production is increasing dramatically up to +2.4% per decade over the western European National Parks (the Netherlands, Flanders). These areas are characterized by a very high anthropogenic environmental pressure. Differences in the trends of entropy production are observed when mean values are computed from 3 x 3 pixels or from 9 x 9 pixels. Generally, the more the pixels, the smoother, the smaller the absolute trend values. For the Landes in France, the trend switches from a small negative value on the larger scale to a more substantial positive value at the smaller scale. Generally, wetter ecosystems tend to lower the Earth´s entropy production thereby increasing the biodiversity of vegetation.

How to cite: Verstraeten, G. J. M. and Verstraeten, W. W.: Evaluating the vegetation biodiversity from space, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-2831, https://doi.org/10.5194/egusphere-egu25-2831, 2025.

Considering Arctic-Antarctic (North-South) carbon dioxide (CO₂) dynamics over time, one can see a ripple of a Vegetation Signal [VRi] on an increasing baseline trend [BTr]. The VRi gradually disappears when moving southwards (towards the Antarctic). It is observed in CDIAC data that the [VRi], superposed on an increasing CO₂ baseline trend increase disappears when traveling southwards. How so?

The observation is intuitively obvious, since the Northern hemisphere contains approximately 68% of the Earth's landmass and is home to about 90% of the global population. In contrast the Southern hemisphere is left with 32% landmass and a tiny 10% of the global population. Due to these differences in landmass and population, the North South dynamics of carbon dioxide mixing ratios is primarily determined by the differences in vegetation density, human population density, and atmospheric circulation patterns, including a variety of factors of less impact on atmospheric CO₂ dynamics.

More sources of CO₂ in the Northern hemisphere determine the global CO₂ trend line over time on the condition that the atmosphere is well mixed in a short time interval. The Intertropical Convergence Zone (ITCZ)  acts as a barrier to the mixing of air masses from the Northern and Southern Hemispheres. This makes CO₂ dynamics in each hemisphere more distinct in a short-term time frame (seasons). The [BTr] is determined more by long-term CO₂ emissions from anthropogenic origin (years).

When comparing global trends in CO₂ mixing ratios obtained with remote sensing estimates by NASA's Orbiting Carbon Observatory (OCO) with measurements from the CDIAC CO₂ monitoring stations over several years, the separation of [VRi] from [BTr] leads to interesting results. 

Ensuring the data are consistent, one is required to remove outliers and perform gap-filling if necessary. Subsequently one has to decompose the CDIAC CO₂ time series into its [VRi] and [BTr] components. This can be done using techniques such as seasonal numerical decomposition of time series. The seasonal component [VRi] represents a regular annual cycle driven essentially by vegetation photosynthesis and respiration. The increasing trend component [BTr], reflects the more long-term changes in CO₂ mixing ratio’s driven by anthropogenic and other sources of CO₂ emissions. A harmonic model is fitted to the deseasonalized and detrended data to quantify the seasonal amplitude and phase of [VRi]. The seasonal amplitude represents the strength of [VRi] due to carbon fixation, while the phase indicates the timing of maximum uptake and release of CO₂, depending on the latitude in the Northern and Southern hemispheres. To validate atmospheric CO₂ dynamics of remotely sensed CO₂ mixing ratios, the CDIAC measured CO₂ mixing ratios are used in a comparison of both types of CO₂ data. Only then, factors, such as climate, land cover and human population densities, can be understood better. It may allow to model the forcing processes determining RS observed and measured trends and variations of CO₂ mixing ratios, and their impact on changes in climate.

How to cite: Veroustraete, F.: A North-South gradient of CDIAC CO2 Mixing Ratios compared with Data from Atmospheric Remote Sensing of CO2 , EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-21074, https://doi.org/10.5194/egusphere-egu25-21074, 2025.

The accurate assessment of vegetation phenology, i.e. the seasonal changes in plant development status, is a critical step for the study of ecosystems’ performance and their response to the ongoing climate change. Satellite products have been used for a quite long time for this purpose, frequently with the combination of data from eddy flux towers. In recent years, the use of digital photography from properly parameterized cameras (phenology cameras), has been introduced as a very promising low-cost methodology for precise monitoring of vegetation phenology.

In this study, we provide an assessment of vegetation phenology for a five-year period (2019 – 2024) of a black locust plantation, located in the restored areas of the Lignite Center of Western Macedonia, Greece. During the study period, repeated RGB images in half – hourly time steps were automatically taken with the use of a phenology camera installed on the top of an eddy flux tower in the study site and timeseries of the Green Chromatic Coordinate (GCC) index were extracted. Furthermore, for the same period, high resolution Sentinel 2 satellite products were used for the estimation of four commonly used vegetation indices (NDVI, EVI, LSWI and SAVI). Finally, carbon flux phenology was assessed from the respective measurements of the eddy flux tower.

Our results indicate that the use of phenology cameras can provide an explicit representation of vegetation phenology and estimation of the respective phenological indices (start, end, peak and length of the growing season). GCC is well correlated with all the studied satellite vegetation indices, however it provides the advantage of the continuous measurements, as the results are not affected by the weather conditions, in contrary to satellite products.  Additionally, it can be used for the distinction of overstory and understory vegetation phenological status, which is very critical especially in deciduous ecosystems, but it cannot be assessed with the use of satellite products. Furthermore, it can provide valuable information for other phenological parameters, such as the start and length of the blooming period, which is also difficult to assess by other methods. Concerning the representation of carbon flux, GCC does not seem to provide any further direct advantages compared to satellite vegetation indices, however the ability of the distinction of understory and overstory vegetation phenology can provide other benefits, such as the more efficient parameterization of productivity models.

How to cite: Markos, N. and Radoglou, K.: Tracking vegetation phenology for a deciduous black locust plantation with the use of a phenology camera, satellite vegetation indices and eddy flux measurements – advantages and limitations, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-12849, https://doi.org/10.5194/egusphere-egu25-12849, 2025.

Monitoring habitat condition is becoming increasingly important in light of the biodiversity crisis. Advances in UAV remote sensing and artificial intelligence are creating opportunities to complement field-based habitat monitoring or provide effective alternatives. As part of MAMBO, an EU-funded project, we aim to develop generic workflows that can deliver crucial habitat condition metrics using affordable drone remote sensing. Shrub cover and biomass in grassland, wetland, and shrub habitats are important for monitoring rewilding or habitat restoration efforts and above ground carbon. Here we describe a workflow, involving deep learning and allometry, developed to map the biomass of individual hawthorn shrub clumps. Our use case is a rewilded farm in Bedfordshire, UK. Results show that (i) U-Net variants are suitable for accurately mapping hawthorn within a complex shrub matrix, and (ii) allometry, based on structure-from-motion derived height, is an effective and affordable solution for shrub biomass mapping.

How to cite: Gerard, F., Zhang, C., Barbedo, R., George, C., Upcott, E., Kelley, D., and Broughton, R.: Shrub species, cover and biomass from affordable UAV observations, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-1632, https://doi.org/10.5194/egusphere-egu25-1632, 2025.

Climate change poses unprecedented risks to agriculture production, thus accurate and timely crop yield forecasting is pivotal for ensuring global food security and agricultural market stabilization, especially in South Asia, a key region for rice production and export. While a growing number of studies have explored the potential of machine learning-based models for rice yield prediction, these efforts are often limited to local scales and ignore the spatiotemporal nature of rice yield in the modeling process. In this study, we propose a graph-based recurrent neural network (GNN-RNN) framework for predicting district-level rice yields in South Asia using publicly available data. The model integrates multi-source datasets, including climate observations, satellite-derived phenological metrics, soil maps, and historical yield records. By aggregating these inputs at the district scale through rice distribution masks, we extract time-series features with a Convolutional Neural Network (CNN) and utilize a GNN-RNN model to process spatiotemporal embeddings. The GraphSAGE algorithm captures geographical relationships, while the RNN component enhances predictions by incorporating temporal dependencies. Validation against five baseline machine learning models (CNN, CNN-RNN, LSTM, gradient boosting, random forest) from 2000 to 2020 shows the GNN-RNN outperforms alternatives, achieving an average R² of 0.75 and RMSE of 288 kg/ha for monsoon-season rice yields. Further tests confirm its robustness in both normal and extreme weather years, with leave-one-year-out RMSEs ranging from 234 to 366 kg/ha (11-18% of the long-term mean yield). The framework also quantifies uncertainty, with over 80% of observed yields falling within the 95% confidence interval, and prediction reliability improving throughout the growing season. This study demonstrates the potential of graph-based AI models for high-resolution crop yield forecasting, offering critical insights for food security and climate resilience. Future research could explore the model's application to extreme weather and pest impacts, as well as the integration of advanced remote sensing datasets to further enhance its predictive power.

How to cite: Zhang, C., Zhao, C., and He, X.: A GNN-RNN Framework for Rice Yield Prediction in South Asia, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-3422, https://doi.org/10.5194/egusphere-egu25-3422, 2025.

The term "invasive noxious weed species” (INWS), which refers to noxious weed plants that invade native alpine grasslands, has increasingly become an ecological and economic threat in the alpine grassland ecosystem of the Qinghai-Tibetan Plateau (QTP). Both the INWS and native grass species are small in physical size and share a habitat. Using remote sensing data to distinguish INWS from native alpine grass species remains a challenge. High spatial resolution hyperspectral imagery provides an alternative for addressing this problem. Here, we explored the use of unmanned aerial vehicle (UAV) hyperspectral imagery and deep learning methods with a small sample size for mapping the INWS in mixed alpine grasslands. To assess the method, UAV hyperspectral data with a very high spatial resolution of 2 cm were collected from the study site, and a novel convolutional neural network (CNN) model called 3D&2D-INWS-CNN was developed to take full advantage of the rich information provided by the imagery. The results indicate that the proposed 3D&2D-INWS-CNN model applied to the collected imagery for mapping INWS and native species with small ground truth training samples is robust and sufficient, with an overall classification accuracy exceeding 95% and a kappa value of 98.67%. The F1 score for each native species and INWS ranged from 92% to 99%. In conclusion, our results highlight the potential of using very high spatial resolution UAV hyperspectral data combined with a state-of-the-art deep learning model for INWS mapping even with small training samples in degraded alpine grassland ecosystems. Studies such as ours can aid the development of invasive species management practices and provide more data for decision-making in controlling the spread of invasive species in similar grassland ecosystems or, more widely, in terrestrial ecosystems.

How to cite: Xing, F., An, R., Guo, X., and Shen, X.: Mapping invasive noxious weed species in the alpine grassland ecosystems using very high spatial resolution UAV hyperspectral imagery and a novel deep learning model, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-20351, https://doi.org/10.5194/egusphere-egu25-20351, 2025.

The decline of wild plant species across European agricultural landscapes threatens biodiversity and vital ecosystem functions. While result-based payments to farmers show promise for species conservation, implementing such programs has been hindered by the high costs of traditional biodiversity monitoring. Our study explored a novel solution using uncrewed aerial vehicles (UAVs) equipped with an RGB camera and deep learning technology to efficiently detect and monitor these important plant species.

We conducted our research across four winter barley fields in Germany under different management intensities. Using the YOLO deep learning model, we analysed UAV imagery to detect segetal flora species across multiple flight altitudes. To validate and enhance our detection methodology, we collected detailed field measurements of plant traits and species coverage, and investigated whether spatial co-occurrence patterns and canopy height variations could help predict the presence of species that are challenging to detect from aerial imagery.

Our findings revealed that UAV-based monitoring could successfully detect 50% of the observed species on-site, with optimal results achieved for developing manual annotations at a ground sampling distance of 1.22mm. Plant height emerged as a crucial factor in detection success, with detection probability increasing with plant height. Based on the trait analysis, we projected similar detection success rates for key indicator species not present in our study area. The YOLO models showed accuracy rates vary between 49% to 100% depending on the management type, and performed effectively at a flight height of 40m enabling rapid field surveys that required only eight minutes per hectare. Notably, we found that both the spatial co-occurrence with easily detectable species and variations in canopy height structure showed potential as predictors for the presence of harder-to-detect species. While these findings are promising, additional research is needed to validate these relationships across broader landscape scales.

This study demonstrates the feasibility of implementing large-scale, cost-effective monitoring of wild plant indicators in agricultural settings. Our results provide a foundation for developing sophisticated 'smart indicators' for future biodiversity monitoring practices. This technological approach could make result-based conservation payments more practical and widespread, ultimately supporting the preservation of vital plant species in agricultural ecosystems.

How to cite: Barrasso, C., Krüger, R., Eltner, A., and Cord, A.: Automated detection of wild plant indicators in agricultural fields: Integrating UAV technology and deep learning for result-based payments, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-18238, https://doi.org/10.5194/egusphere-egu25-18238, 2025.

The loss of biodiversity has been recorded globally at unprecedented rates. Among the various organisms under threat are the European forest dwelling bats, which experienced a significant population decline. One of the causes of this decline is the alteration and destruction of their habitat. To understand how bats interact with forests, a detailed characterisation of their habitat can help target conservation efforts. The common habitat characterisation approach in forests is to carry out field surveys. During these surveys the forest is visually described by qualitative indices of complexity and structure. This detailed surveying requires extensive time investment and highly depends on the field crew who conducts the survey. Naturally, an observer bias is inevitable. Some characteristic parameters, like the volume of gaps or foliage density, cannot be determined by conventional measuring approaches.

In recent years, LiDAR-based 3D point clouds are increasingly used to characterise habitats.  In forest environments, different vegetation density and layering, as well as the changing terrain, make the point cloud-based characterisation particularly challenging. Existing approaches resort to 2.5D raster data, disregarding the full potential of the three-dimensionality that point clouds provide. Given that bat species utilise both tree crowns and the ground, such information is of the utmost importance.

In this work, we present a full 3D point cloud analysis for forest habitats. We quantitatively characterise the habitat and provide a characterisation approach for complex environments. By analysing the acquired point cloud in 3D, we infer the forest structure as a whole. Such a characterisation allows us to assess how much area is potentially used by bats for flying and foraging. The quantitative nature of the characterisation enables the comparison between vegetation structures in different forest stands.

We demonstrate the proposed characterisation on different forest stand types, i.e., beech and mixed forests, in the Vienna Biosphere Reserve. Designated areas were captured with a handheld mobile laser scanner. We show that both for dense and sparse stands the proposed characterisation approach was successfully applied. Therefore, our analysis can be applied to all forested ecosystems, encompassing orchards as well as avenues. The analysis is performed in R and is easy to use. In this way, we can establish better conservation strategies for endangered forest species worldwide.

How to cite: Dammert, L., Suarez-Rubio, M., and Arav, R.: Three-dimensional Habitat Characterisation for European Forest Bats Using High-resolution Point Clouds , EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-12425, https://doi.org/10.5194/egusphere-egu25-12425, 2025.