This study presents advancements in forest change detection by leveraging self-supervised learning (SSL) methods with multi-source and multi-temporal Earth Observation (EO) data. Transitioning from traditional bi-temporal approaches, the developed methodology incorporates multi-temporal analysis and multimodal data fusion using Sentinel-1, Sentinel-2, and PALSAR-2 imagery. Key innovations include mapping the magnitude of forest changes rather than binary classifications, enabling nuanced assessment of disturbance severity.
Experiments demonstrate the effectiveness of SSL-pretrained backbones, such as ResNet architectures, in extracting features for change detection. The integration of multi-temporal Sentinel-1 time series further improved the reliability and accuracy of disturbance tracking over time. These advancements show the potential of SSL to enhance forest change monitoring, providing scalable solutions for continuous and precise assessment of forest dynamics.
How to cite: Kuzu, R., Zappacosta, A., Antropov, O., and Dumitru, O.: Enhancing Forest Change Detection Using Self-Supervised Learning with Multi-Source EO Data, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-294, https://doi.org/10.5194/egusphere-egu25-294, 2025.
Given the impact of climate change on deciduous crop yields, our research focuses on leveraging earth observation remote sensing to accurately detect flowering periods in almond orchards and evaluate a climate-based dormancy model for predicting flowering times. This study addresses the challenge of monitoring almond flowering phenology by employing automated crop mapping techniques to support phenology monitoring across California's Central Valley. Using Sentinel-2 (S2) multispectral satellite imagery, we compare its effectiveness with the carbohydrate-temperature (C-T) dormancy model. The study area encompasses approximately 30,000 almond orchards, precisely identified using the Almond Industry Map. We utilized time-series analyses of the Enhanced Bloom Index (EBI) and the Normalized Difference Vegetation Index (NDVI) to quantify bloom periods and intensity and determine peak bloom times. Leveraging around 4,000 S2 tiles, enhanced vegetation indices, and in situ time-lapse camera data collected from 2019 to 2022, we developed a robust methodology for accurately identifying peak bloom periods. This process created a comprehensive phenological dataset, which was standardized and interpolated to daily resolution for improved time-series analysis. Our approach achieved a mean absolute error (MAE) of just 1.9 days in detecting peak bloom, demonstrating the accuracy of satellite-based phenological monitoring. This underscores both the advantages and limitations of remote sensing technologies in agricultural phenology. The dataset was then used to validate projections from the climate-based carbohydrate-temperature (C-T) dormancy model, offering valuable insights and supporting the refinement of this mechanistic approach. The study revealed significant spatial and temporal patterns in flowering phenology, emphasizing the role of regional climatic conditions in influencing crop development. Results highlight the potential of remote sensing and satellite imagery to detect the start, peak, and end of bloom in almond orchards with high precision, generate valuable phenological datasets, monitor patterns at both regional and field scales, and assess the reliability of dormancy models. This research has critical implications for improving agricultural practices and supporting decision-making in the almond industry. By advancing phenological monitoring techniques, our study presents a scalable and innovative approach to managing perennial crops in the face of climate change.
How to cite: Paz-Kagan, T., Lauterman, O., Kizel, F., Zwieniecki2, M. A., Orozco, J., and Sperling, O.: Leveraging Satellite Earth Observation for Detecting Bloom Shifts and Phenological Patterns in California’s Almond Orchards, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-1993, https://doi.org/10.5194/egusphere-egu25-1993, 2025.
Forest fragmentation disrupts habitat continuity, reshapes ecosystem processes, and threatens biodiversity. Effective conservation efforts in fragmented landscapes rely on precise monitoring of these changes. This study leverages remote sensing through vegetation indices to evaluate forest health and detect fragmentation-induced alterations over time. Focusing on the Tuchola Forest in Poland, an area increasingly affected by windstorms, we analyzed Sentinel-2 imagery from 2016 to 2024 using 19 vegetation indices. Machine learning classifiers—Extra Trees, Random Forest, and LightGBM—were employed to assess which indices best capture fragmentation stress. The Extra Trees classifier outperformed the others in accuracy and generalization, identifying NDWI and GNDVI as the most effective indicators. These indices were particularly responsive to shifts in vegetation water content and canopy density linked to fragmentation. Our findings underscore the utility of targeted vegetation indices for precise ecological monitoring and inform conservation strategies in fragmented forests.
How to cite: Dutt, S. and Kunz, M.: Uncovering Fragmentation Patterns: Optimal Vegetation Indices for Monitoring the Tuchola Forest Ecosystem, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-353, https://doi.org/10.5194/egusphere-egu25-353, 2025.
Illegal charcoal production, by means of indiscriminate logging activities, poses significant threats to the stability of the drylands’ ecosystem in the Somali territory. In addition, the revenues from the charcoal trade often serve further illegal activities, exacerbating the already complex socio-political context of the country. In this work, we investigated the application of freely available multi-sensor remote sensing products (Sentinel-1 and Sentinel-2) and machine learning techniques to detect the presence of charcoal production sites (i.e., kilns) over large areas. Exploiting Google Earth Engine and open-source tools, we were able to develop a binary classification of kilns’ presence-absence for the years 2019, 2020, and 2021 in a remote area (approximately 32000 km2) north-west of Mogadishu, Somalia. Concerning the workflow, we first computed median images, spanning the first three months of each year, composed of numerous optical, SAR (Synthetic Aperture Radar), and combined vegetation indices. Images were then subtracted between consecutive years and a Support Vector Classification (SVC) algorithm was trained and validated on the indices’ values extracted from those. As a reference dataset, we employed known kilns’ locations from a preceding study by FAO-SWALIM, where photointerpretation of very high resolution images was used to individuate the appearance of illegal charcoal kilns. The evaluation of the classifications showed that our approach has great capabilities for the automatic individuation and the monitoring of illegal charcoal production sites, with R2 values and accuracy metrics ranging between 0.80-0.88 for the three considered years (2019, 2020, 2021). Moreover, mappings of the predicted presence-absence of kilns (at 10 m spatial resolution) were produced starting from the trained SVC model, giving a spatial representation of the phenomenon and allowing an assessment of the most impacted areas. In conclusion, our results represent a significant advancement in monitoring illegal charcoal production activities in Somalia, offering a reliable and transferable methodology based on accessible satellite imagery and tools.
How to cite: de Guttry, L., Abdi Olow, I., Paron, P., Bolognesi, M., Leonardi, U., Stendardi, L., Argenti, G., Moriondo, M., and Dibari, C.: A multi-sensor remote sensing approach to monitor illegal charcoal production sites in Somalia’s forests, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-17735, https://doi.org/10.5194/egusphere-egu25-17735, 2025.
Citrus fruit cracking, a physical failure of the peel, causes yield losses of 10% to 35%, peaking during October-November. Water status of the tree and water flow into the fruit influence this phenomenon. with excessive irrigation during critical fruit development stages exacerbates cracking. As part of the EU-Horizon CrackSense project, this study is aimed to link citrus tree plant water status (PWS) to fruit cracking, emphasizing how deficit irrigation can reduce yield loss due to cracking. Using UAV and eco-physiological measurements, we developed models to predict PWS and its relationship with cracking intensity early in the season. The study, conducted in 2023-2024 in a commercial orchard near Kfar Chabad, Israel, tested four irrigation treatments: control, defined as the standard irrigation, two deficits irrigations regimes (50% of control) early and late in the season, and excessive irrigation (150% of control) throughout the season. Ground-based measurements included fruit and trunk diameter, stem water potential (SWP), stomatal conductance, plant area index (PAI), and growth rate (TG). UAV flights integrated multispectral, thermal, and LiDAR sensors to capture spatial-temporal variability in PWS. Canopy metrics, such as height, volume, LiDAR-based PAI, and spectral and thermal indices, were incorporated into PWS models. Results revealed significant differences in TG, SWP, and stomatal conductance for 50% of early and late deficit irrigation treatments compared to other treatments. Random forest models demonstrated strong predictive performance for SWP (R² > 0.77) and TG (R² > 0.76). LiDAR-derived PA correlated highly with field optical measurements (R² = 0.92), yield (R² = 0.67), and cracked fruit percentages (R² > 0.50). This study underscores the importance of precise irrigation management in reducing fruit cracking. It highlights the potential of remote sensing systems for predicting cracking and managing water status at the tree level. The developed models equip farmers with tools to apply controlled water stress, minimizing cracking and improving yield.
How to cite: Dubinin, M. (., Morozov, M., Sadka, A., and Paz-Kagan, T.: Linking Citrus Fruit Cracking Intensity to Plant Water Status: Insights from UAV-Derived Metrics Validated by Ground-Based Data, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-2520, https://doi.org/10.5194/egusphere-egu25-2520, 2025.
Scandinavia is facing climate changes with a predicted increase in mean temperature of 2-4°C. For Swedish forests to be adapted to this challenge, the Swedish tree breeding program aims to select trees that are adapted to different biotic and abiotic conditions. Information on spring phenology, damage and vitality are important variables in the Norway spruce selection process. Traditionally, the data is gathered through manual assessment of each tree, which requires significant resources and limits the number and frequency of variables that can be measured. As an alternative, Remote Sensing is a promising technology to evaluate bud flush and vitality in conifers, offering the advantage of scoring more trees in a shorter time with fewer resources while obtaining data for several time points during the vegetation season, and its use of algorithms to measure variables reduces the risk of human error.
This project aims to develop methods that can be used within the breeding program by collecting information on spring phenology, damage and vitality using high-resolution multispectral drone images of young Norway spruce trees. Data were collected during spring 2023 and 2024. Bud flush is estimated from the spectral values of the tree crowns using manual assessment of the flush in a subset of the trees as training data. The high-resolution multispectral images will also be used to assess the damage and vitality of the new shoots. To ensure capturing the bud flush at a high temporal resolution, images were taken before the vegetation season and up to twice weekly during the period with the most rapid flush. In the final step, the spatial pattern within the study sites will be analyzed and connected to damage and vitality of the young Norway Spruce trees.
How to cite: Eurich, L., López Fernández, S., Elfstrand, M., García-Gil, M. R., Bohlin, J., and Lindberg, E.: Assessment of bud flush and damage in young Norway Spruce trees through airborne high-resolution multispectral images, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-5892, https://doi.org/10.5194/egusphere-egu25-5892, 2025.
Bananas are the tropical fruit with the largest global cultivation area, sales volume, and
international trade. China is the world's second-largest producer and consumer of bananas.
Rapid and accurate acquisition of banana planting range and spatial distribution information
is crucial for promoting the sustainable development of the banana industry in China.
Currently, research on banana classification and identification faces challenges such as
insufficient mechanistic understanding, poor generalizability, and difficulties in large-scale
application. Additionally, banana cultivation areas are often located in regions with cloudy
and rainy climates, limiting the acquisition of optical imagery. To address this, this study
constructs a banana identification model based on phenological characteristics: (1) Sentinel
1/2 imagery is utilized to obtain time series curves of banana spectral and scattering features,
followed by interpolation and filtering of the time series data; (2)A phenological index based
on optical and scattering features is developed according to banana phenological
characteristics. By combining SAR with the index, the model's mechanistic understanding is
enhanced while alleviating the challenges posed by cloud cover in tropical and subtropical
regions; (3)Using the constructed phenological index alongside banana spectral, texture, and
temporal features, a classification model is trained for banana identification in the study area.
This banana forest identification model and the developed phenological index aim to resolve
current issues in banana classification and provide theoretical and practical support for large
scale banana extraction and the study of tropical and subtropical economic crops.
How to cite: wang, Z.: Banana plantation identification using remote sensing data in tropical and subtropical regions, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-2709, https://doi.org/10.5194/egusphere-egu25-2709, 2025.
In the present decade, forest fires have become more common than ever [1]. Efficient strategies to cope with fire situations, and/damage assessments need efficient automatic forest fire detection model. In this research, we propose an unsupervised eXplainable machine learning model to assess the severity of forest fire with remote sensing data. The model, namely, Latent Dirichlet Allocation is a Bayesian Generative model, is capable of generating interpretable visualizations. LDA uncertainty quantifiable and explainable [2]. We do not need labelled data to train the model. Other usefulness of the model is that it is simple to combine any kind of input data (for example, UAV images, wind speed information). In the scope of this contribution, we use Sentinel-2 spectral bands to extract information to compute indices indicating severity of fire [1]. Uncertainty of each prediction of the model is computed to ascertain robustness of the model. As a use case, we have chosen the recent forest fire incident at Los Angeles, USA [6].
The methodological approach is as the following:
1) we acquire pre-fire, post-fire Seintinel-2 images, 2) compute three indices : Normalized Difference Vegetation Index (NDVI), Normalized Burn Ratio (NBR), and Burned Area Index for Sentinel (BAIS) based on state of the art literature and generate index maps, 3) compute difference between the pre-fire and post-fire index maps, 4) apply the unsupervised xAI LDA model to retrieve semantic classes in pre-fire and post-fire Sentinel-2 band images, general corresponding classification maps and plot a binary class-to-class change map, 5) Analyze the maps with visual tool to find the most affected semantic classes (e.g., dense vegetations, urban areas etc.) and produce a data-driven estimation of per-class changes due to fire [7].
In future, we plan to fuse other data sources (e.g., wind speed information [5]) to help practical applications.
Reference:
[1] Lasaponara, A. M. Proto, A. Aromando, G. Cardettini, V. Varela and M. Danese, "On the Mapping of Burned Areas and Burn Severity Using Self Organizing Map and Sentinel-2 Data," in IEEE Geoscience and Remote Sensing Letters, vol. 17, no. 5, pp. 854-858, May 2020, doi: 10.1109/LGRS.2019.2934503.
[2] Karmakar, C. O. Dumitru, G. Schwarz and M. Datcu, "Feature-Free Explainable Data Mining in SAR Images Using Latent Dirichlet Allocation," in IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, vol. 14, pp. 676-689, 2021, doi: 10.1109/JSTARS.2020.3039012.
[3] California Wildfires Live Updates: 24 Dead in L.A. as Dangerous Winds Threaten Fire Growth - The New York Times
[4] Sentinel-2 mission. Available online: https://sentinel.esa.int/web/sentinel/missions/sentinel-2
[5] Global Wind Atlas. Available online: https://globalwindatlas.info/en/about/dataset
[6] ESA news based on Sentinel-2. Available online: https://www.esa.int/ESA_Multimedia/Missions/Sentinel-2/(offset)/100/(sortBy)/published/(result_type)/images
[7] Karmakar, C.O. Dumitru, N. Hughes and M. Datcu, "A Visualization Framework for Unsupervised Analysis of Latent Structures in SAR Image Time Series", IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 16, pp. 5355-5373, 2023.
How to cite: Dumitru, O., Karmakar, C., and Goyal, S.: Explainable Machine Learning for Forest Fire Detection with Remote Sensing for Effective Rescue Planning, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-16843, https://doi.org/10.5194/egusphere-egu25-16843, 2025.
In this paper we study the relationship between vegetative phenology obtained from satellite-derived vegetation indices (VIs) and vegetative and floral phenology based on field observations. The work was conducted during 2023 and 2024 in vineyards belonging to the Designation of Origin Uclés, located in the west of Cuenca province (Castilla-La Mancha region, central Spain). The field work was carried out in seven plots that are frequently ploughed and lack cover crops and green covers. All plots grow under similar conditions and the maximum distance between plots is less than 2 kilometers. Phenological sampling was carried out weekly on 20 grapevines per plot, using the BBCH scale.
Different VIs (NDVI, EVI, SAVI and SAVI2) were calculated using Google Earth Engine (GEE) and Sentinel-2 data, but EVI was selected due to its greater amplitude in the index curves. The R package Phenofit was used to clean the data, curve fitting and extraction of phenology metrics. For curve fitting, the Elmore method was used, and for phenology metrics extraction, the Threshold, Inflection and Gu methods from the Phenofit package were applied. Although Inflection and Gu differ in their approach, they both divide the curve into four phenological metrics: greenup, when index starts to growth; maturity, when the index value remains stable; senescence, when it decreases; and dormancy, when it stops decreasing and remains at a low value. Threshold considers only greenup and dormancy.
The results show that greenup is associated with the inflorescence development. This phase starts in a similar day of the year (DOY) in all plots and in the two studied years. Maturity, marked by Inflection and Gu methods, occurs between flowering and fruit development stages, that is, between DOY 140–198. The senescence period is marked between fruit development and leaf discoloration (178–310 DOY), and despite its amplitude, 75% of the observations place senescence between the final stages of the fruit and leaf discoloration. Finally, dormancy occurs between leaf discoloration and the leaf fall which is correct but usually it is marked excessively late.
Phenological metrics derived from Vegetation Indices (VIs) such as greenup (potentially related to inflorescence development), senescence (potentially related to leaf discoloration), and dormancy (potentially related to leaf discoloration and fall) can be linked to the grapevine cycle on the BBCH scale. However, more studies are needed to accurately link field phenological observations with satellite-derived vegetation indices.
This work has been funded by the Junta de Comunidades de Castilla-La Mancha (JCCM) through the project SBPLY/ 21-180501-000172 and by the University of Castilla-La Mancha (UCLM) through the project 2022-GRIN-34507. EJJ thanks to the Investigo Program for a contract co-financed by the European Social Fund Plus.
How to cite: Jiménez-Jiménez, E., Muñoz-Gómez, G., Lara, B., Fernández-González, F., and Pérez-Badia, R.: Remote sensing applied to phenology monitoring in vineyards: testing through field observations, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-17743, https://doi.org/10.5194/egusphere-egu25-17743, 2025.
Accurate yield estimation and appropriate planting management policies for rubber plantations require their precise information on spatiotemporal change data. Previous studies on mapping of rubber plantations did not employ the dynamic rubber phenology features and had difficulty in obtaining historical samples. Here we attempted to develop a new mapping framework through taking historical sample migration, dynamic phenology, and change detection variables into the classification procedure. An automatic sample migration algorithm was first proposed to generate historical samples. Then, two new variable types, dynamic phenology indices and change detection variables, were developed. Another four commonly used variable types -spectral bands, yearly composite spectral indices, terrains, and textures were also extracted. Five combinations of variable types were designed to explore key variable types. Subsequently, the framework with recommended variable types was applied at an experimental site in China and was finally evaluated to two test sites in Myanmar and Thailand for examining its transferability. Results showed that the average overall accuracy of historically migrated samples reached over 97% at the experimental site. Dynamic phenology indices and change detection variables were found as two crucial variable types for rubber plantations mapping. The average rubber plantations mapping accuracy during 2003-2022 reached 93.68%. Transferring the proposed framework to two test sites confirmed the independent roles of change detection variables and dynamic phenology indices. Their average rubber plantations mapping accuracy during 2003-2022 reached 94.34% and 93.73%, respectively. Good spatial consistency between the classified maps and Google Earth images was observed, displaying clear boundaries between rubber plantations and farmland, evergreen broadleaf forest, and shrub. Overall, the proposed framework has great potential for time series rubber plantations mapping in Southeast Asia.
How to cite: Chen, Y. and Xu, H.: A new framework for mapping time series rubber plantation in Southeast Asia, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-2120, https://doi.org/10.5194/egusphere-egu25-2120, 2025.
The proliferation of remote sensing (RS) data and advancements in mechanistic crop modeling and data assimilation techniques necessitate a framework that digitally represents cropping systems and their spectral properties. Such a framework would enable crop growth simulation, scenario testing, and timely prediction updates using RS data.
In this study, we develop a comprehensive coupling scheme that links a crop model (DSSAT-CROPGRO) with a radiative transfer model (RTMo module in SCOPE). This integration allows for the utilization of reflectance data from all measured spectral bands during data assimilation (DA) into the crop model.
We apply this coupled crop-radiative-transfer model in a DA experiment using a novel particle filter scheme. The assimilated data consists of observed reflectance measurements obtained by a multispectral camera mounted on an unmanned aerial vehicle (UAV). Using multispectral data with a high spatial resolution for analyzing a row crop required a dedicated analysis to fit model simulations to measurements. The suggested DA scheme was implemented in an irrigation and fertilization trial with processing tomatoes to evaluate its effectiveness.
The results showed that applying the DA scheme improved the NRMSE of the Leaf Area Index (LAI) from 59% to 41.8% and yield from 63.6% to 35.4%. The DA scheme performed best when the treatment that included the most severe stress was excluded from weight calculation, resulting in NRMSE of 34.1% and 15.5% for LAI and yield, respectively. After showing promising results, the suggested data assimilation scheme should be tested in large-scale, commercial fields using space-borne RS data to examine its applicability in various scenarios.
How to cite: Weinman, A., Malachy, N., Linker, R., and Rozenstein, O.: Integrating UAV Multispectral Data into a Combined Crop-Radiative Transfer Model for Processing Tomatoes Using a Particle Filter, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-6085, https://doi.org/10.5194/egusphere-egu25-6085, 2025.
Evapotranspiration (ET) is crucial for water resource management, agricultural planning, and understanding land-atmosphere interactions. Numerous approaches are available for estimating ET at various spatial and temporal scales, including ground-based measurements, mechanistic models, and remote sensing. In this study, we aimed to enhance the accuracy and applicability of the Sentinel for Evapotranspiration (Sen-ET) plugin for estimating ET in diverse field crops in Israel. The primary objectives were to validate the Sen-ET method using eddy covariance (EC) measurements across various seasons and crop types, improve Sen-ET estimates by incorporating local weather station data, and illustrate the influence of weather station distance from measurement sites on Sen-ET accuracy.
The research was conducted across eight test sites in Israel, including fields with spring wheat, potato, cotton, and tomato. In applying Sen-ET model, we utilized high-resolution Sentinel-2 and Sentinel-3 imagery, along with ERA-5 meteorological data and local weather station inputs. The ET estimations by Sen-ET involved preprocessing satellite data, resampling meteorological data, and using a Two Source Energy Balance model to derive daily ET values. These estimates were compared against EC measurements.
The results demonstrated that incorporating local weather station data significantly improved the accuracy of the Sen-ET estimates, with most sites showing a substantial reduction in root mean square error (RMSE) of daily ET compared to the standard Sen-ET method. For example, at one of the wheat sites, the RMSE was reduced from 0.60 mm to 0.14 mm day-1. On the other hand, one of the tomato sites showed a slight deterioration, with an increase of 0.01 mm day-1 in RMSE when data from a weather station 7 km away was used. However, when a closer weather station at 1.17 km was used, the RMSE was reduced by 0.34 mm day-1, thus demonstrating the importance of employing representative weather data in the model.
This study underscores the contribution of localized meteorological data in refining satellite-based ET models and provides a robust approach for precise ET estimation in agricultural landscapes. The findings have significant implications for improving water resource management and irrigation practices.
How to cite: Rozenstein, O., Kwame Dickson, J., and Tanny, J.: Improving satellite-based actual evapotranspiration estimations using data from local weather stations , EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-7430, https://doi.org/10.5194/egusphere-egu25-7430, 2025.
Leaf Area Index (LAI) and Leaf Chlorophyll Content (LCC) are key vegetation indices for modeling energy and mass exchange between the atmosphere and land surfaces and can therefore be utilized for yield prediction. Consequently, suitable methods have been developed to retrieve LAI and LCC from remotely sensed data. Among these, the inversion of Radiative Transfer Models stands out as a promising approach, as it addresses the issue of limited transferability and minimizes the need for extensive field measurements also accounting for crop variability.
The objective of this study is to assess the applicability of the Soil Canopy Observation of Photochemistry and Energy Fluxes (SCOPE) model for estimating LAI and LCC of potato (Solanum tuberosum L.) using time series of hyperspectral images captured by an uncrewed aerial vehicle. A field experiment was conducted in Belgium from June to October 2024, involving two potato varieties, early and late, subjected to two nitrogen fertilization levels and six different biostimulants. Throughout the crop growth cycle, hyperspectral UAV images were captured biweekly using the Specim AFX10 camera. On the same days, in situ measurements of LAI and LCC were performed. LAI and LCC were estimated using a look-up table (LUT) approach based on the inversion of the SCOPE model. A cost function (norm2 distance) was employed to sort the LUT and identify a set of spectra that minimized the distance between measured reflectance and simulated reflectance in the LUT. The estimated LAI and LCC values were then compared with their corresponding in situ measurements.
Preliminary results indicate that simulated LAI and LCC showed potential for use in designing models to predict measured LAI and LCC (R2=0.26 and R2=0.30, respectively, p<0.001). In conclusion, simulated LAI and LCC correlated well with measured values for the late variety at the beginning of the crop cycle. Future work will focus on extending the analysis to cover the entire season, incorporating remote sensing observations into the parametrization of a crop growth model for yield predictions.
How to cite: Rivosecchi, C., Amar, A., Deligios, P. A., Eeckhout, E., Francioni, M., Haesaert, G., Ledda, L., Mancini, A., and Maes, W. H.: Leaf Area Index and Leaf Chlorophyll Content estimation from hyperspectral imaging using SCOPE model inversion, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-21604, https://doi.org/10.5194/egusphere-egu25-21604, 2025.
Advancements in sensing technology and in machine and deep learning have expanded UAV remote sensing applications in agriculture. Most of these applications rely on supervised techniques, but generalization remains a critical and underexplored challenge. Agricultural datasets often exhibit variability across fields, sensors, crops and growth stages. While models such as convolutional neural networks (CNNs) perform well when trained on millions of samples, this approach is impractical with UAV-based agricultural data. This suggests that a location-specific, unsupervised approach might be more effective.
This study proposes a generally applicable method to map weed densities in row crops using high resolution RGB UAV data. The workflow first starts with a vegetation masking based on the Excess Green index, followed by a novel row detection model that separates intra- and interrow vegetation. Pseudo-labels generated from this step are used to train the CNN segmentation model Deeplabv3.
The method was applied on 12 maize datasets collected across multiple locations in Belgium, at different growth stages, and using three different UAV cameras, leading to ranges in ground sampling distance (GSD). The model was also applied on a public sugar beet dataset, PhenoBench, covering 3 dates was used to validate the model. Model performance was evaluated against manually annotated ground truth segmentation maps from each field (n = 50).
Semantic segmentation of crops achieved consistent mean Intersection over Union (IoU) values, exceeding 0.7 (F1-score > 0.89). Weed detection performance was relatively low in very early growth stages (IoU>0.4, F1-score > 0.6) due to limited plant sizes, but improved as weeds grew, with IoU reaching 0.63 (F1-score = 0.83) in later stages. The model was equally performant on maize and on sugar beet.
Despite these early-season limitations, the lower weed detection accuracy had minimal impact on field-level weed density maps, which are primarily used for relative density comparisons to guide site-specific herbicide applications. Regression analyses of predicted crop and weed areas against ground truth annotations showed strong linear relationships. Early-season datasets exhibited slight underestimates of weed area, whereas later-season datasets demonstrated a near-perfect 1:1 relationship (R² > 0.80). GSD proved to be a reciprocal indicator of accuracy, with the highest accuracy at GSDs below 1mm/pixel. GSD above 3 mm/pixel showed a rapid decrease in accuracy.
Overall, the proposed approach effectively generates accurate field-level weed density maps, offering a robust tool for precision weed management in agriculture.
How to cite: Van Poucke, A., Verwaeren, J., and Maes, W.: Generally applicable method for unsupervised weed detection in row crops using UAV-based high-resolution RGB imagery, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-15760, https://doi.org/10.5194/egusphere-egu25-15760, 2025.
The invasive tomato pest Tuta absoluta (Meyrik) poses a significant threat to global agriculture, often resulting in severe yield losses if not detected and managed early. This study investigates the application of artificial intelligence (AI) to develop an automated system for detecting T. absoluta (Meyrik) lesions on tomato plants. Leveraging open-source computational tools such as Google Colab, the research aims to provide an accessible and efficient solution through computational experiments, without requiring field trials.
A curated dataset of tomato plant images is prepared for training and evaluation. The YOLO (You Only Look Once) model is utilized for its proven effectiveness in small-object detection tasks, making it an ideal choice for identifying pest lesions. Model performance is assessed using metrics such as mean Average Precision (mAP), precision, recall, and F1-score, ensuring robust and reliable results across varying conditions. Prior research has highlighted the success of similar AI-based approaches in agricultural pest detection, achieving high accuracy while supporting sustainable farming practices
This work emphasises leveraging multi-source data and advanced modelling approaches to enhance agricultural sustainability. By integrating sensing data and AI techniques, the study supports improved Integrated Pest Management (IPM) strategies, offering a scalable and environmentally friendly solution for pest monitoring in tomato production. Furthermore, the approach demonstrates how AI-driven insights from remote sensing can contribute to the broader goals of ecosystem productivity and nature-based solutions for climate change mitigation.
Acknowledgements: The authors acknowledge the support of the Project “LIFE23-CCA-ES-LIFE ACCLIMATE: Cultivating Resilience: Climate Change Adaptation Strategies for Greenhouses to Enhance Yield and Resource Efficiency from the Programme for the Environment and Climate Action (LIFE-EU) (project number: 101157315).
How to cite: Almeida-Ñauñay, A. F., Sanz, E., Martín-Sotoca, J. J., Moratiel, R., Hernández-Montes, E., and Tarquis, A. M.: Automated detection of tuta absoluta (Meyrik) lesions on tomato plants using artificial intelligence, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-17615, https://doi.org/10.5194/egusphere-egu25-17615, 2025.
Light Detection and Ranging (LiDAR) in precision agriculture is gaining traction as the technology becomes both accessible and affordable, particularly for assessing biophysical characteristics of vegetation. This study investigates the potential of unmanned aerial vehicle (UAV)-based LiDAR data for modeling Leaf Area Index (LAI), a key indicator of crop health and productivity. We explore laser penetration indices to model LAI and compare these results with machine learning models using various LiDAR return types (e.g., ground, vegetation, first, last). In both approaches, in-situ LAI measurements obtained with a LiCOR LAI-2000 were used as ground truth. The study was conducted over two years with a multi-date planting of corn (Zea mays L.) in Temple, TX. Our findings indicate that LiDAR-based methods, both through penetration indices and machine learning, hold promise for accurately modeling LAI and other biophysical crop traits in precision agriculture.
How to cite: Flynn, K. C., Baath, G., Sapkota, B. R., and Smith, D. R.: LiDAR-based indices and machine learning efforts to model biophysical estimations of corn (Zea mays L.), EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-2602, https://doi.org/10.5194/egusphere-egu25-2602, 2025.
Winter and spring wheat are among key agricultural crops in the Republic of Armenia, and represent a significant share of grain production. However, their yield is threatened to substantially decline due to the negative impact of various biotic factors, including weeds and phytopathogens such as rust, powdery mildew, and tan spot. Remote sensing methods, particularly multitemporal dynamics of plant spectral imagery, offer opportunities for early detection and monitoring of these diseases. Early identification allows for timely management interventions to stabilize crop conditions, preserve yields, and enable mapping of problem areas before scheduled applications, allowing more effectively application of herbicides and fungicides.
Hyperspectral spectrometry of winter wheat crops under increased pathogen stress, together with control plots without increased pathogen stress, were studied in experimental fields in southern Russia (Krasnodar Krai) between 2017-2023. The results show that the temporal dynamics in reflectance during the spring-summer growth period of winter wheat likely indicate disease levels, where the period between stem elongation and heading was identified as crucial. A series of high-frequency spectral measurements (every 2–3 days) allowed the classification of areas with infected and healthy plants (accuracy of 70–88%) but also reasonably accurate predictions of the maximum development stage of various pathogens (R² = 0.48–0.55) 10–12 days before peak development. Moreover, these patterns were confirmed using data from ground-based spectrometry, UAVs, and satellite imagery.
Additionally, this methodology was tested on spring wheat fields in the Republic of Armenia (Aragatsotn, Nerkin Sasnashen) in 2024. Using a series of multitemporal UAV surveys, the fields were divided into zones based on the temporal behavior of spectral imagery that successfully identifies zones of weed emergence and negative consequences of agronomic errors. However, identification of more sensitive spectral regions with pathogen hotspots was hindered by the high heterogeneity of the fields.
Based on these methodologies, we defined the optimal dates for initiating phytosanitary monitoring for different regions in Armenia. This part of the investigation shows that zoning territories by the timing of the phenophase "stem elongation" with an error <10 days is crucial for the start of intensive spectral monitoring, and can be achieved by combining NDVI data with meteorological and topographical parameters.
Altogether, the results demonstrate the early diagnosis of biotic stress in plants is feasible using spectral data and can improve decision-making for field treatments in the long term. The early detection of biotic stress in plants enhances the potential of precision agriculture, as time is a crucial factor in addressing these challenges. Furthermore, the described methods have shown the capability to be scaled from local experiments, as is currently the case in most studies, to a regional scale.
How to cite: Sereda, I., Medvedev, A., Ayvazyan, G., and Asmaryan, S.: Using Remote Sensing Spectral Image Dynamics for early prediction of biotic stress in wheat: lessons from Armenia and southern Russia, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-8765, https://doi.org/10.5194/egusphere-egu25-8765, 2025.
Monitoring forests in hard-to-reach locations and under extreme climatic conditions requires reliable, long-term data collection systems. Low-cost devices are increasingly being developed for this purpose; however, deploying these systems without thorough characterisation and calibration can compromise data quality. This work emphasises the importance of fully characterising and calibrating such systems prior to installation to ensure accuracy and reliability over extended periods. This study was conducted as part of the RemoTrees project, which aims to develop a unique IoT tree monitoring system equipped with satellite communication and designed to withstand extreme environmental conditions. A set of the alpha version prototypes, developed within the project, was evaluated in this work. The evaluation focused mainly on a set of low-cost environmental monitoring devices equipped with radiometric sensors measurements. The key performance parameters were assessed, including signal-to-noise ratio (SNR), irradiance sensor detector nonlinearity, sensitivity to temperature variations, and angular response influenced by the diffusive optics. Each parameter was analysed to determine system performance under close to real-world conditions, using both laboratory and in situ validation setups. Key findings revealed that without proper optics used the accuracy of irradiance measurements are significantly influenced. Improvements on the system design and on calibration procedures were implemented to address these issues, improving the overall accuracy and stability of the systems. By addressing these challenges, the systems demonstrated enhanced robustness and suitability for long-term environmental monitoring in extreme conditions. This study underscores the necessity of rigorous pre-deployment testing and calibration for low-cost monitoring devices, particularly when deployed in challenging environments. The findings contribute to advancing the development and deployment of cost-effective technologies for environmental monitoring, enabling more sustainable and accessible data collection practices in forests under extreme climatic conditions.
How to cite: Mihai, L., Toma, C., Mihalcea, R., Sakowska, K., Vescovo, L., Marchesini, L. B., Coppola, V., Renzi, F., and Valentini, R.: Characterisation and Calibration of Low-Cost IoT Monitoring Systems for Extreme Environmental Conditions , EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-4248, https://doi.org/10.5194/egusphere-egu25-4248, 2025.
The European Union’s Carbon Removal Certification Framework (CRCF) establishes robust quality standards and transparent monitoring, reporting, and verification (MRV) systems to ensure the credibility of carbon removal initiatives. Reliable MRV systems are critical for maintaining the environmental integrity of European carbon farming efforts and building stakeholder confidence in carbon accounting and reporting. Achieving these objectives requires the integration of innovative technologies with traditional methods to enhance accuracy and scalability carbon stock estimations.
Within this framework, growing attention is being directed toward methodologies for estimating carbon stocks across various pools in agroecosystems. While soil carbon estimation methods are well-established, the estimation of above-ground biomass (AGB) in agroforestry systems remains underexplored. Significant challenges in this domain include the difficulty of conducting destructive sampling in productive agricultural systems, the lack of species-specific allometric equations for woody crops, and the variability in tree structure introduced by pruning and other anthropogenic interventions.
This study applies terrestrial laser scanning (TLS) in a plum (Prunus domestica L.) orchard to address these challenges and perform non-destructive sampling of AGB for carbon stock assessment. The research employs quantitative structure modeling (QSM) to estimate tree volume and AGB with high precision, demonstrating TLS's ability to overcome limitations associated with destructive sampling, offering a scalable and repeatable approach for accurate biomass estimation in agroforestry systems. Furthermore, the study highlights the role of agroforestry in carbon sequestration efforts.
The findings highlight TLS as a valuable tool for improving the precision and reliability of carbon accounting in agroforestry systems. Its ability to provide accurate, non-destructive AGB estimates supports the effective implementation of the CRCF and advances the EU’s climate goals. Moreover, the scalability and adaptability of TLS make it a promising addition to MRV frameworks, offering stakeholders practical solutions for enhancing carbon removal initiatives.
How to cite: Leoni, M., Chiriacò, M. V., Castaldi, S., and Valentini, R.: Innovative Approaches to Carbon Stock Assessment in Agroecosystems: The Potential of TLS, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-18174, https://doi.org/10.5194/egusphere-egu25-18174, 2025.
Selecting the appropriate unmanned aerial vehicle flight height is beneficial for increasing the monitoring efficiency. We firstly used an unmanned aerial vehicle to explore the scale effect on monitoring rice aboveground biomass. The results confirmed the feasibility of using vegetation indices and textures from hyperspectral images to improve the estimations at different spatial resolutions. The monitoring accuracy of combining vegetation indices and textures was the highest, and exhibited a decreasing trend as the spatial resolution decreased with the greatest accuracy appearing at 13 cm. Two new concepts were proposed: “appropriate monitoring scale domain” to define the range of spatial resolution where the monitoring accuracy was less affected by scale effect, and “appropriate monitoring scale threshold” to define the spatial resolution where accuracy dropped noticeably. The appropriate monitoring scale domains varied at different growth stages and the appropriate monitoring scale thresholds of using vegetation indices and textures were lower than those using textures: 39 cm, 52 cm, and 65 cm at the pre-heading, post-heading, and entire growth stages, respectively when using textures, and 52 cm, 65 cm, and 78 cm at the corresponding growth stages when combining vegetation indices and textures. In terms of aboveground biomass level, growth stage and error value, the relatively lower aboveground biomass levels, earlier growth stages of the multi-temporal models, and overestimations were more likely to yield notable accuracy changes when the spatial resolution converted to lower level on both sides of appropriate monitoring scale threshold. Vegetation indices containing red-edge or near-infrared bands were effective for estimation. Yellow/green band textures and vegetation indices containing green bands with near-infrared/red-edge bands also obtained inspiring performances. MEA was indispensable in estimation while more diverse textures were incorporated into the models of the entire growth stages and models established at lower spatial resolutions. These findings are essential for understanding the scale effect in estimating rice aboveground biomass, facilitating efficient monitoring at field scale.
How to cite: Xu, T., Wang, F., and Shi, Z.: Multi-scale monitoring of rice aboveground biomass by combining spectral and textural information from UAV hyperspectral images, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-2093, https://doi.org/10.5194/egusphere-egu25-2093, 2025.
Monitoring urban, peri-urban, and rural temperatures, along with greenhouse gas (GHG) emissions, is crucial for understanding local climate dynamics, especially in rapidly urbanizing areas. This study leverages advanced remote sensing techniques and environmental analysis to enhance high-resolution Land Surface Temperature (LST) mapping. It further investigates the relationship between LST and methane (CH₄) emissions - a significant driver of climate change - and their combined impact on Urban Heat Island (UHI) effects.
Leveraging multispectral atmospherically corrected imagery from LANDSAT 8-9 and SENTINEL-2 satellites, spectral harmonization techniques and Convolutional Neural Network (CNN)-based super-resolution models were applied to improve the spatial resolution and accuracy of LST calculation. These methods are further refined through the integration of key environmental indices, including soil characteristics, land cover, Normalized Difference Vegetation Index (NDVI), Normalized Difference Built-up Index (NDBI), and Normalized Difference Water Index (NDWI), which capture land use characteristics and their impact on thermal variations. The resultant LST at 1m was statistically validated against meteorological datasets by calculating Root Mean Squared Error and Mean Absolute Error, showing errors consistently below 2°C, with 75% of the values within 1°C. Making use of the accurate LST readings, air temperature (Ta) was derived using polynomial regression models, ultimately resulting in LST-derived air temperature maps with R² values exceeding 0.75.
Building upon this high-resolution thermal mapping, the study examines how agricultural zones are influenced by urban thermal dynamics exacerbated by GHG emissions creating a negative feedback loop where increased temperatures further impact agricultural practices and lead to additional GHG emissions. Seasonal and phenological variations in CH₄ emissions from major crops cultivated in the Loiret region including wheat, were analyzed. Results reveal that land use, crop phenology and soil characteristics significantly modulate LST, influencing both the intensity and distribution of urban heat anomalies. Moreover, the thermal contributions of these areas are analyzed within the context of their dual role. On one hand, these areas can act as potential moderators of UHIs by providing vegetative cover and cooling effects. On the other hand, they contribute to regional methane fluxes due to agricultural practices. This dual role highlights the complexity of peri-urban and rural zones, as they can simultaneously alleviate and exacerbate environmental challenges.
The presented framework can be considered as a contribution to bridging the gap between remote sensing advancements and climate science by providing actionable insights into the interactions between urban and rural thermal dynamics. The methodology not only offers a scalable approach for improving LST and Ta monitoring in data-sparse regions but also highlights the implications of land management practices for mitigating urban heat and reducing GHG emissions. By combining cutting-edge data processing techniques with environmental analysis, the study underscores the importance of integrating thermal mapping with greenhouse gas emission assessments to inform sustainable planning and climate adaptation strategies. In conclusion, this study contributes to the broader understanding of urban-rural thermal interdependencies and their role in shaping regional climate resilience, while also aiming to develop a new approach that leverages remote sensing to GHG emissions across wide areas.
How to cite: El Beyrouthy, N., Al Sayah, M., Der Sarkissian, R., and Nedjai, R.: Integrating High-Resolution Thermal Mapping and Greenhouse Gas Emission Analysis for Climate Resilience in Urban, Peri-Urban and Rural Areas, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-670, https://doi.org/10.5194/egusphere-egu25-670, 2025.
Traditional approaches for crop type classification from optical satellite images typically evaluate algorithms using training and test datasets from the same year and based on calendar days. However, this experimental setup is not practical for real-world applications due to (i) year-to-year variations in crop growth caused by climate, which limit generalization, and (ii) the inability to apply a model to the current year if trained on current-year data. This work addresses these challenges by introducing a cross-year experimental setting and incorporating thermal calendars into our deep learning model. Specifically, we train an attention-based deep learning model on the 2021 Swiss crop dataset, validate it in 2022, and test it in 2023. Thermal calendars, derived from accumulated daily average temperatures, align crop growth with thermal time instead of calendar time, addressing temporal shifts caused by climatic variations. Our results demonstrate that integrating thermal calendars improves performance compared to baseline using standard calendar encodings, achieving better generalization across years and showcasing the potential for large-scale operational crop classification.
How to cite: Turkoglu, M. O. and Aasen, H.: Cross-Year Crop Mapping with Thermal Calendar from Optical Satellite Image Time Series, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-2532, https://doi.org/10.5194/egusphere-egu25-2532, 2025.
Process-based agricultural system models (PBMs) are pivotal tools for evaluating the environmental impacts of agricultural practices. However, their large-scale application is constrained by significant computational demands, extensive time requirements, and data availability. These challenges hinder policymakers and land managers in implementing sustainable agricultural practices at scales meaningful for decision-making. Recent advancements in machine learning (ML) offer a promising solution by providing computationally efficient alternatives, yet the lack of interpretability regarding agro-environmental processes remains a critical barrier.
In this study, we address this challenge by developing a machine learning-based surrogate model for LandscapeDNDC (LDNDC) framework. The surrogate model predicts key agro-environmental variables, including yield, nitrous oxide (N2O) emissions, nitrate leaching (NO3-), and soil organic carbon (SOC), at a national scale for Denmark. Synthetic data were generated using a factorial design based on observed crop practices in Denmark, utilizing field-level data collected across six Danish catchments between 2013 and 2019 as part of the National Monitoring Program for Water Environment and Nature (NOVANA; LOOP-program). This approach incorporated crop rotations as well as spatially disaggregated information on soils and weather, resulting in a dataset comprising approximately 2 billion rows. To enhance the dataset's versatility and account for potential future scenarios, factors like manure amount and synthetic fertilizer amount were extrapolated beyond its current observed ranges. The synthetic dataset was subsequently simulated using the LDNDC modelling framework, and the resulting outputs were employed to train a variety of machine learning algorithms utilizing multi-task learning, optimizing predictions for multiple agro-environmental variables of interest.
Our results demonstrate that the ML-based surrogate model not only significantly reduces computational cost and processing time but also enables the exploration of multiple cropping scenarios with greater efficiency. This approach facilitates rapid scenario testing and optimization, making it accessible to policymakers and farmers without the constraints imposed by traditional PBM frameworks. We propose this methodology as a scalable and practical tool for advancing sustainable agricultural decision-making.
How to cite: Aderele, M. O., Haas, E., Butterbach-Bahl, K., and Rahimi, J.: A Machine Learning-based Surrogate Model for Optimization of Cropping Systems in Denmark, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-4513, https://doi.org/10.5194/egusphere-egu25-4513, 2025.
Effectively tracking drought effects using satellite data can be conducted by combining atmospheric data with additional information of vegetation indices (VIs) from optical data. While VIs detect drought when plant damage is often irreversible, information about the plant physiological status can help detect drought effects much earlier. Remotely-sensed solar-induced chlorophyll fluorescence (SIF), emitted directly from the photosynthetic apparatus (Drusch et al., 2017), provides such information. When abiotic stress occurs due to an increased dissipation of thermal energy through the process of non-photochemical quenching (NPQ), the fluorescence yield is decreased, which can be measured as SIF (Berger et al., 2022, Damm et al., 2018).
Top of canopy (TOC) SIF is available from Sentinel-5P’s TROPOMI sensor since 2018 (Guanter et al., 2021, Köhler et al., 2018). This data, however, is affected by incoming radiation and canopy structure. These effects need to be removed In order to calculate the fluorescence yield in form of the quantum efficiency at leaf level (ΦF), which provides the pure information on the actual physiological status of the plant. Equation (1) uses the vegetation index NIRv (NDVI*NIR (Badgley et al., 2017)) to serve as a combined proxy of the fraction of absorbed photosynthetically active radiation (fAPAR) and the fluorescence escape probability (fesc) (Dechant et al. 2020, Liu et al. 2023). Both SIF data at 743 nm and the reflectance used to calculate the NIRv come from TROPOMI, while the photosynthetically active radiation (PAR) is provided by MODIS.
ΦF = π*SIF743canopy/(NIRv*PAR) (1)
This study presents a new multi-year (2018-2023) ΦF dataset at 0.05° resolution covering Germany with daily temporal resolution. To assess ΦF’s potential as an early drought stress indicator for agricultural and forest ecosystems, it is compared to the anomaly of subsurface water storage (sss), which serves a reference parameter for plant water availability generated by combining the hydrological model PARFLOW and common land model (CLM) (Belleflamme et al., 2023). ΦF and sss anomaly data were split into periods of prolonged negative sss anomaly indicating drought events (cross-referenced as watch/warning periods using the Combined Drought Indicator (European Commission)). Cross-correlation coefficients for different time lags were calculated to compare the datasets. The data was spatially aggregated daily and temporally averaged using a two-day rolling average.
Results show that cross-correlation coefficients for ΦF and sss anomaly are highest at a 2-day lag, dropping again after 3 days, indicating that ΦF follows the negative sss anomaly trend with a 2-day delay in both agricultural and forest ecosystems. Non-normalized canopy SIF and vegetation indices (NIRv, NDVI) showed no pattern and low cross-correlation coefficients during the observed periods. Our findings prove that ΦF has the ability to detect insufficient plant water availability and thus can be used for early drought stress detection in agricultural and forest ecosystems. The comparison of the capabilities of ΦF and TOC SIF to track short-term changes in subsurface water storage illustrates that a proper downscaling and normalization of canopy SIF is essential to use SIF satellite measurements for the early detection of drought events.
How to cite: Herrera, D., Rascher, U., Belleflamme, A., and Siegmann, B.: On the Potential of a Novel Satellite-Based Time-Series of Normalized Far-Red Solar-Induced Fluorescence to Track Short-Term Changes in Subsurface Water Storage, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-15038, https://doi.org/10.5194/egusphere-egu25-15038, 2025.
Air pollution, particularly surface ozone, has become a significant threat to agriculture in China, severely impacting the productivity of essential staple crops like winter wheat. However, the spatiotemporal variability of ozone concentrations and its interactions with other environmental factors—such as temperature and droughts—remain inadequately understood regarding their impact on agricultural productivity. To address this gap in knowledge, this study integrates multi-source remote sensing data with advanced statistical analysis and machine learning techniques to quantitatively examine the spatiotemporal variation of ozone pollution and its interactions with climate change and other environmental factors on winter wheat productivity.
The study first employs the Geographically and Temporally Weighted Regression (GTWR) model, utilizing high-resolution remote sensing data from 2013 to 2019, to assess the spatiotemporal response of winter wheat productivity to ozone pollution. To further investigate the interactions between ozone and other environmental factors, an interpretable machine learning framework is applied, specifically using the eXtreme Gradient Boosting (XGBoost) algorithm augmented by SHapley Additive exPlanations (SHAP). Additionally, a structural equation model is developed to elucidate the underlying mechanisms of these interactions. The results indicate that the negative impact of surface ozone on winter wheat has intensified annually, with significant spatial variation. Particularly in high-pollution areas, such as eastern Henan and northern Anhui provinces, the effects of ozone on winter wheat are most pronounced. Furthermore, the study reveals that the impact of ozone on winter wheat productivity varies across different growth stages, with the most severe effects observed during the later stages in May. Additionally, the research reveals the complex interactions between ozone and other environmental factors, such as temperature and aerosol concentration. Notably, the harmful effects of ozone are exacerbated under conditions of high aerosol concentration and elevated temperatures. Interestingly, drought conditions were found to partially mitigate the negative impact of ozone on productivity.
This study provides a systematic and actionable analytical framework for quantitatively evaluating the effects of ozone pollution and its interactions with climate change and other environmental factors on crop productivity. The findings underscore the need for targeted agricultural measures and pollution control strategies, particularly in high-pollution regions and during critical growth stages. These results provide theoretical support for sustainable agricultural development and climate adaptation management. Furthermore, the study contributes valuable insights into the application of remote sensing technology for large-scale agricultural monitoring, thereby enhancing the management efficiency and adaptive capacity of agricultural ecosystems in response to environmental challenges.
How to cite: Du, C.: Evaluating Air Pollution Impacts on Agricultural Productivity in China: Insights from Remote Sensing Data and Geospatial Modeling, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-14198, https://doi.org/10.5194/egusphere-egu25-14198, 2025.
Remote sensing vegetation indices play a vital role in agricultural zoning by providing detailed insights into crop health, productivity, and environmental conditions. They enable researchers and professionals to monitor environmental changes, urban expansion, and natural events with exceptional accuracy and precision. This progress has been fueled by major technological developments in satellite sensors, data processing algorithms, and analytical methods, enabling the capture of more detailed information and increased observation frequency across expansive regions. Despite these excellent opportunities, numerous image processing techniques have been suggested, each customized for particular applications, datasets, and user needs, yet no widely recognized standard methods have been established. This absence of standardization creates difficulties of interoperability, reproducibility, and consistency in analytical results. Researchers and practitioners frequently encounter challenges choosing the most suitable methods, since the effectiveness of these techniques can fluctuate based on factors like spatial resolution, temporal frequency, and the type of landscape under examination. As a result, there is an increasing demand for the creation of thorough guidelines and uniform procedures that can facilitate the use of remote sensing instruments while ensuring dependable and comparable outcomes across various studies and fields. In this research, we analyze zonation outcomes obtained from remote sensing images captured at different times, using several vegetation indices and applying various clustering techniques. The objective is to evaluate how time-related changes, the selection of vegetation indices, and the use of different clustering methods affect the precision and dependability of land classification. Through the examination of these combination performance, this comparative examination underscores both the advantages and drawbacks of each approach while offering important insights for improving classification methods in varied and changing environments.
How to cite: Hasanli, G., Emamalizadeh, S., Mazzoleni, R., and Baroni, G.: Comparison of zonation approaches by means of remote sensing vegetation indices for agricultural applications , EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-12606, https://doi.org/10.5194/egusphere-egu25-12606, 2025.
In recent decades, space-borne hyperspectral sensors have demonstrated significant potential for agricultural monitoring by providing rich spectral information, improved feasibility, and cost-effectiveness compared to multispectral satellite imagery. In this study, we investigated the consistency of two hyperspectral satellite sensors, PRISMA and EnMAP, for agricultural mapping during the 2025 growing season in the Meknes region: one of the most fertile and productive areas for cereals and vegetables at the national level of Morocco. The primary objective was to conduct a comparative analysis of the two datasets and perform a binary classification (crop vs. no-crop) to support land use monitoring, inform decision-making, and enable advanced crop type mapping.
Our methodology included a correlation analysis of reflectance values across the visible to near-infrared (VNIR) and shortwave infrared (SWIR) ranges, as well as the evaluation of NDVI indices using two methods: band averaging and hyperspectral NDVI (hNDVI). Classification was performed using three machine learning algorithms—Support Vector Machine (SVM), Random Forest (RF), and CatBoost—based on 16 optimal hyperspectral narrow-bands (i.e., 427, 535, 567, 714, 775, 805, 839, 913, 977, 1175, 1246, 1295, 1717, 2077, 2191, 2343 nm) from PRISMA and EnMAP that best capture the variability of vegetation biophysical and biochemical characteristics.
Results demonstrated high Pearson correlation coefficients between the two sensors, with r=0.93 in the VNIR and r=0.91 in the SWIR ranges. NDVI comparison also showed strong consistency results, with correlations of r=0.84 using the hNDVI method and r=0.85 using band averaging. The utilization of optimal hyperspectral narrow-bands achieved superior classification accuracies of 99.95% with PRISMA and 99.65% with EnMAP, with SVM outperforming other algorithms, followed by RF and CatBoost. Moreover, an Explainable Artificial Intelligence (XAI) based analysis indicated that bands in the NIR and SWIR regions were the most critical features driving these high classification performances.
These findings highlight the consistency and complementarity of PRISMA and EnMAP for agricultural monitoring. They also demonstrate the potential of these sensors to produce hyperspectral time-series essential for tracking crop phenology and enhancing crop type mapping, thereby overcoming the constraints posed by limited revisit intervals in current imaging spectroscopy missions.
How to cite: Bourriz, M., Laamrani, A., El-Battay, A., Hajji, H., Elbouanani, N., Ait Abdelali, H., Bourzeix, F., Amazirh, A., and Chehbouni, A.: An Intercomparison of Two Satellite-Based Hyperspectral Imagery (PRISMA & EnMAP) for Agricultural Mapping: Potential of these sensors to produce hyperspectral time-series essential for tracking crop phenology and enhancing crop type mapping, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-18417, https://doi.org/10.5194/egusphere-egu25-18417, 2025.
This study presents a novel deep-learning approach for estimating Soil Water Content (SWC) with high spatial resolution across multiple soil depths. Additionally, the study identifies critical field points based on their drying-out times analyzed by SWC estimations over extended periods. Understanding potential critical points regarding SWC allows operators of heavy agricultural equipment to gain insight into the field's traits and prevent excessive soil compaction. Additionally, this information can support more strategic and efficient harvesting plans by accounting for the impact of varying drying patterns on crop growth and soil strength to not only minimize soil degradation but also maximize yield production, offering a more productive and sustainable crop production.
In this regard, our proposed method offers a practical approach to integrating diverse data types, including:
- Spatial data: remote sensing data (Synthetic Aperture Radar (SAR) and vegetation index), land elevation, and soil profiles at various depths (soil content and bulk density).
- Temporal data: historical weather information (precipitation, temperature, wind, and global radiation).
- Contextual data: date, groundwater level, and crop type.
Previous machine learning and numerical models primarily used temporal and contextual data alongside point-based parameter values as inputs. In contrast, we incorporated spatial information instead of point values, allowing the model to capture better the surrounding influences—such as elevation, water flow, and vegetation shadows—on SWC.
To be able to estimate the SWC using the comprehensive analysis of spatial, temporal, and relevant contextual factors, these inputs are processed by a novel multi-model deep learning framework comprising:
- U-Net to capture spatial features and the impacts of 2D image data.
- Temporal Convolutional Network (TCN) to extract temporal dependencies from weather data.
- Feed-Forward Network (FFN) to model the influence of contextual inputs.
Our model is trained and validated using ground truth data from site measurements in the HOBE dataset. These measurements are conducted at 30 locations within the Skjern River Catchment in Western Denmark, with each data sample containing SWC at different depths: surface, 20cm, and 50cm. By utilizing data collected between 2014 and 2018 from point 1.09 in the HOBE dataset, we demonstrated that the proposed model achieved a mean absolute error (MAE) of 0.0207. For comparison, a numerical model (Daisy) and a machine learning approach that did not account for spatial context produced higher MAEs of 0.0382 and 0.0269, respectively.
Subsequently, the developed model is employed to estimate SWC over extended periods and identify critical points within fields. To achieve this, we collaborated with several farmers who manually classified their field maps into regular, late-drying, and critical parts. The distinction between the latter two categories is crucial, as our observations revealed that "not every wet point is a critical point." The collected temporal SWC data is analyzed with land elevation to differentiate between these two classes. This aspect of the study remains under investigation, and further research is being conducted to refine the classification process and validate its effectiveness.
How to cite: Rafiei, M., Asif, M. R., Nørremark, M., and Sørensen, C. A. G.: Temporal and Spatial Analysis of Critical Field Points Using High-Resolution Soil Water Content Estimation Employing Remote Sensing and Deep-Learning, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-18129, https://doi.org/10.5194/egusphere-egu25-18129, 2025.
Agriculture is one of the primary consumers of freshwater globally. However, precise data on irrigation water use (IWU) at the regional scale is often lacking, which hampers the development of effective water management plans. This information gap is particularly crucial in water-stressed regions, resulting in significant resource waste. Remote sensing datasets offer a valuable opportunity to monitor irrigation patterns over extended periods at a regional scale. Since irrigation affects both soil moisture (SM) and actual evapotranspiration (ET), increases in SM and ET values following irrigation events can be leveraged to frequently retrieve IWU from remotely sensed data. In this regard, we first developed an irrigation-free soil water model in the root zone to simulate SM dynamics during non-growing periods. We then computed the residuals between the modeled SM and the 9 km root zone SM retrieved from SMAP L3, as well as the residuals between the modeled ET and both 30-m OpenET and 500-m PML, to estimate IWU. We used annual IWU data from Arizona State, USA, in 2017 to examine model performance. The simulated SM by our soil water model showed strong agreement with SMAP, evidenced by R2 = 0.68 and RMSE = 0.015 [mm3/mm3]. The estimated IWU using OpenET closely aligned with benchmark data, showing a bias of -17%. However, IWU retrieved by PML led to a much higher bias of -56%, indicating the deficiency of course-resolution ET products in capturing irrigation signals. We further found that over 97 % of the estimated IWU was attributed to ET rather than SM residuals, which is due to SMAP’s low spatial resolution, which limits its ability to resolve farm-scale irrigation volumes.
How to cite: Khamseh, F. and Danesh-Yazdi, M.: Detecting irrigation amount from the integration of remote sensing data in the soil water model , EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-7950, https://doi.org/10.5194/egusphere-egu25-7950, 2025.
Accurate sensor placement is critical in precision agriculture to collect high-resolution data essential for effective monitoring and decision-making. This study presents a comprehensive methodology for optimizing the spatial placement of sensors, focusing on determining the number of sensors needed and their optimal positions to ensure data quality and adequate area coverage. This methodology addresses the challenges posed by terrain restrictions, cost constraints, and data resolution needs. It is versatile, supporting in-situ monitoring, UAV-based sensing, and soil sampling for applications such as soil health analysis and soil organic carbon prediction models.
In many Research and Innovation Labs (RILs), the resolution of Earth Observation (EO) data, such as Sentinel-5 imagery with a resolution of 5×3 km, is often insufficient for the specific needs of agricultural parcels. To complement EO data, additional information must be gathered using in-situ sensors or UAVs. These additional data collection methods can provide higher resolution and more diverse data types, which are crucial for localized agricultural applications. However, the placement of sensors significantly impacts the quality and adequacy of the collected data. Dense sensor deployment across an entire area is often infeasible due to terrain challenges, budgetary limits, and the specific nature of the data being collected.
The methodology developed to address these challenges combines convex optimization, soft clustering, and cost-minimization techniques. The process begins by analyzing the statistical properties of the dataset, such as maximizing variance and maintaining the mean value, to ensure comprehensive data representation. This approach identifies key locations within the parcel that can adequately describe distributed values, reducing the need for excessive sensor deployment while maintaining data integrity.
For areas with existing spatial maps or datasets, the methodology applies weighted subsampling and soft clustering to identify optimal sensor locations. Weighted distributions prioritize critical areas for data collection, ensuring that key zones receive sufficient coverage. In cases where spatial maps are unavailable, an in-house cost-minimization algorithm guides the placement of sensors or UAVs. This algorithm incorporates factors such as terrain, accessibility, and installation costs to balance logistical constraints with data coverage requirements.
This methodology is compatible with diverse data sources, including EO data, hyperfield data, and in-situ sensor data from IoT networks. For instance, it can leverage data from soil moisture monitoring systems. Additionally, the methodology can guide soil sampling strategies for soil health assessment and serve as input for soil organic carbon prediction models. Its adaptability allows it to meet the needs of various agricultural monitoring applications, ranging from broad-scale field evaluations to detailed soil property studies.
Moreover, it enhances data quality by ensuring optimal sensor placement that captures maximum variability within the monitored area and it reduces costs and improves efficiency by minimizing the number of sensors needed. The approach is scalable and flexible, accommodating parcels of varying sizes and adapting to different data collection requirements and its integration with multiple data sources provides a comprehensive and cost-effective solution for advancing precision agriculture and sustainable resource management.
Acknowledgement:
This research has been funded by European Union’s Horizon Europe research and innovation programme under ScaleAgData project (Grant Agreement No. 101086355).
How to cite: Papachristos, E., Vlachos, M., and Amditis, A.: Sensor Spatial Planning Methodology for Optimal Coverage and Data Accuracy in Agricultural Parcels, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-6071, https://doi.org/10.5194/egusphere-egu25-6071, 2025.
Assessing agricultural production in the context of climate change is a global concern. In the recent decades, variable rate technology (VRT) for agricultural machinery has made it possible to adjust fertiliser rates on-the-go, allowing the within-field crop management. In this context, in order to select the most effective management practices, it is essential to identify the driving factors that determine yield variability, mapping the spatial distribution of these driving factors and to determine the local yield variability potential.
Mapping the homogeneous within-field areas of yield potential is used to define management zones. Remote sensing data provide a practical means of delineating these zones. The crop biophysical variable, cumulative evapotranspiration (ETccum), derived from NDVI time series and climate data, was analysed to evaluate its ability to estimate yield. In the semi-arid conditions of the Spanish Central Plateau, wheat ETccum maps were correlated with yield maps by non-linear regression with an R2 of 88%. ETccum serves as an effective proxy for yield estimation and the statistical analysis to determine the level of homogeneity within the field, the driving factors that determine yield variability, and mapping the spatial distribution of these driving factors. Nevertheless, the observed saturation effect in the biophysical variable highlights limitations that require further analysis.
Additionally, during the wheat season, expected potential yields can fluctuate in response to actual weather conditions. Consequently, updating yield predictions early in the season is critical for informed irrigation and fertilisation management decisions. The ability of ETccum to forecast yields at early phenological stages, such as flag leaf and flowering—key stages for yield formation—is examined. Finally, the stability of spatial variability patterns, compared to those derived from ETccum at maturity, is analysed as an indicator of the spatial distribution of yield drivers.
Acknowledgments: this work was supported by the research project NSBOIL (Horizon, GA 101091246).
How to cite: Pareja-Serrano, E., González-Piqueras, J., and Chanzy, A.: Early prediction of within-field variability wheat productive potential using Sentinel2 satellite data., EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-14999, https://doi.org/10.5194/egusphere-egu25-14999, 2025.
Scalably sensing/estimating local information of environment, management, and crop at the field level is the first step of a “System-of-Systems” solution to quantify field-level agroecosystem dynamics (Guan et al., Earth-Science Reviews, 2023). This sensing effort involves two major and inherently connected tasks: (1) ground truth collection, and (2) cross-scale sensing. Agricultural ground truth is scarce and expensive to collect, however, the need for ground truth data is non-negotiable and should be a major investment with public funding. We have developed “cross-scale sensing” approaches to scale-up “ground truth” collection to large scales. In this talk, we will review our recent progress in using "cross-scale sensing" to accurately estimate critical variables of agroecosystem dynamics, covering management practices (e.g. tillage practice, crop rotation, cover crop adoption, irrigation), environmental conditions (e.g. soil properties), and crop traits and conditions (e.g. LAI, Vmax, phtosynthesis, crop yield). We will also identify current challenges and future opportunities to further advance remote sensing for sustainable and precision agriculture.
How to cite: Guan, K.: Recent progress in remote sensing for advancing sustainable and precision agriculture, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-19065, https://doi.org/10.5194/egusphere-egu25-19065, 2025.
This study explores the impact of climatic variability on the generalization capabilities of a deep learning model for pixel-level crop classification using multi-temporal Sentinel-1 SAR data in Denmark. With agriculture accounting for 61% of Denmark’s land area, accurate and timely crop mapping is essential for providing insights into crop distribution, offering valuable information to advisors and authorities to support large-scale agricultural management, and address challenges posed by changing climatic conditions.
Our study leverages a novel deep learning architecture that combines a 3-D U-Net with a conv-LSTM module to effectively capture both spatial and temporal dependencies in crop growth patterns. We consider 14 crop types over an eight-year period (2017–2024) and growth season (May to August), with ground truth data derived from Denmark’s Land Parcel Identification System (LPIS). Our analyses reveal that climatic variables such as precipitation, temperature, and humidity significantly influence model performance across years. Notably, extreme years like 2018 (characterized by drought and high solar radiation) and 2024 (marked by record precipitation) challenge the model’s ability to generalize effectively. By correlating inter-annual model accuracy trends with climatic data, the study demonstrates the necessity of incorporating environmental context into AI-driven agricultural monitoring systems.
We also evaluate the benefits of training the model on multi-year datasets to enhance robustness against climatic variability. Our findings reveal that leveraging temporal diversity improves model performance but highlights persistent difficulties in generalizing to outlier years with extreme climate conditions. While training on multi-year datasets helps capture a broader range of crop phenological variations, the results underscore that this approach alone is not sufficient, and underscores the importance of integrating auxiliary data, such as local climatic variables, to enable models to better adapt to evolving crop growth patterns influenced by changing environmental conditions.
This work represents one of the most comprehensive evaluations of deep learning for crop classification, spanning eight years and covering over 1.5 million hectares of agricultural land. By linking model performance to climatic variability, this study provides critical insights for improving the generalization capabilities of deep learning models in precision agriculture. These findings not only pave the way for enhanced crop monitoring under diverse climatic scenarios but also emphasize the potential of integrating climate-resilient AI technologies to address global agricultural and environmental challenges.
How to cite: Asif, M. R., Rafiei, M., Jørgensen, R. N., Nørremark, M., and Teimouri, N.: Assessing generalization of deep learning models for crop classification under climatic variability in Denmark, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-9634, https://doi.org/10.5194/egusphere-egu25-9634, 2025.
Accurately modeling the impacts of climate stress on crop growth and yield is crucial for ensuring food security. Data-driven models are increasingly utilized for yield estimation because they can learn effective crop growth features from vast amounts of remote sensing and meteorological data. However, extreme climate stress conditions have few yield labels available for these models to modeling the interaction in crop responses. The response of crops to extreme climate stress often exhibits varied delays which are captured in remote sensing observations. In this study, we explicitly encode the time lag effect quantified by remote sensing and climate stress indicators into a two-stream fusion framework for estimating crop yield under extreme climate stress. Each stream employs a pyramid structure that progressively aggregates remote sensing and climate time series into feature embeddings. A time-lag-encoded cross attention mechanism fuses feature embeddings between the two streams, while phenology-sensitivity-guided linear attention is applied on top of the pyramid structures for processing ultimate time-lag encoded features. The proposed model is evaluated across nine Midwestern states within the US Corn Belt at the county level from 2006 to 2012, simulating climate stress situations with fewer samples. End-of-season results demonstrate that the knowledge-encoded two-stream model (RMSE=1.17 Mg ha-1) outperforms both the feature-stacking-based two-stream model (RMSE=1.43 Mg ha-1) and random forest (RMSE=1.68 Mg ha-1) under extreme climate stress. The improved estimation performance indicates that knowledge-encoded data fusion is more effective than merely stacking multi-source input data. In-season results reveal that our model proficiently captures extreme events and effectively predicts yield 8 weeks in advance. The time-lag knowledge could be extended to other forms of climate stress. Also, cross attention enables integration with additional data sources to enhance the interaction modeling of complex biomass accumulation and yield formation.
How to cite: Xiong, X., Zhong, R., Tian, Q., Athanasiadis, I., and Lin, T.: Knowledge-encoded deep fusion for yield estimation under extreme climate stress, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-16547, https://doi.org/10.5194/egusphere-egu25-16547, 2025.
Mountain pastures are a biodiversity rich and heterogeneous ecosystem of the Alps influenced by a complex topography and a variable climate. Understanding the impact of these factors on pasture productivity is of primary importance for forage production and ecosystem preservation.
We present here a regional analysis covering the alpine pastures in the Grisons Canton (eastern Switzerland, 1997 sq. km), for which we developed a collection of high-resolution (10-m) annual growth indicators based on the Enhanced Vegetation Index (EVI) derived from Sentinel-2 images, from 2016 to 2024. We correlate our growth maps to a 1-km gridded climate dataset (Meteoswiss) and a 10-m digital elevation model (Swisstopo) to understand which weather factors - rainfall, temperature, or radiation - influence the most the growing season and from which period of the year. In addition, we explore the variability of these dependencies in space, in relation to elevation and derived topographic descriptors, e.g. slope or valley orientation.
This analysis shades light on the climate dynamics impacting the most the growing season in conjunction to a complex local topography. The results can be used to identify vulnerability levels along the elevation profile, influenced by soil depth and valley orientation, where growth varies the most from year to year in function of annual weather variations. In these zones, pasture management will need extra flexibility measures and real-time monitoring to adapt to annual fluctuations of a future climate change.
How to cite: Oriani, F., Aasen, H., Schneider, M., and Calanca, P.: What influences alpine pasture productivity? Exploring the relation among topography, climate, and biomass using remote sensing., EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-11964, https://doi.org/10.5194/egusphere-egu25-11964, 2025.
Wheat is one most widely grown and consumed crops globally. Region-wise, the north Indian plains are one of the largest producers of wheat in the world. However, there remains a substantial variation in the sowing dates and thus the phenology of wheat grown in the area owing to variation in cropping patterns, soil type and agricultural practices. In this study, field data including the extent of wheat crops along with their sowing and harvest dates were collected in the Gautam Buddha Nagar district of Uttar Pradesh, India during the 2022-23 crop season. The study region is then classified into croplands and further into wheat and non-wheat areas using the random forest classifier in the Google Earth Engine. On the basis of the sowing dates, the study region is divided into early sowing (sowing date before 10 November 2022) and late sowing (sowing date after 25 November 2022) areas. The phenology of the wheat agricultural fields is analyzed using the normalized difference vegetation index (NDVI) derived using the Sentinel 2 surface reflectance data product available in the Google Earth Engine. Results showed that the early sowing wheat has the largest period (6 to 7 weeks) in which canopy cover was near maximum. The same period for late sown wheat was found as 4 to 5 weeks for late sown wheat. In general, the peak vegetation density for the crop season decreased as the sowing time of the wheat was delayed. The average value of peak normalized difference red-edge index (NDRE) was found as 0.67 (in second week of February 2023) and 0.62 (in first week of March 2023) for the early and late sown wheat, respectively. The lengths of the crop seasons of the early and late sown wheat were found as 140 and 120 days, respectively. The findings of the present study can be extrapolated to understand the phenology as well as the yield patterns of the wheat in one largest wheat producing regions in the world.
How to cite: Upreti, H., Sriyodh, C., and Yadav, M.: Assessment of phenology of winter wheat using Sentinel 2 multispectral data for varying sowing dates , EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-19560, https://doi.org/10.5194/egusphere-egu25-19560, 2025.
In East Africa, lack of agriculture inputs and unstable climates lead to 50% yield gaps, making intercropping—the planting of more than one crop in the same parcel of land—a common agricultural management practice among smallholder farmers to improve land-use efficiency and reduce risks. In Kenya, where maize is the staple food, maize is often intercropped with beans, legumes, and potatoes. Despite its widespread, agricultural statistics on intercropping are currently sparse, and remote sensing approaches for large-scale crop monocultures are often unsuitable for intercropping monitoring. Mapping intercropping at national scale is extremely challenging because of heterogeneous landscapes, lack of cloud-free satellite imagery, and the scarcity of high-quality ground-based situ data in these regions. This study addressed these challenges using a phenology-assisted automated mapping framework on Google Earth Engine (GEE) to create 10m-resolution maps of monoculture and intercropped maize across Kenya for the long and short rainy seasons of 2023.
First, we computed 10-day median composites of Sentinel-2 optical reflectance data for each pixel in the region to build monoculture/intercropped/non-maize Random Forest (RF) classifiers. Several thousand crop ground labels were collected during field surveys in 2023, including monoculture maize (mono-maize), intercropped maize (in-maize), and other crops (e.g., wheat, rice, coffee, tea, sugarcane, potatoes, beans, etc.). To address the limited availability of intercropped maize samples, a novel phenology-based approach was implemented. Maize was first differentiated from other crops by analyzing TCARI and OSAVI during the vegetative phase and ARI during maturity. Additionally, lower greenness and moisture levels in intercropped systems, which have larger planting width and more short-term crops, were detected using the SWIR1/NDVI ratio, effectively distinguishing mono-maize from in-maize. Automatically derived monoculture/intercropped maize samples and 40% of ground samples were used for training, while the remaining ground data were used for accuracy assessment.
For the long rainy season, the overall accuracy (OA) was 0.88, with an F1-score of 0.87 for mono-maize and 0.78 for in-maize. For the short rainy season, OA dropped to 0.85, with F1-scores of 0.82 for mono-maize and 0.72 for in-maize. Misclassification primarily arose from phenological similarities between mono-maize and in-maize and increased planting of other crops with similar patterns during the short rainy season. Results revealed that 854,432 hectares of mono-maize were concentrated in the Western region and Rift Valley plateau during the long rainy season, while 1,061,701 hectares of in-maize were widely distributed across the region, particularly near Mount Kenya and the Eastern region. In the short rainy season, reduced and erratic precipitation led to decreased maize planting, with more farmers opting for intercropped systems and short-term crops to reduce risks of crop failure.
We are convinced that this study is a crucial first step to demonstrate the potential of Sentinel-2 data and phenology-based automated mapping for large-scale monitoring of intercropping, providing critical insights for agricultural monitoring in sub-Saharan Africa. It serves as a foundation for developing a regional archive of monoculture and intercropped crop systems and addressing key agricultural challenges across the region.
How to cite: Chen, Y., Zuo, L., Wei, X., Wang, X., and Xu, J.: Mapping 10-m monoculture and intercropped maize of Kenya with phenology knowledge and Sentinel-2 data, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-4910, https://doi.org/10.5194/egusphere-egu25-4910, 2025.
Precision agriculture (PA) has emerged as a key strategy for optimizing agricultural production. Using data-driven technologies such as sensors and satellite imagery, PA improves the efficiency of agricultural processes. Accurate crop yield estimation is an essential component of PA. An important aspect of yield estimation within PA is the ability to assess and map spatial variations in yield in an agricultural field. Understanding these spatial patterns enables more precise management decisions and targeted interventions.
Therefore, this study aimed to develop two regression approaches, multiple linear regression (MLR) and random forest regression (RFR), to estimate crop yield using sixteen input variables with a 6 m resolution. These variables were obtained using different sensors, reflecting the soil and crop spatial variability. The estimation performance of the studied approaches was assessed using the coefficient of determination (R²), showing very satisfactory results (R² > 0.85) for both approaches.
The spatial distribution of barley yield was assessed, focusing on identifying areas of high and low productivity within the field. RFR demonstrated its ability to capture yield patterns. By incorporating spatial factors, RFR effectively modelled the varying yield potential in the crop field.
Keywords—precision agriculture, multiple linear regression, random forest regression, spatial pattern, barley
Acknowledgments: Financed by Ministerio de Ciencia e Innovación, Spain (PID2021-124041OB-C22)
How to cite: Ksantini, F., Quemada, M., Almeida-Ñauñay, A. F., Sanz, E., and Tarquis, A. M.: Barley Yield Estimation Using Regression Models and Spatial Pattern Analysis, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-9672, https://doi.org/10.5194/egusphere-egu25-9672, 2025.
India is one of the world's leading exporters of wheat grain, making monitoring its growth and yield one of the country's top economic priorities. This study aimed to develop a methodology for delineating wheat cultivation areas and estimating wheat yields using Landsat 8 (30 m spatial resolution) data for the Nainital and Udham Singh Nagar districts of Uttarakhand, India. The cultivated wheat fields were identified using a supervised classification-based Random Forest (RF) algorithm during the growing season from November 2020 to April 2021. To characterize the wheat class, a total of 239 and 226 wheat points, along with 201 and 166 non-wheat geometry points, based on NDVI time series were allotted for Nainital and Udham Singh Nagar districts, respectively. The calculated wheat area was found to be 778.94 sq. km and 209.48 sq. km, compared to the actual reported areas by the Agriculture Department, Government of Uttarakhand of 1059.61 sq. km and 212.78 sq. km for Udham Singh Nagar and Nainital, respectively. The RF algorithm showed an underestimation for both districts, achieving a kappa coefficient of 0.97, producer accuracy of 0.97, user accuracy of 0.96, and overall accuracy of 0.98 for the Nainital district. For the Udham Singh Nagar district, the kappa coefficient was 0.89, with producer accuracy of 0.89, user accuracy of 0.93, and overall accuracy of 0.93. The study also utilized weather data along with Landsat 8 imagery as inputs for the Carnegie-Ames-Stanford Approach (CASA) to estimate wheat yields and get spatial wheat yield maps. The estimated mean yields were 3.73 t ha⁻¹ and 3.37 t ha⁻¹, whereas the actual mean yields were 3.82 t ha⁻¹ and 4.45 t ha⁻¹ for Nainital and Udham Singh Nagar districts, respectively. The study demonstrates the potential of combining remote sensing and supervised classification techniques for reliable wheat yield estimation in data-scarce regions, which can be a promising tool for agricultural policy and decision-making.
Keywords: Crop classification, Landsat 8, random forest, wheat, spatial yield map
How to cite: Singh, P. and Kothari, K.: Remote Sensing-based Wheat Area and Yield Estimation: Insights from Uttarakhand, India, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-9872, https://doi.org/10.5194/egusphere-egu25-9872, 2025.
Pan-European Network of Green Deal Agriculture and Forestry Earth Observation Science (PANGEOS) funded by the European Cooperation in Science and Technology (COST) organisation brings together researchers to share their expertise and bring up a new generation of scientists. In October 2024 PANGEOS conducted an intensive 5-day summer school where more than 20 participants learnt how to propagate uncertainty of spectral measurements to uncertainty in higher-level products. The training material in the form of Python Jupyter notebooks is publicly available on GitHub https://github.com/pangeos-cost/uq-training.
This presentation is going to highlight the steps of uncertainty propagation from ground measurements through vegetation indices and retrieved plant traits towards higher-level model estimates, like gross primary productivity and evapotranspiration. All three pathways of retrieval uncertainty estimation, regression-based (vegetation indices), radiative transfer model-based and hybrid, are demonstrated. In addition, challenges of uncertainty propagation through satellite imagery are discussed in a separate block.
Finally, a highlight of current and future activities of the PANGEOS COST action will be given.
Acknowledgement
This abstract is supported by the EU COST (European Cooperation in Science and Technology) Action CA22136 “Pan-European Network of Green Deal Agriculture and Forestry Earth Observation Science” (PANGEOS).
How to cite: Prikaziuk, E., Llewellyn, G., Mihai, L., Bialek, A., Hueni, A., Werfeli, M., Gomez-Dans, J. L., Verrelst, J., Garcia-Soria, J. L., Fennell, J., Ganeva, D., and Kefauver, S. C.: PANGEOS COST action: Uncertainty propagation in remote sensing , EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-21750, https://doi.org/10.5194/egusphere-egu25-21750, 2025.
The accurate evaluation of crop quality is vital for sustainable agriculture and optimized production. Raman spectroscopy, renowned for its insensitivity to water interference and its ability to deliver molecular-specific information, presents significant potential as a remote sensing technology. This study explores the feasibility of adapting advanced Raman spectroscopy as a remote crop quality sensor for the precise assessment of carotenoids. Carotenoids were chosen due to their dual role as key stress indicators in crops and their well-established antioxidant benefits for human health.
To explore carotenoid variability, Arabidopsis thaliana and Spinacia oleracea were analyzed. Raman spectroscopy measurements were performed on two leaves per plant using a 785 nm laser. For the carotenoid quantification, Linear Discriminant Analysis (LDA) was adapted. The spectra were processed through smoothing, background removal, and normalization, followed by modification with an amplifying factor. This study evaluated the impact of these processing methods, particularly the application of the amplifying factor, on the accuracy of the model. High-Performance Liquid Chromatography (HPLC) was employed as the reference method for validation. Three-quarters of the samples were used to construct the model, while the remaining one-quarter was reserved for validation. As a result, the model utilizing spectra modified with the amplifying factor in most cases achieved higher validation accuracy compared to models based on unmodified spectra.
This study introduces a novel Raman spectroscopy-based remote sensing approach for crop quality assessment, establishing an enhanced model for interpreting spectral data. By enabling precise detection of stress-induced changes in plant chemical composition, including carotenoids, this technique paves the way for scalable, real-time monitoring through Raman-equipped machinery or drones, advancing sustainable agriculture practices.
How to cite: Park, M., Somborn, A., Schlehuber, D., and Keuter, V.: Development of a Remote Crop Quality Sensor: Advancing Carotenoid Assessment with Raman Spectroscopy, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-1883, https://doi.org/10.5194/egusphere-egu25-1883, 2025.
The spectral invariants theory (p-theory) has received much attention in the field of quantitative remote sensing over the past few decades and has been adopted for modeling of canopy solar-induced chlorophyll fluorescence (SIF). However, the spectral invariant properties (SIP) in simple analytical formulas have not been applied for modeling canopy fluorescence anisotropy primarily because they are parameterized in terms of leaf total emissions and scatterings, which precludes the differentiation between forward and backward leaf SIF emissions. In this study, we have developed the canopy-SIP SIF model by combining geometric-optical (GO) theory to account for asymmetric leaf SIF forward and backward emissions at the first-order scattering and by modeling multiple scattering based on the p-theory, thus avoiding the dependence on radiative transfer models. The applicability of the model simulations especially over 3D heterogeneous canopies was improved by incorporating canopy structure through multi-angular clumping index, and by modeling single scattering from the four components of the scene in view according to the GO approach. The results show good consistency with both the state-of-the-art SIF models and multi-angular field SIF observations over grass and chickpea canopies. The coefficient of determination (R²) between the simulated SIF and field measurements was 0.75 (red) and 0.74 (far-red) for chickpea, and 0.65 (both red and far-red) for grass. The average relative error was approximately 3% for 1D homogeneous scenes when comparing the canopy-SIP SIF model simulations to the SCOPE model simulations, and around 4% for the 3D heterogeneous scene when comparing to the LESS model simulations. The results indicate that the proposed approach for separating asymmetric leaf SIF emissions is a robust way to keep a balance between satisfactory simulation accuracy and efficiency. Model simulations suggest that neglecting the leaf SIF asymmetry can lead to an underestimation of canopy red SIF by 16.1% to 43.4% for various canopy structures. This study presents a simple but efficient analytical approach for canopy fluorescence modeling, with potential for large-scale canopy fluorescence simulations.
How to cite: He, Y., Zeng, Y., and Hao, D.: Combining geometric-optical and spectral invariants theories for modeling canopy fluorescence anisotropy, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-4726, https://doi.org/10.5194/egusphere-egu25-4726, 2025.
The snow cover occurrence index (SCOI), deffned as the ratio of the number of times that a pixel is classiffed as snow to the number of times that the pixel is observed in optical remote sensing data over a given year, can effectively mitigate the inffuence of clouds and holds great potential for extracting the annual snow duration and glacier extent in mountainous regions. The SCOI of the Qinghai–Tibet plateau (QTP) is calculated and analyzed on the basis of Landsat images from 1985 to 2021. The results indicate the following: 1) the evaluation based on station snow depth reveals that the SCOI is stable when the number of combined years reaches 5; 2) the SCOI has a strong correlation with snow cover days (SCD) determined from Moderate Resolution Imaging Spectroradiometer (MODIS) snow cover products; and 3) the SCOI has good potential for glacier extraction and exhibits a high level of consistency with glacier boundary survey data. Overall, owing to the higher spatial resolution and longer duration of the Landsat-based SCOI, it can accurately describe the distribution characteristics and changes in snow cover and glaciers in complex mountainous areas.
How to cite: Wang, X.: A Novel Snow Cover Occurrence Index (SCOI) for the Dynamics of Snow Duration and Glacier Extent in Mountainous Regions, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-2205, https://doi.org/10.5194/egusphere-egu25-2205, 2025.
A study conducted in a northern Jordanian arid Mediterranean grassland between 2017 and 2021 examined the relationship between remotely sensed Normalized Difference Vegetation Index (NDVI) and modeled standing crop biomass. The research sought to determine the utility of high-resolution (10-meter) Sentinel-2 imagery, coupled with the PHYGROW model, for biomass estimation in this challenging environment, and to assess the potential of NDVI as a cost-effective alternative to traditional ground-based methods. Data were aggregated into 10-day intervals for temporal analysis. Results indicated a significant positive correlation (p < 0.001) between NDVI and standing crop (kg/ha), described by the linear model: Standing crop = 60.40 + 3567.56 × NDVI (R² = 0.52). This finding suggests that NDVI offers a reliable and time effective approach to biomass estimation in such settings.
The strong positive correlation between NDVI and standing crop highlights the potential of remote sensing for large-scale rangeland health monitoring. Tracking NDVI changes over time provides insight into vegetation responses to climate, grazing, and conservation efforts. This understanding supports decision-making for sustainable grazing, water management, and conservation strategies. Future research should validate these findings on larger scales and explore integrating NDVI with other data, like soil moisture, to refine predictive models and improve accuracy. The study advocates adopting NDVI-based monitoring in arid rangeland management.
How to cite: Alhamad, M. N.: Integrating Sentinel-2 Imagery and PHYGROW Model for Biomass Estimation in Arid Rangelands of Northern Jordan, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-5178, https://doi.org/10.5194/egusphere-egu25-5178, 2025.
Soil moisture is vital for agricultural fields as it determines water availability for crops, directly affecting plant growth and productivity. It regulates nutrient uptake, root development, and microbial activity, ensuring efficient use of fertilizers and soil resources. Proper soil moisture levels prevent water stress, reduce crop failure risks, and optimize water irrigation efficiency. Accurate soil moisture monitoring supports sustainable farming practices, helps mitigate drought impacts, and enhances climate resilience. By maintaining optimal soil moisture, farmers can improve resource use, boost crop yields, and promote long-term agricultural sustainability. This study aims to develop an approach for retrieving soil moisture from Sentinel-1 A Synthetic Aperture Radar (SAR) data. The SAR data were processed for the 2024 dry season using a triangle-based approach in the Mekong Delta, Vietnam, following three main steps: (1) data preprocessing to convert raw radar backscatter values into the sigma naught (σ₀) backscatter coefficient in decibels (dB). This involves radiometric calibration, noise removal, and logarithmic scaling to enhances data interpretability, allowing for better comparisons across different radar acquisitions and improved analysis accuracy, (2) soil moisture retrieval by means of a triangle-based method developed based on the dual-polarization modes of the vertical transmit and vertical receive polarization (VV) and vertical transmit and horizontal receive polarization (VH). This method employs the triangular feature space created by using change in VV backscatter coefficients and the radar vegetation index (RVI), in which RVI helps distinguish vegetation effects while VV backscatter provides information on soil moisture. Combining both parameters thus allows for more precise moisture estimation even in complex environments, and (3) error verification. The results of soil moisture retrieval compared with the reference data showed moderate positive correlation, with the values of correlation coefficient (r) greater than 0.5 and the root mean square error (RMSE) smaller than 0.05, respectively. The lower soil moisture levels were especially observed in coastal areas, where higher evaporation rates, saline intrusion, and limited rainfall contribute to drier soils. These conditions create challenges for agriculture in coastal regions, as crops are more susceptible to drought stress and water shortages. Consequently, managing soil moisture becomes crucial for maintaining crop productivity and ensuring sustainable farming in coastal provinces. Eventually, soil moisture data was spatially aggregated with cropping areas to improve management practices in the region, allowing precise monitoring of soil conditions relative to specific crops and enabling tailored irrigation and water management strategies. This approach, leveraging dual-polarization SAR data with aid of the triangle-based method, could enhance soil moisture monitoring in agriculture and is completely transferable to other regions across the globe for soil moisture monitoring.
How to cite: Son, N.-T., Chen, C.-F., Chen, C.-R., Zhang, Y.-T., Chen, S.-L., and Chen, S.-H.: Soil Moisture Retrieval Over Agricultural Fields Using Synthetic Aperture Radar (SAR) Data, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-5987, https://doi.org/10.5194/egusphere-egu25-5987, 2025.
Natural environments face substantial challenges from human activities related to food, feed, and energy production. Unsustainable nutrient management is a key issue, with excess nutrients leaching into the groundwater cycle or escaping intended cropland through other pollution pathways ending up in the atmosphere or in nearby coastal systems. This nutrient loss depletes soil health, contributes to the climate crisis and impacts water quality, especially when combined with intensive farming practices lacking conservation efforts. Innovative mitigation actions, such as the Nature-based Solutions framework, designed to enhance water quality and advance sustainability in agricultural management, require thorough assessment and monitoring to encourage stakeholder participation in these strategies. Conducting research to explore the extent of their effects is thus essential, with a deeper understanding of the nutrient cycle playing a pivotal role in achieving these goals.
With the cumulatively increasing availability of remote sensing data sources and advancements in machine learning technologies, automating monitoring and assessment efforts has become a hot and important topic. The challenge is to construct transparent and transferable models capable of working with real-time data to accurately predict crop types, crop status or other desired features. The primary goal of this study is to investigate how an automated multisource data analysis approach, with a focus on remotely sensed data, can support the quantification and mapping of sustainability efforts in agricultural crop management while enhancing the understanding of nutrient flow within large-scale agricultural catchments. Centered on the Hjarbæk Fjord in Denmark, the study also aims to assess the transferability of its models across different sites in Europe. This research is part of a broader project investigating the potential of integrating permanent grasslands into crop rotations as a Nature-based Solution in the catchments surrounding Hjarbæk Fjord. The project aims to develop a decision support tool to guide the planning and optimization of grassland implementation in terms of extend and location. This tool is designed to maximize benefits across various parameters, including the number of stakeholders impacted, economic considerations, crop yield, biodiversity, and other critical factors. The output of the current study, involving the training of a deep learning model to predict cropland trends related to grassland implementation, can in turn be integrated as input for the described decision support tool.
This is an explorative study that relies on the availability of accurate ground truth data to train and validate a deep learning model, providing insights into trends associated with the implementation of sustainable management strategies. A key challenge lies in acquiring knowledge of and access to comprehensive datasets that capture relevant parameters, such as actual yield values, quantitative values of nutrients in different stages of the growth season and different nutrient pools within the cropland environment, accurate accounts of management actions and other contributors to the nutrient cycle. Additional challenges involve preprocessing satellite data to establish a robust pipeline for the automated collection of satellite imagery, ensuring a coherent time series. This includes addressing temporal and spatial data gaps through extrapolated estimations to create a consistent dataset.
How to cite: S. Vesterdal, M., Dalgaard, T., and Gislum, R.: Multisource data analysis at the catchment scale to quantify and map sustainable agricultural management practices, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-6243, https://doi.org/10.5194/egusphere-egu25-6243, 2025.
The increasing global demand for food and the pressing need for sustainable agricultural practices have made technological innovations essential in modern agriculture. Satellite imagery, as a cornerstone of precision agriculture (PA), provides valuable tools for monitoring crops and optimizing resource management. This study evaluates the potential of PlanetScope’s (PS) advanced 8-band multispectral sensor (SuperDove) for citrus orchard monitoring. The primary objectives are to investigate the effectiveness of PS data in assessing orchard health and dynamics and to explore its utility in detecting spatial variability within citrus orchards. The methodology involves preprocessing SuperDove data to derive key vegetation indices (VIs), such as NDVI, SAVI, and EVI, which are widely used to gain insights into the vigor and condition of citrus orchards. To assess the reliability and practicality of PS data, the study includes a comparison with free and open-source alternatives, such as Sentinel-2. This research emphasizes the importance of integrating high-resolution satellite imagery into citrus orchard management practices. While still in the early stages, the study aims to provide insights into how advanced satellite data can support sustainable agriculture.
How to cite: Rahali, L., Pratico, S., and Modica, G.: Can SuperDove Multispectral Satellite Data Optimize Citrus Orchard Monitoring?, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-6285, https://doi.org/10.5194/egusphere-egu25-6285, 2025.
Abstract: The universal contamination of arable land with potentially toxic elements (PTEs) poses a severe threat to food security and jeopardizes worldwide efforts to meet the United Nations Sustainable Development Goals (SDGs). How to obtain information on PTEs in regional agricultural soils more reliably is a priority problem to be solved. Multispectral satellite remote sensing, with its advantages of high spatial and temporal resolution, broad coverage, and low cost, offers the potential to acquire distribution information of PTEs over large areas. However, owing to the complexity of soil environments and the insufficiency of spectral information, the mechanism for retrieving concentrations of soil PTEs via multispectral satellites is not yet clear, and the accuracy needs to be improved. In this study, we aimed to assess whether employing a fusion of spectral information and environmental covariates, results in more accurate predictions of PTEs, specifically chromium (Cr) and mercury (Hg), in croplands than does employing spectral information alone. Three machine learning algorithms—kernel-based support vector machine (SVM), neural network-based back propagation neural network (BPNN), and tree-based extreme gradient boosting (XGBoost)—were developed to retrieve soil Cr and Hg concentrations. The results showed that the fusion of spectral information and environmental covariates combined with the XGBoost model performed best in retrieving both Cr and Hg concentrations with coefficient of determination (R2) values of 0.73 and 0.74, respectively. Environmental covariates are important variables for determining Cr and Hg concentrations in agricultural soils, but the ability to retrieve these element concentrations by utilizing multispectral information alone is limited. High Cr concentrations occurred in central towns and southern hilly mountains. High Hg concentrations were detected in the northwestern region and southern hilly mountains. The potential of fusing multispectral data and environmental variables to precisely retrieve soil PTE concentrations can serve as a reference for agricultural information monitoring worldwide.
Keywords: Potentially toxic elements; Sentinel-2; Environmental covariates; Machine learning; Farmland
How to cite: Zha, Y.: Retrieval of chromium and mercury concentrations in agricultural soils: Using spectral information, environmental covariates, or a fusion of both?, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-7525, https://doi.org/10.5194/egusphere-egu25-7525, 2025.
Remote sensing and machine learning techniques enable precise diagnosis of crop growth anomalies, providing an effective means to mitigate production losses caused by disease outbreaks while supporting sustainable agricultural management. This study aims to detect rice diseases using satellite, drone, and weather data in a timely manner. A random forest model for rice disease detection was developed using drone imagery collected in 2023 year, where disease-damaged pixels were classified through K-means clustering, and the corresponding damaged areas were used for rice paddy disease classification model training. This model has been applied to agricultural fields in 2024 year as follows. First, Sentinel-1 and Sentinel-2 satellite data were utilized to classify paddy rice fields, with irrigated areas identified through the normalized difference vegetation index, land surface water index, and VV polarization. Second, the risk of rice disease occurrence was calculated based on air temperature, relative humidity, and precipitation. These variables represent weather conditions that can cause crop diseases. Third, drone measurements were conducted to monitor the abnormal growth of paddy rice when the risk score increased. Fourth, the location of disease outbreaks was detected using the random forest model, which uses surface reflectance at blue, green, red, red-edge, and near-infrared wavelengths as input data. Subsequently, drone spraying operations were carried out to reduce crop damage caused by the identified diseases. These results highlight the potential of agricultural management using remote sensing techniques.
Acknowledgments: This research was funded by RDA, grant number RS-2022-RD010059.
How to cite: Ryu, J.-H., Lee, K.-D., Jeon, Y., Kwak, G.-H., Lee, S.-J., and Choi, L.-Y.: Machine Learning-Based Rice Disease Diagnosis Through Joint Utilization of Satellite, Drone, and Weather Data, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-7922, https://doi.org/10.5194/egusphere-egu25-7922, 2025.
Understanding cropland utilization is essential for improving agricultural productivity and efficiently managing cropland resources. Analyzing region-specific cropping systems enables the establishment of sustainable agricultural policies tailored to environmental conditions. However, conducting field surveys over extensive agricultural areas presents significant challenges. Satellite data for agricultural monitoring provides continuous and large-scale information for cropland. The purpose of this study is to develop a cropping pattern product for annual crops using satellite data. The study area is ‘Gimje-si’ in the Republic of Korea. Sentinel-2 Level-2 data was acquired from 2022 to 2024. The normalized difference vegetation index (NDVI) was calculated after eliminating cloud and contaminated pixels, and then the monthly mean NDVI was computed. Cropland was extracted using a farmland boundary map in vector file format. Types of cropping patterns were classified into single and sequential (e.g., double, triple) cropping, and non-cultivated land, based on the number of peaks in the time-series NDVI data. The threshold for NDVI peaks was set to 0.4, and the minimum distance between NDVI peaks was set to 3. The final product was generated in vector format and includes monthly NDVI values, cropping patterns, and peak month information for each field. The annual map for 3 years showed changes in cropping patterns. These products were useful for detecting changes in cropland and confirming whether it was being cultivated. There was an increasing trend in the number of fields with sequential cropping from 2022 to 2024. Our results help comprehend the use and change of cropland spatiotemporally.
Acknowledgments: This research was funded by RDA, grant number PJ01676802.
How to cite: Choi, L.-Y., Ryu, J.-H., Ahn, H.-Y., Lee, S.-J., Kwak, G.-H., Jeon, Y.-A., and Lee, K.-D.: Development of Cropping Pattern Product Using Sentinel-2 Satellite Data, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-8077, https://doi.org/10.5194/egusphere-egu25-8077, 2025.
Accurate cultivated land parcels (CLPs) information is essential for precision agriculture. Deep learning methods have shown great potential in CLPs delineation but face challenges in detection accuracy, generalization capability, and parcel optimization quality. This study addresses these challenges by developing a high-generalization multi-task detection network coupled with a specialized parcel optimization step. Our detection network integrates boundary and region tasks and design distinct decoders for each task, employing performance-enhancing modules as well as more balanced training strategies to achieve both accurate semantic recognition and fine-grained boundary depiction. To improve network's ability to train more generalized models, our study identifies the variations in image hue, landscape surroundings, and boundary granularity as the key factors contributing to generalization degradation and employ color space augmentation and attention mechanisms on spatial and hierarchy to enhance the generalization. Additionally, the parcel optimization step repairs long-distance boundary breaks and performs object-level fusion of delineated regions and boundaries, resulting in more independent and regular CLP results. Our method was trained and validated on GaoFen-1 images from four diverse regions in China, demonstrating high delineation accuracy. It also maintained stable spatiotemporal generalization across different times and regions. Comprehensive ablation and comparative experiments confirmed the rationale behind our model improvements and demonstrated our method's effectiveness over existing single-task models (SegNet, MPSPNet, DeeplabV3+, U-Net, ResU-Net, R2U-Net), and recent multi-task models (ResUNet-a, BSiNet, SEANet).
How to cite: Zhu, Y. and Pan, Y.: A deep learning method for cultivated land parcels (CLPs) delineation from high-resolution remote sensing images with high-generalization capability, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-8391, https://doi.org/10.5194/egusphere-egu25-8391, 2025.
Precision irrigation is one of the fundamental areas of modern agriculture that aims to manage water use more efficiently and sustainably. Continuous monitoring of crop status is essential for the optimisation of irrigation systems, in which spectral-based monitoring methods play a key role. These methods use the spectral properties of the light reflected or absorbed by plants to determine vegetation indices, soil moisture and other plant life parameters. Measurements in the optical and infrared (IR) wavelengths are particularly important as these wavelengths are sensitive to the biochemical and physical properties of plants, such as chlorophyll content, nitrogen levels and water content.
The primary aim of the study is to expand the area of remote sensing in agricultural monitoring using laboratory, field scale proximal sensors, field an UAV imaging by creating a new rapid non-invasive approach for predicting crop health and water demand using spectral data. The study seeks to close the gap where chlorophyll estimations are generally not plant-specific by offering an integrated and refined approach to improve reliability and accessibility in chlorophyll estimation. Besides Integrating VI and thermal imaging with UAV technology can be used in precision agriculture in a number of areas, such as crop monitoring, yield forecasting and optimisation of irrigation water allocation. Furthermore, using several VIs were found to be optimal in crop coefficient estimation, so as to more precise calculation of crop evapotranspiration The ultimate result is giving new approaches to farmers and agricultural stakeholders for more precise and dependable tools for measuring crop evapotranspiration, crop health while promoting sustainability, efficiency, and scalability in irrigation practices.
The research presented in the article was carried out within the framework of the Széchenyi Plan Plus program, with support from the RRF 2.3.1 21 2022 00008 project. This research was supported by the János Bolyai Research Scholarship of the Hungarian Academy of Sciences
How to cite: Nagy, A., Szabó, A., Nxumalo, G. S., Budayné Bódi, E., and Tamás, J.: Spectral-based monitoring methods to optimise precision irrigation in maize, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-9499, https://doi.org/10.5194/egusphere-egu25-9499, 2025.
The development of UAVs and the reduction in the weight of payload-bearing devices is making remote sensing of crops possible. This technology is cheaper, more time-efficient and produces higher resolution images in a non-destructive way. Another important feature of drone imagery is its ability to monitor crops on a regular basis. The raw data collected by drones can be integrated into models for analysis and further corrective measures can be created to improve crop yields. Drones are capable of assessing soil conditions, assisting in irrigation, fertilizer application and monitoring crop health. The Normalized Difference Vegetation Index (NDVI) was used to quantify the greenness of vegetation to assess changes in vegetation density and health. When near-infrared light reaches the leaves of a healthy plant it is reflected back into the atmosphere, as the amount of chlorophyll produced by the plant decreases, less near-infrared radiation is reflected back. The result can then be used to assess the overall health of the plant. The values are calculated for each pixel of your map, giving you an index in the range -1 to 1.
4 sampling points (A-D) were selected in the sample area Nyírbator, Hungary. Soil moisture and soil temperature probes were deployed at three depths in the points and data were downloaded during bi-weekly sampling and measurements. The vegetation monitoring of the irrigated and non-irrigated area was carried out by taking NDVI images every 2 weeks using UAV remote sensing. During the NDVI processing of the irrigated area, only the first half of the area was captured for the initial images, at the beginning of the vegetation. NDVI images were processed in Pix4D and ArcGIS Pro software. In ArcGIS Pro, the minimum, maximum, mean and standard deviation values for the study area were observed and subsequently evaluated separately point by point using a zonal statistics algorithm.
In the study area, a larger temperature variation is observed for the deployed soil probes at a depth of 10 cm, which underlines the sensitivity of the surface temperature to environmental conditions. With increasing depth, a gradual decrease in temperature is observed, indicating the influence of soil properties on heat retention and dissipation. Consistently fluctuating moisture levels near the surface (at a depth of 10 cm) were observed in response to precipitation or irrigation events. The fluctuation of the curves gradually decreases with increasing depth. At all depth levels, a more consistent linear gradient is observed, reflecting the prolonged drought conditions in the soil. This observation is consistent with the low mean NDVI values observed simultaneously in the same zone. The data show that the irrigated area tends to have higher average NDVI values than the non-irrigated area, which has significantly lower values.
This research was supported by the János Bolyai Research Scholarship of the Hungarian Academy of Sciences. The research presented in the article was carried out within the framework of the Széchenyi Plan Plus program with the support of the RRF 2.3.1 21 2022 00008 project.
How to cite: Szabó, A., Budayné Bódi, E., Blessing, A. B., Kun, S., Kiss, É. N., Tamás, J., and Nagy, A.: Correlation between NDVI and soil sensor data collected by UAV, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-10478, https://doi.org/10.5194/egusphere-egu25-10478, 2025.
Understanding and monitoring crop growth is crucial for addressing global food security challenges and promoting sustainable agricultural practices. Traditional methods of observing crop traits in plot experiments are labor-intensive, limiting their spatial and temporal resolution. While conventional satellite platforms like Sentinel-2 and Landsat have proven valuable for large-scale agricultural monitoring, their spatial resolutions and temporal gaps are insufficient for time series of small experimental plots. Recent advancements, such as PlanetLabs’ SuperDove constellation, provide an alternative by offering daily imagery at a 3 m resolution, making them suitable for small-scale plot-level analysis. Despite their high spatial detail, these images face challenges related to radiometric stability, spatial co-registration accuracy, and quality masks, which must be resolved for effective small-scale monitoring. Addressing these limitations, this research investigates the use of PlanetScope data to estimate canopy cover (CC) and leaf area index (LAI) in plot experiments. High-resolution Unmanned Aerial System (UAS) RGB imagery was used as a reference to estimate early-stage CC. By applying a machine learning-based segmentation technique, we distinguished foliage from background pixels. This segmentation enabled us to integrate UAS-derived CC estimates with 8-band multispectral imagery from PlanetLabs’ SuperDove constellation. After improving the radiometric stability and spatial accuracy of the satellite imagery, we used the multispectral data along with UAS-derived canopy cover estimates as inputs to identify the most sensitive satellite-derived vegetation indices (VIs) for estimating CC during the early growth stages. In conjunction with LAI, we generated model-based time-series growth curves covering all phenological stages. The method was validated on experimental plots in northern Switzerland, with varying soil compaction and fertilization treatments. The study demonstrates successful segmentation of high-resolution UAS-based RGB imagery, providing a robust baseline for validating satellite-derived data and training novel retrieval methods for canopy cover. Comparative analyses identify vegetation indices from PlanetScope imagery that correlate with early crop growth. This research highlights the potential of high-resolution satellite data for generating time-series growth curves, offering a valuable tool for improving crop management and optimizing resource use across diverse farming systems.
How to cite: Boos, T. and Aasen, H.: Using High-Resolution Satellite Data to Estimate Canopy Cover and Leaf Area Index in Plot Experiments, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-12456, https://doi.org/10.5194/egusphere-egu25-12456, 2025.
Like sun-induced fluorescence (SIF), LED-induced fluorescence (LEDIF) became frequently used to establish and analyze leaf- and canopy-level stress responses. Different plant phenotypes (trees, understory shrubs, crops, vineyards, etc.) were subjected to, in most of the studies, blue LED illumination during the night or in darkened boxes for assessing either the entire broad-band (650-850 nm) spectrum of LEDIF or one of the wavelength bands of the red (~ 690 nm) and far-red (~ 740 nm) peak emissions. It seems to be however less common to apply close to “white” LED lighting, mixed from different wavelength ranges all below 650 nm (to overcome spectral overlap of red excitation and emission) as a light source. Moreover, stress manipulation in microcosm experiments is also scarce within studies while detecting LEDIF signal changes.
In our study, we established a microcosm experiment with four treatments on sunflowers: well-watered – no heat stressed, well-watered – heat stressed, water-stressed – no heat stressed, and water-stressed - heat stressed. The plants were gradually exposed to the treatments during the two months of the experiment between October and December 2024. We captured reflectance and the broad-band fluorescence spectra above the canopy with a VIS-NIR spectrometer facing downwards toward the canopy between the LED panels. We followed the response of the plants to the imposed stress by weekly/bi-weekly measurements and analyzed the changes in the shapes of the curves. We also captured the canopy architecture with side-view photos and leaf area growth with top-view photos. There was a clear increase in the LEDIF signal during the canopy development, and then a heterogeneous response depending on the treatment.
How to cite: Fóti, S., Mészáros, Á., Guettala, I., Péli, E., Pintér, K., Nagy, Z., and Balogh, J.: LED-induced chlorophyll fluorescence during heat and drought stress as assessed in a microcosm experiment on sunflower, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-13239, https://doi.org/10.5194/egusphere-egu25-13239, 2025.
Soil organic carbon (SOC) stocks represent the second-largest natural carbon reservoir globally, surpassed only by the oceans. SOC plays a vital role in maintaining ecosystem health, offering numerous benefits such as enhancing soil structure, increasing nutrient availability, and boosting water retention capacity. Beyond its ecological significance, SOC is integral to climate change mitigation, given its ability to sequester atmospheric carbon dioxide effectively. Additionally, SOC contributes to improving the physical, chemical, and biological properties of soil, making it indispensable for sustainable land management. Taiwan, an island in the western Pacific Ocean, spans an area of approximately 35,800 square kilometers. Shaped like a tobacco leaf, the island extends 400 kilometers in length and 150 kilometers at its widest point. Taiwan’s landscape is characterized by a Central Mountain Range running north to south, steep slopes, and geologically fragile formations. In recent decades, Taiwan has experienced significant changes in land use and land cover, particularly in urban areas where cropland and forest land on city outskirts have been replaced by infrastructure development. These transformations have directly impacted SOC levels across the island, underscoring the need for accurate mapping to estimate SOC stocks and assess soil functionality, particularly in agricultural regions. Traditional ground sampling methods for estimating SOC, though precise, are often costly and labor-intensive. To address these limitations, alternative approaches, such as remote sensing, offer cost-effective solutions. Among various predictive modeling techniques, machine learning algorithms like Random Forest (RF) have emerged as highly effective tools for SOC estimation. RF models excel due to their ability to minimize correlation among individual decision trees and provide reliable error estimates, ensuring robust predictions.
In this study, we combined field sampling data (2010–2021) with remote sensing, topographic, and climatic datasets to estimate SOC stocks in the topsoil layer (0–30 cm) of Taiwan’s agricultural areas. Using the RF algorithm, we initially employed 23 explanatory variables and subsequently refined the model by eliminating less significant predictors, reducing the final set to 12 variables. The refined model demonstrated strong predictive accuracy, with R² values exceeding 0.70 for agriculture land in Taiwan. Our findings revealed spatial variations in SOC levels, with mountainous regions exhibiting higher SOC stocks compared to suburban and low-lying agricultural areas, where values were notably lower. SOC levels for agricultural lands ranged from a maximum of 7.14 kg/m² to a minimum of 2.55 kg/m², with an average value of 3.43 kg/m². Agricultural practices incorporating agroforestry techniques showed relatively higher SOC stocks, emphasizing the role of sustainable practices in enhancing soil carbon storage. The results of this study hold significant implications for long-term monitoring of SOC in Taiwan and provide a crucial reference for policymakers aiming to develop effective carbon sequestration strategies. By integrating field data with advanced modeling and remote sensing technologies, this research contributes to a deeper understanding of SOC dynamics and supports efforts to promote sustainable land management and climate resilience.
How to cite: Valdez Vasquez, M. C., Chen, C.-F., Syu, J.-H., and Chen, L.-C.: Mapping Soil Organic Carbon Dynamics in Taiwan’s Agricultural Land Using Field and Remote Sensing Data., EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-14174, https://doi.org/10.5194/egusphere-egu25-14174, 2025.
Methane (CH₄) emissions from paddy rice fields significantly contribute to greenhouse gas emissions and global climate change. In Taiwan, rice cultivation occupies approximately 20% of agricultural land. This study utilizes Sentinel-2 and Sentinel-5P satellite data to monitor methane emissions from these fields. The research follows four key steps: 1) classifying rice cropping areas; 2) detecting the phenological stages of rice; 3) correlating spatial and temporal data with rice cultivation and methane emissions; and 4) validating the results with in-situ data. The preliminary findings identify methane emission hotspots during the rice-growing seasons, revealing substantial temporal variability linked to agricultural practices such as water management, organic matter application, and rice phenology. Peak emissions occur during the early to mid-growing stages. The adoption of satellite data for monitoring emissions offers a cost-effective and scalable alternative to traditional methods, which are often labor-intensive and geographically limited. The research can also enhance the sustainable agricultural management strategies for achieving local greenhouse gas reduction targets.
How to cite: Chen, C.-R., Chen, C.-F., Son, N.-T., Chen, L.-C., Liu, T.-S., and Kuo, Y.-C.: Monitoring Methane Emissions from Rice Paddies in Middle Taiwan Using Remote Sensing Data., EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-14688, https://doi.org/10.5194/egusphere-egu25-14688, 2025.
The species Bituminaria bituminosa (L.) C.H. STIRT has been studied over the past two decades to be integrated as a forage crop in agro-silvo-pastoral systems due to its nutritional qualities and low water requirements (<200 mm). These efforts have led to the development of new varieties using genotypes from the Canary Islands. These varieties are expected to be utilized in mixed or monoculture systems, leveraging their drought tolerance to extend the availability of high-quality feed, thus reducing costs during the forage shortages of the summer season. The ability of Bituminaria to fulfill this role in Mediterranean basin farms will largely depend on its adaptation to environmental conditions.
This study aims to explore the circum-Mediterranean distribution of Bituminaria using Species Distribution Models (SDMs) and 33,132 occurrences from the GBIF platform on natural populations of the species. Bioclimatic, edaphic, geomorphological, and satellite-derived variables were used in model development through the biomod2 package in R, achieving ensemble model metrics with a mean True Skill Statistic (TSS) of 0.78. Eight clusters have been proposed to group occurrences based on the most important variables identified in the ensemble model, which also aids in identifying isolated populations or localized scenarios that may serve as a foundation for breeding programs aimed at improving specific traits. These results will contribute to a deeper understanding of the ecology, phenotypic plasticity, population dynamics, movement patterns, and evolutionary processes within the genus Bituminaria.
How to cite: San Martin Loren, J., Fernandez Habas, J., and Fernandez Rebollo, P.: Species Distribution Models: Application to the Identification of Populations and Potential Distribution Areas of the Forage Plant Bituminaria bituminosa, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-18748, https://doi.org/10.5194/egusphere-egu25-18748, 2025.
The increasing frequency of drought periods and the intensification of precipitation distribution extremes in Central Europe, particularly in eastern Hungary, pose significant challenges for water resource management. The Great Hungarian Plain (Alföld) experiences an annual precipitation deficit of 150–250 mm, exacerbating the adverse effects of drought. The Eastern Main Canal (Keleti-főcsatorna) plays a crucial role in water supply, transporting 300–400 million m³ of water annually as part of the Civaqua program. This initiative aims to channel water from the Tisza River to the Tócó stream, ensuring sustainable water supply for the region and maintaining critical water levels in local reservoirs, including the Vezér Street Retention Basin. The basin serves not only water retention and flood control purposes but also provides recreational opportunities for the local community.
This study aims to evaluate strategies for maximizing the capacity and efficiency of retention basins by optimizing the water supply from the Tisza River and the Eastern Main Canal, particularly during drought periods. Additionally, the research explores the potential of basin retention for the storage of precipitation and excess water within the basin and surrounding landscapes. Such retention solutions contribute to efficient water resource management, mitigating drought impacts and enhancing the long-term sustainability of water management practices.
The research employed active remote sensing technologies, including the Apache 3 unmanned surface vessel equipped with a monobeam sonar, providing depth measurement accuracy within 1% of the measured depth. For terrestrial surveys, the Stonex X120GO SLAM Laser Scanner was utilized, delivering millimeter-level precision in 3D mapping. The integration of these technologies enabled the development of detailed basin models, capturing both underwater and aboveground features of the retention basin. The primary focus was the Vezér Street Retention Basin, which serves flood control, water retention, and recreational functions in the Debrecen area.
The lowest point of the Vezér Street Retention Basin is at an elevation of 110.65 m above Baltic Sea level, while the highest point of the basin crown is 114.39 m, resulting in a maximum depth of 3.74 m. The basin’s total storage capacity, when fully saturated, is 39,213.59 m³, with a water surface area of 16,354.93 m². At the average water level of 113.69 m, the basin holds approximately 28,253.2 m³ of water, with a water surface area of 15,000.08 m². During the summer, under conditions of 20°C, average atmospheric pressure, and humidity, evaporation rates reach 3 mm/day/m², resulting in a daily water loss of 45,000.24 mm/day. The aquatic biodiversity of the basin is characterized by the presence of Typha species, which serve as critical ecological indicators.
The preliminary findings highlight that active remote sensing methods, such as sonar and the Stonex X120GO SLAM Laser Scanner, provide reliable tools for maximizing basin capacity and developing efficient water retention strategies.
The research presented in the article was carried out within the framework of the Széchenyi Plan Plus program with the support of the RRF 2.3.1 21 2022 00008 project.
How to cite: Pásztor, D., Nagy, A., Fehér, Z., and Tamás, J.: Assessment of Retention Basin Potential Using Active Remote Sensing, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-19406, https://doi.org/10.5194/egusphere-egu25-19406, 2025.
Anticipating the response of grapevines to environmental variability is crucial for opti-mizing field management practices. This study explores the interaction between vines and their habitat across the growing cycle to inform more effective vineyard management. The research was conducted at the "Alhambra" plot in Ciudad Real (38.8089720, -3.0705830), which spans approximately 6 hectares of irrigated Tempranillo (Vitis vinifera L.) vines. Vine spacing is 3.05x1.54 m², and the training system is a double guyot pruned, vertical shoot positioning. The study utilizes data collected over 2024.
Within the plot, three replicates of 30 plants each were sampled. Measurements were taken from consecutive rows, 15 plants each. At the phenological stage of separated clus-ters, the number of clusters was recorded, while berry weight and the number of berries per cluster were assessed at veraison and harvest. Yield partitioning was determined at harvest. Additional parameters were also measured, including total soluble solids, surface area, pruning and shoot weight.
A custom script was developed to analyze the orthophotos of the vineyard to quantify the trellis length occupied by vines, excluding gaps where vines were missing. This method enables precise calculation of the vine-covered productive area. By combining these or-thophoto analyses with field-estimated yields per linear meter of vine, the study could provide accurate vineyard yield predictions. The accuracy and effectiveness of this inte-grated methodology are thoroughly evaluated.
Acknowledgements BigPrediData
How to cite: Novellón, M., Lacalle, S., Tarquis, A. M., and Baeza, P.: Integrating Orthophotos and Field Data for Precision Vineyard Yield Prediction: A Case Study of Tempranillo Grapevines, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-19779, https://doi.org/10.5194/egusphere-egu25-19779, 2025.
Andean communities in central Peru play a key role in the conservation of vicuñas (Vicugna vicugna), a protected species that depends on puna grass and flooded vegetation for food and access to water throughout the year. This study focuses on seven communities of Lucanas in Ayacucho, a dry mountainous region of Peru, emphasizing the need for accurate information to monitor resources in a context of climate change and support community decision-making. In this research, based on Google Earth Engine (GEE), we evaluated the performance of classification algorithms using Sentinel-1 (S1) and Sentinel-2 (S2) image data for rangelands classification. The process used ground-based and image-based points to train and validate the models, a filter to minimize spatial autocorrelation between training and validation sets; and spectral separability measurements using the Jeffries-Matusita (JM) distance, all of steps allowed an adequate discrimination and representation of the classes. Additionally, we used 64 feature variables (including vegetation, texture, topographic, snow, water, minerals, radar features) and applied Cloud Score+, quality assessment (QA) processor in S2 image collection, to improve classification accuracy. Random Forest (RF) algorithm achieved an overall accuracy (OA) of 92% and a Kappa coefficient of 0.908 outperforming the Support Vector Machine (SVM) algorithm, which obtained an OA of 90.9% and a Kappa coefficient of 0.895. The results show that, in the semi-captivity sectors, 1,777.5 hectares of puna grass and 319.1 hectares of flooded vegetation were identified, while in wild management areas 5,431.1 hectares of puna grass and 843.8 hectares of flooded vegetation were recorded. These findings highlight the importance of integrating remote sensing tools and machine learning algorithms to generate key information in the management of natural resources in communities.
How to cite: Ochoa, J., Juarez, H., Sotomayor, D., and De Haan, S.: Mapping Rangeland Vegetation Using Sentinel-1 and Sentinel-2 Imagery with Machine Learning: A Case Study of Vicuña Conservation in the Central Andes of Perú, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-21573, https://doi.org/10.5194/egusphere-egu25-21573, 2025.
Rapid advancements in technology, particularly the rise of artificial intelligence (AI) and the integration of uncrewed aerial vehicles (UAVs) equipped with RGB, multi- and hyperspectral sensors, have boosted agricultural research on crop disease detection. This has led to a surge in studies exploring high-technology approaches to detecting crop diseases. While numerous studies have demonstrated high accuracy in detecting specific diseases or pests in crops, concerns arise regarding their reproducibility and generalisability.
We conducted a meta-analysis of over 100 research papers to examine how models are trained and validated, with a focus on how datasets for training, validation and testing were handled. In principle, a model can only be considered robust and widely applicable if it performs well on an entirely new dataset, i.e., a dataset it wasn’t specifically trained one. Otherwise, AI models risk overfitting to specific datasets or fields, potentially detecting signals that are not universal or not related to the targeted pest or disease. This issue arises when datasets are randomly split in training, validation and test subsets.
Our analysis revealed significant limitation in current practices. Nearly half of the reviewed papers relied on a single dataset (one single field, one single flight) for both model training and validation. About one-quarter of the studies used data from a single field with repeated flights during the same growing season. Only another quarter utilized datasets from multiple fields; however, the majority of these studies still used a random split for training and testing, meaning their models were not evaluated on independent datasets. In addition, a handful of studies using RGB data, applied transfer learning, with models pretrained on public (non-UAV) datasets and then applied to UAV datasets.
Overall, only 10% of the reviewed papers validated their models on fully independent datasets, i.e, using transfer learning or using an independent (untrained) separate field to test the model. We found that particularly models constructed with multispectral or hyperspectral data did not use independent datasets. On top of that, none of the studies explicitly tested whether their models were pest- or disease-specific, i.e., whether the models were sensitive only to the pest or disease they were trained to detect.
These findings highlight a critical limitation in the robustness and scalability of current AI-approaches to crop disease detection with UAVs. To address this, we call on researchers to include independent test datasets in their studies, and urge journals and reviewers to prioritize this criterion during evaluations. Additionally, we advocate for the public sharing of datasets to enable the development of robust and generalisable methods.
How to cite: Eeckhout, E., Spanoghe, P., and Maes, W.: UAV-based disease and pest detection using AI: Time to reconsider our approach?, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-6149, https://doi.org/10.5194/egusphere-egu25-6149, 2025.
Herbicides are extensively used for weed management worldwide. However, their use is a significant cause
of environmental pollution and human health problems. Efficient Site-Specific weed management (SSWM)
practice attempts to reduce herbicide use and its negative impacts by adjusting herbicide application based
on weed composition and coverage. Such an application requires high-resolution data in spatial and spectral
domains, which is not always available. Consequently, Mixed pixels are likely to exist, creating a challenge
to generate accurate weed maps. In this regard, Spectral Mixture Analysis (SMA) can mitigate this challenge
by exploiting subpixel information. This study assesses the potential benefits of four SMA methods for
estimating weed coverage of different botanical groups. We examined four methods- Constrained Least
Squares Unmixing (FCLSU), Sparse Unmixing via variable Splitting and Augmented Lagrangian (SUnSAL),
Sparse Unmixing via variable Splitting and Augmented Lagrangian and Total variation (SUnSAL-TV) and
the Vectorized Code Projected Gradient Descent Unmixing (VPGDU). Each suggests a distinct advantage
for spectral unmixing. We used controlled hyperspectral and multispectral field datasets to compare the four
methods. The controlled data included weed species characterized by distinct botanical groups, while the
field dataset included a corn field with weeds at varying densities. We assessed the performance of the
different methods in estimating weed coverage and composition at various spatial resolutions. Our results
demonstrated the advantages of the total variation regularization of SUnSAL-TV and the superiority of the
SAM-based method, VPGDU, over other approaches. VPGDU was the best-performing method, with MAE
values consistently lower than 8.6% at all resolutions, underscoring the advantage of its objective function
in unmixing weed botanical groups and the significant effect of illumination on the results. This result was
also consistent in the field data as VPGDU yielded the lowest MAE of 11.95%,
How to cite: Ronay, I., Lati, R. N., and Kizel, F.: Comparing Different Unmixing Methods for weed detection and identification, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-15653, https://doi.org/10.5194/egusphere-egu25-15653, 2025.
Abstract
Ravines in arid lands are affected by various soil erosion processes caused by inconsistent rainfall regimes, flooding patterns, and anthropogenic interventions. These effects are expressed in the geomorphological and vegetation patterns of the ravine's land segments. To address these changes, a study was planned with the following objectives: [1] Modeling the effects of ravine erosion processes on its land-segments vegetation using high-resolution satellite imagery; [2] Suggesting analysis schemes based on remote sensing to suit land management practices for the ravine parts. The study site is located in Migda Ravine, Northern Negev, between Gerar and Patish ephemeral streams. Due to the loess soil and extreme arid conditions, the area suffers from soil erosion and land incision. Using imaging from PlanetScope® satellite constellation (spatial resolution: 3m pixel-1, temporal: Image per 3 days, and spectral: Red-Green-Blue-Near Infra-Red bands) between 2017 and 2024, from January to August each year, NDVI median and quartiles ranges of the ravine land segments were calculated and normalized against a stabilized reference plot. Thirteen erosion processes were defined, and classified into ravine surrounding areas, banks, and ephemeral stream water flow. The findings indicate erosional processes that dramatically decreased the Normalized Fresh Vegetation Reflectance (NFVR)in 2019, with a lighter decrease in 2020. Some erosion processes were characterized by a subsequent NFVR increase after the soil erosion event, while others, such as subsurface erosion, showed a continuous NFVR decrease. Stream plots were characterized by soil deposition, which resulted in vegetation flocculation. Using vegetation change patterns, NDVI normalization, and multi-year temporal analysis can aid in formulating land management practices for the ravine land segments and predicting long-term erosional patterns.
How to cite: Mor-Mussery, A., Zaady, E., and Blank, L.: Modeling and managing erosion in arid ravines using high-resolution satellite imagery, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-8008, https://doi.org/10.5194/egusphere-egu25-8008, 2025.
Accurate and timely seasonal yield predictions before harvest are becoming ever more relevant due to increasing pressure on the agricultural sector under climate change. Especially for agricultural planning, logistics, and food markets, seasonal predictions are of significant importance in the context of food security and price stability.
A novel approach to enhance early-season yield forecasts at the regional scale will be presented. Earth observation (EO) data from the Copernicus Sentinel-3 satellite are able to trace spatio-temporal vegetation dynamics (e.g., crop phenological status, crop growth, photosynthesis via FAPAR, or chlorophyll indices) in near real-time. By deriving daily satellite composites and combining these data with physical modelling using the Lund-Potsdam-Jena managed Land (LPJmL) dynamic global vegetation model (DGVM) in a newly developed assimilation process, enhanced yield forecasts can be achieved. There are currently no interfaces for continuous assimilation of EO data for the LPJmL model, thus, approaches such as parameter forcing and ensemble methods allowing for continuous parameter optimization during the course of the growing season are presented and compared conceptually to improve the LPJmL model for seasonal yield predictions. Existing methods for model parameter calibration and optimization with EO data using machine learning are applied to agricultural areas in the study area.
While these results focus on the study area of Bavaria, southern Germany, the approach is scalable also on national or European scale. For demonstration purposes, the year 2018 – a comparably dry year – was chosen due to the availability of detailed land use data. LPJmL was designed for global simulations, hence, a regional downscaling is necessary for its application at the regional scale.
Integrating different remote sensing data sources enables a more detailed picture of plant growth, which will allow a regional early warning system for food security and farmer’s turnover in the future. The combination of process- and data-based approaches is likely to improve accuracy and lag time.
How to cite: Jörges, C., Hank, T., and Fader, M.: Chances and Challenges of Data Assimilation for Seasonal Yield Predictions Using Sentinel-3 Satellite Data and the Agro-Ecosystem Model LPJmL, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-13318, https://doi.org/10.5194/egusphere-egu25-13318, 2025.
Posters virtual: Fri, 2 May, 14:00–15:45 | vPoster spot 4
EGU25-18606 | ECS | Posters virtual | VPS20
Deep Learning based Paddy Land Abandonment Detection Using Multitemporal Polarimetric SAR PatternsFri, 02 May, 14:00–15:45 (CEST) | vP4.11
The abandonment of agricultural land in India, especially paddy fields, has emerged as a significant challenge for food security and ecosystem sustainability in the country. Although rice production is vital for national food security, research on paddy land abandonment in India remains limited. Some Indian states have reported an alarming decline in paddy cultivation area over the past two decades. The study employs the Udupi district of Karnataka, India, a high-rainfall coastal region where paddy has traditionally been the dominant crop and where paddy land abandonment has been observed, as the study area. This study addresses crucial research gaps by framing these objectives for the study: (1) developing a deep learning framework that utilizes both intensity and phase information from polarimetric Synthetic Aperture Radar (SAR) data for abandoned paddy land detection, (2) leveraging recurrent neural networks (RNNs) to capture temporal patterns in abandonment, and (3) demonstrating an automated, all-weather monitoring approach that overcomes the limitations of traditional optical remote sensing in tropical regions.
Conventional monitoring approaches struggle with persistent cloud cover in tropical regions which limits effective assessment of abandonment patterns. SAR data provides unique capabilities for continuous monitoring under all weather conditions, making it particularly well-suited for tropical regions. However, previous studies have primarily underutilized SAR's potential by concentrating solely on backscattering intensity from ground range detected (GRD) products, overlooking the valuable phase information that could offer deeper insights into land use changes. In this study, we employ Sentinel-1 Single Look Complex (SLC) data, which offers both intensity and phase information. Considering the temporal nature of paddy land abandonment, we developed a deep learning framework utilizing RNNs viz. LSTM, BiLSTM and BiGRU to effectively capture time-series patterns in the data. This framework analyzes backscattering coefficients (VV and VH polarizations) and polarimetric parameters (entropy, anisotropy and alpha angle) derived from SLC data collected during the Kharif seasons from 2017 to 2024. We carried out extensive ground truth data collection of active and abandoned paddy lands to train and validate our models. The backscattering coefficients were processed through orbit correction, radiometric calibration, TOPSAR deburst, multi-looking, speckle filtering and terrain correction. For deriving the polarimetric parameters, after basic preprocessing steps, the covariance matrix was generated followed by the polarimetric decomposition of the phase-preserved data. Results indicate that our RNN models show promising performance in detecting temporal patterns of paddy land abandonment. The method exhibits a robust ability to produce reliable abandoned land maps in regions prone to cloudy and rainy conditions. Future research should explore polarimetric features across various vegetation types in abandoned lands, expand the methodology to other agricultural systems, and examine the impact of socio-economic and topographical factors on abandonment patterns to support evidence-based land management policies.
How to cite: Kasture, S., Hegde A, A., and Umesh, P.: Deep Learning based Paddy Land Abandonment Detection Using Multitemporal Polarimetric SAR Patterns, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-18606, https://doi.org/10.5194/egusphere-egu25-18606, 2025.
