Abstract: 

In this work, we develop a hydrological model designed to simulate the water balance and runoff processes at the catchment-scale. Instead of using a rectangular grid discretization, the model represents the catchment using a non-homogeneous two-dimensional triangular mesh framework (similar to a triangular mesh in the Finite Element method). This discretization fits a more flexible representation of complex topography and land boundaries. The model is implemented in the Fortran programming language. It depends on the Digital Elevation Model (DEM) to extract the flow pathways starting from upstream and reaching downstream. That guarantees a physically consistent and explicit flow-routing structure across the triangular mesh.

Evapotranspiration is calculated using the Penman–Monteith equation, as the parameters are considered to suit coastal climate conditions. The model utilizes temperature, solar radiation, wind speed, and vapor pressure as atmospheric inputs. The SCS Curve Number method is used to estimate the surface runoff, considering slope, land cover, and soil properties. Meteorological data measurements, including precipitation, temperature, humidity, as well as inflow and outflow discharges, are integrated into the simulations.

Due to its efficient numerical structure, the model supports simulations with numerous spatial elements and long time series while maintaining the computational cost at its lowest limits. This makes it well-suited for large-scale watershed applications and provides a strong basis for future high-performance computing developments.

Keywords: hydrological modeling, watershed triangulation, flow routing, numerical simulation

How to cite: Dali, N.: Hydrological Modelling Framework for Large-Scale Catchments using triangular nonhomogeneous spatial discretization, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-8171, https://doi.org/10.5194/egusphere-egu26-8171, 2026.

A major threat to Falkland Islands (FI) biodiversity and livelihoods is a drying climate. As warming continues, FI water security is a growing concern, but we lack baseline data to inform mitigation and adaptation. This study applies an innovative remote sensing approach to monitor long-term surface water variability across the Falkland Islands, aiming to support climate-resilient water management. Using the Google Earth Engine (GEE) platform, historical dynamics from 1999–2021 were derived from the Global Surface Water (GSW) Explorer dataset, while recent trends (2021–2025) were assessed using Harmonized Sentinel-2 MSI Level-2A imagery. Together, these enable the first continuous, multi-decadal assessment of pond, wetland, and lake dynamics across East Falkland, West Falkland, and Lafonia. Preliminary results show a relative decline in surface water extent across East Falkland, West Falkland, and Lafonia from 1999 to 2021. More recent Sentinel-2 observations reveal regionally distinct trends from 2021 to 2025: East Falkland remains relatively stable, West Falkland shows a modest increase, and Lafonia exhibits a pronounced rise with strong seasonal variability. These results align with limited ground observations from the water level monitoring site, where satellite-derived surface water area strongly correlates with recorded maximum water levels, confirming the hydrological consistency of the satellite data. Despite limited ground validation, this proof-of-concept highlights the capability of cloud-based remote sensing tools to monitor hydrological variability at regional scale. This approach illustrates how open-access Earth observation data and hydroinformatics tools can aid early detection of climate-driven water changes and strengthen water management in data-scarce areas. It also establishes a basis for future studies linking satellite data with peatland hydrology and ecosystem resilience.

Keywords: Climate Change Impacts, Surface Water Variability, Remote Sensing, Sentinel-2, Google Earth Engine

How to cite: Ko, N. T., Baylis, A., Davies, G. M., Barlow, D., and Evans, C.: Surface Water Dynamics under Changing Climate: Integrating Multi-Sensor Satellite Observations (1999–2025) across the Falkland Islands, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-245, https://doi.org/10.5194/egusphere-egu26-245, 2026.

Effectively characterizing uncertainty and error in flood prediction is essential for informed decision-making. This study combines advanced deep neural network architectures, i.e., Neural Hierarchical Interpolation for Time Series Forecasting (N-HiTS), and Long Short-Term Memory (LSTM), with multiple uncertainty quantification frameworks to evaluate flood forecasts across several watersheds in the southeastern United States. Bayesian inference, Monte Carlo–based methods, and quantile regression are applied to estimate predictive uncertainty. The comparative analysis examines how different uncertainty approaches perform across a range of flood magnitudes, highlighting their respective advantages and limitations at multiple scales. Results indicate that N-HiTS generally yields narrower and more reliable uncertainty bounds than LSTM. The findings further demonstrate that prior specification in MCMC sampling strongly influences uncertainty estimates and requires careful calibration. While Monte Carlo dropout, which is an approximate Bayesian technique, primarily captures uncertainty near flood peaks, MCMC offers a more complete characterization across the full hydrograph. In addition, this study investigates multi-site training to evaluate model adaptability under diverse hydrological regimes. Collectively, these results advance the integration of deep neural networks and uncertainty quantification to enhance flood modeling capabilities and risk management.

How to cite: Saberian, M., Samadi, V., Wagener, T., and Popescu, I.: Uncertainty-Aware Flood Prediction Using Deep Neural Networks Across Multiple Watersheds, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-15829, https://doi.org/10.5194/egusphere-egu26-15829, 2026.

Abstract. Predicting flood risk is a complex phenomenon. Several factors influence flood behavior generation and intensity such as intricate interactions between hydrological dynamics, meteorological variability, the overarching influence of climate change and land-use changes. This study explores flood risk within the watershed of Tassaout River located in the central region of Morocco. Three advanced machine learning algorithms were chosen to evaluate flood risk. These algorithms are Multi-Layer Perceptron Artificial Neural Networks (MLP-ANN), Random Forest (RF) and Support Vector Machine (SVM). The models are trained based on 11 different factors derived from remote sensing data. From ALOS digital elevation model, 8 factors are developed: Elevation, Slope, Aspect, Plan Curvature, Profile Curvature, Stream Power Index (SPI), Topographical Wetness Index (TWI), and Surface Roughness. In addition, from Landsat 9 imagery, three flood susceptibility factors are extracted: Normalized Difference Vegetation Index (NDVI), Normalized Difference Water Index (NDWI) and Land Surface Temperature (LST). The predictive performance of each model was assessed using standard classification metrics: accuracy, recall, and F1-score. Results indicate that the RF model performed the best with an accuracy of 100%, SVM algorithm achieved good performance, attaining 68% in accuracy and more than 80% in f1-score. However, the ANN model underperformed compared to the other algorithms, with an accuracy of only 59% in accuracy and 70% in f1-score highlighting its limitations in capturing the decision boundaries within the current data configuration. Furthermore, the Shapley Additive exPlanations model (SHAP) was used to enhance the transparency and interpretability of the modelling results.

How to cite: Legsabi, H., Tiai, S., Boussabou, S. M., Najaoui, N., El Mansouri, B., and Erraioui, L.: Modeling Flood Risk in Kalaa Sraghna Region in Morocco Using Explainable Artificial Intelligence Techniques, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-19676, https://doi.org/10.5194/egusphere-egu26-19676, 2026.

Farm roadway networks are an important infrastructure in grassland farms providing access between the farmyard and grazing fields. However, during livestock movement, excreta is deposited on the roadways, especially on bends, T-junctions and at corners where their movement is impeded. Nutrient-enriched soiled runoff generated on these roadways can contribute significantly to water quality degradation if connected to waters (including man-made open drainage ditches). Quantifying the risk associated with farm roadway runoff delivery to waters includes mapping the roadway and drainage networks and identifying sections which contain high pollutant loads and have the potential of generating, mobilising and delivering surface runoff to the drainage channels. In this study, a deep learning (DL) approach was employed to automatically identify internal farm roadway networks and open drainage channels in 5 grassland farms. Aerial imagery and LiDAR-derived digital terrain models were used to train the DL models for identifying farm roadways and open drainage ditches, respectively. The flow direction and flow accumulation were determined using digital elevation models to map farm roadway sections that have the potential to generate and deliver runoff to the drainage network.

Across the 5 farms, a total of 16.7 km of roadway and 13.5 km of drainage channels were identified by the DL models, achieving precisions of 79 % and 64 %, and accuracies of 90 % and 96 %, respectively. Flow accumulation maps were established for each farm to assess delivery pathways and the potential of roadway runoff connectivity to waters. Flow pathways through roadway junctions and at corners were considered critical outranking those on straight roadway sections. Breaking the runoff pathway at these locations will help prevent delivery to waters. The findings of this study indicate that mapping of open drainage channels and internal farm roadways in grassland farms can be automated by using deep learning models. Integrating the automated mapping and hydrological modelling enables more precise identification of critical roadway sections, supporting targeted mitigation to reduce soiled runoff from entering waters and thus enhance water quality protection in grassland farming systems.

How to cite: Sifundza, L. S., Murnane, J. G., Daly, K., Adams, R., Tuohy, P., and Fenton, O.: Integrating deep learning and hydrological modelling to assess farm roadway runoff risk to inform targeted mitigation in grassland systems, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-6937, https://doi.org/10.5194/egusphere-egu26-6937, 2026.

Monitoring of inland water bodies is considered crucial for effective water resource management. In this study, a combination of satellite imagery and altimetry products was utilized to monitor changes in water level and surface extent during the different operational filling phases of the Grand Ethiopian Renaissance (GER) Dam. The primary objective was to utilize diverse remote sensing products to provide an accurate estimation of water volume changes over time. Sentinel-1 data were processed using an unsupervised edge Otsu algorithm to map reservoir extents. These output maps were validated against Planet and Sentinel-2 water masks, and a high level of agreement was observed, with overall accuracy values ranging from 0.97 to 0.99. Furthermore, various Surface Water and Ocean Topography (SWOT) satellite products were evaluated for the estimation of reservoir extents. It was found that the SWOT Lake Single Product performed poorly, with an Intersection over Union (IOU) value of approximately 0.33 being recorded. In contrast, moderate agreement with validation sets was demonstrated by the SWOT water mask raster and pixel cloud products, with overall accuracy values ranging from 0.78 to 0.89 being observed. Volume variation across different dam operational phases was estimated through the application of satellite-based observations and a DEM contouring method. Although a high correlation (R² value of 0.98) was exhibited by both methods, significant differences in absolute values were identified (RMSE value of 2736.35 km³). These discrepancies are attributed to a potential scaling error and the inherent water slope present within the GER Dam reservoir.

How to cite: Tazwar, M., Rietbroek, R., Maathuis, B. H. P., and Shakya, A.: Tracking The GER Dam Impoundment Stages Using SWOT and Other Radar Altimetry Products, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-8732, https://doi.org/10.5194/egusphere-egu26-8732, 2026.

Climate variability represents one of the most severe threats to lacustrine ecosystems worldwide, leading to water loss in nearly half of the world’s lakes and reservoirs. On the one hand, this variability is associated with drought conditions, manifested by a decline in precipitation. On the other hand, it is linked to rising temperatures, which enhance evaporation rates at lake surfaces.

In Senegal, a Sahelian country, the prolonged drought period of the 1970s led to the desiccation of several water bodies, including Lake Tanma. In this context, the present study contributes to a better understanding of the drying process of Lake Tanma under climate variability conditions, using remote sensing techniques. Lake Tanma is located in Thiès region, approximately 70 km from Dakar.

To achieve this objective, Landsat Earth observation products were used at the beginning and end of each decade between 1984 and 2024. The time series consists of multispectral images acquired in October, corresponding to the end of the rainy season in Senegal. This choice ensures the capture of the lake’s maximum water extent, thereby minimizing seasonal fluctuations. All data were acquired and processed using the Google Earth Engine platform. The Modified Normalized Difference Water Index (MNDWI) was computed for the entire time series to accurately delineate and characterize water-covered surfaces.

The results reveal a highly variable evolution of the inundated surface area of Lake Tanma, with a variation coefficient of 57.8%. The largest flooded area was observed in 1984, covering 969.33 ha, while the smallest extent was recorded in 2024, with only 76.18 ha. Analysis of intra-decadal variations shows a slight decrease (7%) in the flooded surface, between 1984 and 1989. In contrast, subsequent decades exhibit a marked and progressive regression of the lake’s water surface, reaching 21% during the 2000–2009 decade, 61% during 2010–2019, and up to 89% over the 2020–2024 period.

These decrease trends highlight the influence of hydro-climatic parameters, particularly precipitation and evaporation, which constitute the primary drivers of lake recharge and desiccation. Consequently, further investigation, of hydro-climatic factors, namely rainfall and temperature, is required, to better understand the drying process of Lake Tanma and to assess the impacts of hydro-climatic variability on its long-term dynamics.

How to cite: Ndeye, M. D. B., Diene, S. M., Ndao, S., Sarr, S., Guèye, A., and Tamba, S.: Contribution of Remote Sensing to the Analysis of the Drying Process of Lake Tanma under Strong Climate Variability, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-10679, https://doi.org/10.5194/egusphere-egu26-10679, 2026.

Bofedales are high-altitude wetlands whose functioning is closely linked to surface water availability, playing a key role in hydrological regulation and ecosystem resilience in the Andes. Despite their importance, spatially explicit information on their surface water dynamics remains limited, particularly in data-scarce mountain regions. This study investigates the ecohydrological dynamics of bofedales in the Alto Pampas sub-basin (Huancavelica, Peru) over the 2015–2024 period using a multitemporal remote sensing approach combined with climatic information. Seasonal patterns of bofedal extent and surface water presence were mapped from Landsat 8 imagery using vegetation and moisture indices (NDVI and NDII), together with topographic variables. Bofedales were identified through a Random Forest classification framework and subsequently categorized as permanent or seasonal based on the temporal persistence of hydric signals. Changes in surface water extent and bofedal productivity were quantified, and temporal trends were assessed using the Mann–Kendall test. In addition, generalized additive models were applied to examine potentially nonlinear relationships between climatic drivers and key ecohydrological indicators. The results reveal contrasting surface water trajectories among bofedales, reflecting heterogeneous sensitivity to climate variability within the sub-basin. These findings demonstrate the value of satellite-based monitoring for assessing surface water dynamics in high-Andean wetlands and provide relevant insights for water resources management and climate change adaptation in data-poor mountainous regions.

How to cite: Sansoni-Koga, S., Laurente-Torres, J., Ccaico-Atoccsa, M., Flores-Quispe, S., Meza-Fajardo, G., and Cárdenas-Gaudry, M.: Ecohydrological and multitemporal analysis of Andean wetlands under climate variabilitiy, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-16435, https://doi.org/10.5194/egusphere-egu26-16435, 2026.

Baseflow is a crucial component of streamflow, essentially driven by changes in groundwater storage, and is vital for sustaining flows during dry periods. Traditional techniques for baseflow quantification using graphical analysis or digital filters require long-term river discharge observations, which are often limited in their spatial extent. However, with the launch of the Surface Water and Ocean Topography (SWOT) mission, global estimates of river discharge are now available over a period of two years, offering a high-resolution dataset with at least one observation every 21 days. Despite its relatively coarse temporal resolution, prior studies have demonstrated SWOT’s ability to accurately estimate average baseflow even at one observation per cycle, based on synthetic SWOT discharge estimates. The high spatial resolution provided by ‘SWOT discharge’ can be utilized to estimate baseflow at a reach-scale and gain new insights into groundwater-surface water interactions in water-stressed river basins.

In this study, we will utilize SWOT’s discharge products over Indian river basins to characterize baseflow dynamics at reach-scale resolution and examine the effects of climate variability and land-use changes on baseflow. By accurately estimating the baseflow recession parameter (k), this study will be able to identify the gaining-to-losing transition in a basin. Furthermore, the research will explore SWOT’s ability to detect temporal shifts in the baseflow recession parameter (k) during the pre-monsoon period and evaluate the effects of anthropogenic extractions on the groundwater table. Finally, these estimates will be integrated into a mass-balance model, baseflow will be converted into upstream groundwater storage (GWS) changes and validated against independent GWS anomalies derived from the Gravity Recovery and Climate Experiment (GRACE) satellites. This study will demonstrate the capability of SWOT to bridge the gap between reach-scale hydraulics and basin-scale storage, providing a vital tool for sustainable water resource management in water-stressed regions.

How to cite: Deshpande, R. S., Vidushi, V., and Syed, T. H.: Characterizing Baseflow in Indian River Basins Using SWOT Discharge Observations, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-10090, https://doi.org/10.5194/egusphere-egu26-10090, 2026.

How to cite: Barella-Ortiz, A., Quintana-Seguí, P., Cid-Giménez, J., Clavera-Gispert, R., Altés-Gaspar, V., Villar, J. M., Munier, S., Laluet, P., Olivera-Guerra, L. E., and Merlin, O.: Comparison of Irrigation Scenarios in the Ebro Basin Using the SASER Modelling Chain, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-22992, https://doi.org/10.5194/egusphere-egu26-22992, 2026.

The Souss-Massa region is known as the most important agricultural area in Morocco, and one of the most affected regions by climate change and over-exploitation. This situation has required the intervention of new tools to improve water resource management. In this context, the Unmanned Aerial Vehicles (UAVs) images data were used for weeds detection in a Citrus orchard farm. Two sites were considered, the first one planted with 12-years-old and 1.5 years-old clementine trees. After a panoply of image processing from the data collection, following by the georeferencing, the creation of the digital elevation model, the digital surface model, and the elaboration of the orthomosaic image, the machine learning algorithms (MLA) such as Maximum Likelihood Classification, Minimum Distance Classification, Support Vector Machine, were applied for weeds detection and mapping. For both sites, all MLA showed a Cohen’s kappa coefficient higher than 0.6 and an overall accuracy higher than 60%. This study demonstrates how this emerging technology offers farmers opportunities to enhance production while optimizing water usage.

How to cite: El Hafyani, M., Chaaou, A., Sadik, A., Labbaci, A., Hssaisoune, M., Tairi, A., Abdelfadel, F., Taia, S., Ait-Ichou, H., Elhaid, I., and Bouchaou, L.: A combined approach of UAV data and machine learning algorithms in weeds detection , EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-2155, https://doi.org/10.5194/egusphere-egu26-2155, 2026.

Tunisian agriculture remains a crucial component of the country’s economic development and faces considerable constraints related to increasing water demand and reducing water resources’ availability. Improving the assessment of irrigation water use is a prerequisite for sustainable water management. The present study aims to evaluate the quality of water consumption estimates in the Public Irrigated Area of Lakhmess using open-source data. High-resolution (10 m) Sentinel 2 images, combined with ERA5-land meterological data, were used to assess monthly and seasonal actual evapotranspiration (ET) and water use through the implementation of the Surface Energy Balance Algorithm for Land (SEBAL) in the Google Earth Engine (GEE) environment. The calculated water uses were combined with the seasonal supplied water to the PIA Lakhmess, collected at plot level.

This study was conducted over eight agricultural campaigns from 2015-2016 to 2022-2023. The method is validated for three sectors Sidi Jaber, Gantra and Gabel, comparing the seasonal water use estimates to water meter observations. Correlation analysis between estimated water use from open-access data and  in-situ measurement yielded correlation coefficients of 0.76, 0.75 and 0.73, with corresponding RMSE values of 0.461, 0.425, and 0,391 mm/day, respectively. In addition, SEBAL-derived evapotranspiration estimates were evaluated through comparison with reference evapotranspiration computed using the FAO-56 Penman-Monteith, resulting in an R²  of 0,68 and an RMSE of 0.315 mm/day. Overall, the methods were deemed satisfactory, as they facilitated the monitoring of excessive water usage by identifying areas where water losses occurred.

Key words: Evapotranspiration, Irrigation, water use, Remote sensing, GEE, SEBAL.

How to cite: Belhaj Kilani, A., Alonso, A., Bousselmi, A., Khlifi, S., and Vanclooster, M.: Estimation of Crops Water Consumption by Remote Sensing: SEBAL Model Calculations Versus Ground Observation In The Irrigated Area of Lakhmess (Siliana, Northern Tunisia), EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-12323, https://doi.org/10.5194/egusphere-egu26-12323, 2026.

Reliable validation of satellite altimetry over inland waters relies on long-term, high-quality in-situ water height measurements over different types of waterbodies. The strategy implemented in the St3TART Follow-On (FO) project relies on controlled super sites to produce high quality Fiducial Reference Measurements (FRMs) and on a high number of data provided by public national hydrological networks considered as opportunity sites.

However, these measurements from national hydrological networks remain highly heterogeneous in terms of formats, units, and metadata description, limiting their direct large-scale use for Cal/Val activities. The first step of data uniformization has been performed by vorteX-io team during St3TART-FO project. As an adaptation of the validation strategy for Sentinel-3 is considered for CRISTAL inland waters products, this uniformization work should be extended to cover more virtual stations for other altimetry missions.

This contribution presents the hydrological component of the Hydro-Cryo in-situ platform, INSIGHT, an ESA-funded project, extension of CRISTAL IN-PROVA project, aiming at the centralization and harmonization of publicly available in-situ water surface height data across Europe. This work participates in the preparation for the Cal/Val phase of the future CRISTAL mission and in support of ongoing Sentinel-3 validation activities, with support from the European Environment Agency (EEA) as coordinator of the Copernicus In-Situ component.

In this first phase, the platform will integrate data from twelve national hydrological networks covering France, Switzerland, Belgium (Wallonia), Ireland, Portugal, Norway, Poland, Italy, Slovenia, Croatia, the Netherlands and Germany. The data from fixed in-situ sensors deployed on Cal/Val super sites for Sentinel-3 will also be integrated in the platform. The back-end architecture is designed to easily integrate additional networks in Europe and all over the world. Native temporal resolutions provided by in situ sensors are preserved without aggregation or resampling, and up to ten years of historical observations are considered when available.

The harmonized hydrological datasets will be disseminated on a dedicated Data Hub developed by NOVELTIS together with reference Cryosphere data for satellite altimetry validation. This open-access platform is designed to serve the Cal/Val community by providing a unified entry point for inland water and cryosphere reference measurements relevant to multiple altimetry missions.

The core objective of the hydrological processing chain is the harmonization of in-situ water height measurements by standardizing measurement units and metadata across heterogeneous national public datasets. Attention is given to the consistency of the altimetric reference of the in-situ sensors. This harmonization is essential for the use of in situ stations as FRMs for the validation of both Sentinel-3 and CRISTAL, as well as for others satellite altimetry missions.

Beyond the altimetry community, this platform addresses the broader hydrological community by providing access to a standardized water height dataset from public national networks. By lowering technical barriers to data use, the infrastructure supports cross-border hydrological studies and contributes to the reuse of public hydrological observations.

This project, currently under development, establishes the data infrastructure for the needs of inland water altimetry validation, while simultaneously enabling wider scientific exploitation of harmonized in-situ water level observations at the European scale.

How to cite: Dechamp-Guillaume, M., Fouqueau, V., Hahn, J., Gil, P., Grenier, E., Poisson, J.-C., Le Merle, E., El Hajj, M., Restano, M., and Catapano, F.: Centralizing in-situ Hydrological measurements for satellite altimetry validation: the INSIGHT platform , EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-20408, https://doi.org/10.5194/egusphere-egu26-20408, 2026.

Arid and semi-arid regions are facing more frequent and severe droughts, with annual rainfall often below 200 mm. Large-scale, intensive irrigation further strains these limited water resources. Under these conditions, growers need practical tools to estimate yield and monitor tree health at high spatial detail so they can better manage irrigation and inputs. This work develops and tests an automated, data-driven pipeline for estimating citrus yield at the individual-tree level using UAV imagery and Deep Learning. The pipeline comprises three main components. First, individual trees and orchard rows are segmented using a lightweight Tiny U‑Net model. Second, a CNN-based model predicts tree-level yield from vegetation indices and field measurements. Third, these predictions are validated through detailed fruit sampling.

The study was conducted in a commercial citrus orchard in a semi-arid region under climate and water stress. High‑resolution UAV imagery was processed into orthomosaics and vegetation index maps, and the Tiny U‑Net was optimized for fast, near real‑time semantic segmentation, enabling precise tree crown delineation and accurate tree and row counts. For yield prediction, the CNN model exploited spatial features from vegetation indices combined with in‑situ data. The validation relied on direct comparison between UAV‑based yield estimates and yields obtained from field sampling and laboratory weighing. Both mean and median yields per tree were computed to capture tree‑level variability. The final dataset, consisting of 34 trees and approximately 340 fruit samples, provided a robust basis for assessing model performance. The Tiny U‑Net segmentation model reached high accuracy, with precision and recall of 94.74% and 94.88%, and an inference time of 12.55 ms per image tile. This shows the model is suitable for real‑time or on‑board use and can reliably map orchard structure at large scale. Tree and row counts derived from the segmentation achieved an R² greater than 0.99, confirming the robustness of the approach. For yield estimation, the CNN model outperformed other machine learning methods, achieving an R² of 0.88 at tree level. Field validation confirmed the practical usefulness of the pipeline, UAV‑predicted yields closely matched ground‑truth values, with both indicating an average yield of roughly 50 kg per tree. Most trees fell between 40 and 70 kg, and the model’s output histogram mean 50.9 kg, and median 51.4 kg aligned well with these field observations.

This robust agreement between model outputs and independent field validation data underscores the system's reliability and operational readiness for accurate, tree-level yield mapping. By integrating precise tree segmentation, high-resolution vegetation indices, and rigorously collected ground truth measurements, this study demonstrates that automated yield maps can be produced with sufficient accuracy to support operational decisions in orchards. This offers a cost-effective and scalable tool for precision agriculture, enabling optimized resource allocation, improved harvest planning, and adaptive management under increasing climate stress.

How to cite: Bakas, K., Saddik, A., Dliou, A., Hssaisoune, M., El Hachemy, S., Ait Ichou, H., Hmache, F., El Hafyani, M., Labbaci, A., and Bouchaou, L.: Real-Time UAV-Deep Learning System for Citrus Orchard Structure and Yield Assessment, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-2054, https://doi.org/10.5194/egusphere-egu26-2054, 2026.