The production of cocoa beans contributes to 7.5 % of European Union (EU) driven deforestation. As a result, the recent European Union Deforestation-free Regulation (EUDR) mandates producers to track cocoa farm extents comprehensively. While Remote Sensing has enormous capacity in dynamic crop monitoring, cocoa crop detection shows challenges due to cocoa complex canopy structure, spectral similarity to forest, variable farming methods, and location in frequently cloudy regions. Previous research on cocoa crop detection has mainly focused on pixel-based classification, disregarding spatial context. In this research we have performed a semantic segmentation approach to incorporate spatial configuration and enhance cocoa crop detection. We have applied Convolutional Neural Network (CNN) for the to semantic segmentation of cocoa parcels, considering both spectral and spatial characteristics. Additionally, we have evaluated the impact of combining Synthetic Aperture RADAR (SAR) and MSI (Multi-Spectral Imagery) data in the training of a CNN to demonstrate the importance of texture, moisture, and canopy characteristics in identifying cocoa canopies. The impact of MSI dataset stack with different SAR polarizations, seasons and temporality has been evaluated. The methodology is tested on Sentinel 1 and 2 data over an area of 100 × 100 km in Ghana for which an extensive ground truth data set of almost 90,000 polygons was available for training and validation. The results show that the addition of single-day and temporal SAR to a single-day MSI image can improve the predictions, reaching an F1 score of 86.62 %. This research demonstrates the influence of SAR measurements, seasons, polarization, and ground truth classes on the semantic segmentation of cocoa.

Historical aerial imagery provides valuable data from regions and periods with limited geospatial information. A common method to utilize this data is through the generation of ortho-photos and 3D models using Structure-from-Motion (SfM) techniques. However, many of these images were scanned decades after their acquisition and require geometric calibration, along with internal and external camera parameter estimation, for accurate reconstruction. Manual identification of key features, such as fiducial marks and text annotations, is labour-intensive, while existing automated methods struggle with poor-quality datasets. This paper presents an automated workflow that combines computer vision and machine learning techniques to detect and extract these key features from historical aerial images. To address challenges related to image quality, we also introduce estimation protocols that compensate for missing or unreliable detections by leveraging redundancy across multiple flight paths. The methodology was evaluated on the TMA (Trimetrogon Aerial) archive, a collection of historical images from the Antarctic Peninsula. Our test dataset comprised over 7000 images from 20 different flight paths. The workflow demonstrated high success rates in detecting and extracting fiducial marks, image subsets, and textual annotations. Approximately 70% of the images provided usable focal length data, while fiducial mark detection exhibited high accuracy except in cases of severe scanning artifacts. Altitude data extraction proved to be the most challenging, with successful results in only 15% of images due to degraded altimeter readings. Despite these limitations, the automated workflow effectively estimated missing parameters, ensuring robust image reconstruction across flight paths. The code for this workflow is open-source and publicly available on GitHub at https://github.com/fdahle/hist_meta_extraction.

Arsenic contamination in shallow aquifers of Holocene alluvial basins is a serious health risk affecting millions of people [1]. Detection of arsenic hotspots is a slow and tedious process based on the analysis of groundwater samples. This study improves arsenic risk prediction by incorporating geomorphological features such as oxbow lakes and clay plugs into a machine learning (ML) approach. Advances in remote sensing [2], often combined with ML, enable the efficient detection of these and other proxy features, significantly reducing reliance on labour-intensive fieldwork. By combining these features with environmental and demographic data, the approach provides more accurate and cost-effective risk assessments, enabling better-targeted interventions in vulnerable regions and supporting proactive environmental monitoring.

Fitting a smooth curve to 2D, a surface to 3D, and a manifold to 4D irregular point cloud data is becoming a common practice in many engineering and science applications. Piecewise-polynomial spline functions provide a powerful tool applicable to interpolation and approximation problems. This study presents the least squares B-spline approximation (LSBSA) theory, which is a generalized version of the spline interpolation and can be applied to any irregularly scattered point cloud data at knots specified by the user. The formulation allows to apply the well-established body of knowledge of least squares theory to the B-spline approximation. This for example has the benefit of embedding quality control measures such as hypothesis testing and proper error propagation to assess the quality of the approximation problem. The method is applicable to many 1D curve, 2D surface and 3D manifold fitting problems of which both simulated and real data are used to illustrate the efficacy of the proposed theory. In particular, its real-world applications to multi-beam echo-sounder bathymetric data, digital terrain modeling and Greenland ice sheet deformation monitoring will be highlighted. The performance of the method for linear, quadratic, cubic and quartic spline functions will be investigated. The primary application of LSBSA lies in its ability to perform 3D manifold fitting for deformation monitoring. This capability provides the possibility of monitoring changes in continuous spatial and temporal domains. The Python and Matlab source codes of LSBSA are freely accessible at https://github.com/tudelft4d/lsbsa.

Many topographic scenes exhibit complex dynamic behavior that is difficult to map, quantify, predict and understand. A terrestrial laser scanner fixed on a permanent position can be used to monitor such scenes in an automated way with centimeter to decimeter quality at ranges of up to several kilometers. Laser scanners are active sensors, and are therefore able to continue operation during night. Their independence from texture conditions ensures that in principle they provide stable range measurements for varying surface conditions. Recent years have seen a strong increase in the employment of such systems for different scientific applications in geosciences, environmental and ecological sciences, including forestry, glaciology, and geomorphology. At the same time, this employment resulted in a new type of 4D topographic data sets (3D point clouds + time) with a significant temporal dimension, as systems are now able to acquire thousands of consecutive epochs in a row. Extracting information from these 4D data sets turns out to be challenging, first, because of insufficient knowledge on error budget and correlations, and, second, because of lack of algorithms, benchmarks, and best-practice workflows. This paper provides an overview of different 4D systems for near-continuous laser scanning, and discusses systematic challenges including instability of the sensor system, meteorological and atmospheric influences, and data alignment, before discussing recently developed methods and scientific software for extracting and parameterizing changes from 4D topographic data sets, in connection to the different applications.

Satellite-derived bathymetry (SDB) provides a cost-effective solution for coastal mapping, but challenges remain in model interpretability and uncertainty quantification. This study investigates the applicability of the least-squares-based deep learning (LSBDL) framework for SDB, leveraging its hybrid structure that integrates neural networks with the available least-squares theory to enhance model transparency. ICESat-2 photon-counting LiDAR was used to train depth estimation from Sentinel-2 multispectral imagery over an approximately 30 km × 30 km region of near-coastal bathymetry at Anegada, British Virgin Islands. ICESat-2 provided high-precision depth information, of which 80% were used for training and the remainder for validation. LBSDL depth estimation achieved a root-mean-square error (RMSE) of 2.74 m, representing around 10% of the maximum observed depth, with the best performance in the 2–15 m depth range. These findings demonstrate the potential of LSBDL for interpretable and reliable bathymetric mapping, highlighting ICESat-2 as a globally accessible training and validation source and advancing SDB capabilities for data-sparse coastal regions.

Monitoring freshwater resources is essential for assessing the impacts of drought, water management and global warming. Spaceborne LiDAR altimeters allow researchers to obtain water height information, while water area and precipitation data can be obtained using different satellite systems. In our study, we examined 5 years (2018–2022) of data concerning the Euphrates–Tigris Basin (ETB), one of the most important freshwater resources of the Middle East, and the water bodies of both the ETB and the largest lake of Türkiye, Lake Van. A multi-sensor study aimed to detect and monitor water levels and water areas in the water scarcity basin. The ATL13 product of the Ice, Cloud, and Land Elevation Satellite-2 (ICESat-2) was used to determine water levels, while the normalized difference water index was applied to the Sentinel-2 optical imaging satellite to monitor the water area. Variations in both water level and area may be related to the time series of precipitation data from the ECMWF Reanalysis v5 (ERA5) product. In addition, our results were compared with global HydroWeb water level data. Consequently, it was observed that the water levels in the region decreased by 5–6 m in many reservoirs after 2019. It is noteworthy that there was a decrease of approximately 14 m in the water level and 684 km2 in the water area between July 2019 and July 2022 in Lake Therthar.

Sandy beach-dune systems make up a large part of coastal areas world wide. Their function as an eco-system as well as a protective barrier for human and natural habitat is under increased threat due to climate change. A thorough understanding of change processes at the sediment surface is essential to facilitate prediction of future development and management strategies to maintain their function. Especially slow and small scale processes happening over several days up to weeks at cm level, such as aeolian sand transport are difficult to identify and analyse. Permanent laser scanning (PLS) is a useful tool in the study and analysis of coastal processes as it captures a data representation of the evolution of the sediment surface over extended periods of time (up to several years) with high detail (at cm-dm level). The PLS data set considered for this study, consists of hourly acquired 3D point clouds representing the surface evolution of a section of the Dutch coast during three years. However, it is challenging to extract concrete information on specific change processes from the large and complex PLS data set. We use multiple hypothesis testing in order to reduce the PLS data set to a so-called inventory of trends, consisting of 12.8 million partial time series with associated rate of change and elevation. The inventory of trends proofs to be a suitable tool to identify natural processes such as storms and aeolian sand transport in our test area in the aeolian zone of a sandy beach-dune system on the Dutch coast. We identify these processes and provide a tool to derive summarising data from the complex PLS data set. We find that all partial time series identified as most likely representing aeolian sand transport, result in 1354 m³ of sand deposition in our study area over the course of three years. We also show a comparison with transects from JarKus data and find a correlation between anthropogenic activities and erosion in our test area with a correlation coefficient of 0.3.

Coastal dunes provide an important role to society by fulfilling various ecosystem functions. These are however under pressure due to sea level rise, climate change and urbanization. Shoreward sand transport can partly mitigate sea level rise and climate effects by contributing to dune growth but increasing urbanization of beaches can block this transport and restoration of ecosystem functions. In this paper, we investigate the long-term effects of buildings and other anthropogenic influences on dune development up to a decadal temporal scale.

A series of three topographic datasets is used to study the effect of anthropogenic actors on local dune development at an urban beach in Noordwijk, the Netherlands. Datasets range from a 100 to 3000-meter spatial scale and from a weekly to yearly temporal scale. On the smallest spatio-temporal scale topographic measurements of the effects of two containers placed on Noordwijk beach are studied. The intermediate dataset is obtained from the 2-year CoastScan project monitoring surface elevation around one beach pavilion at (bi) monthly intervals. Finally, 15 years of annual airborne lidar data along a 2.7-kilometer stretch of the beach/dune system in Noordwijk is used to evaluate the effect of 17 pavilions.

The small-scale experiment shows that horseshoe-shaped deposition patterns developed on the leeside of the containers. These depositions follow daily wind changes and leave deposits corresponding to the residual wind direction over the whole measuring period. Similar patterns are found around the beach pavilion, but, due to anthropogenic influences like bulldozing and beach shaping, longer term patterns in the direct vicinity of the pavilion and the dunes are hard to discern.

Evaluation of the longer term dataset reveals large variations in dune height and volume in the neighborhood of beach pavilions. Dune height/volume increases vary between 1-8 m in dune height and vary between 0-200 m3 in dune volume after 15 years along 2.7 km of coast. An autocorrelation analysis shows that the alongshore variability length scale in dune volume of urbanized dunes can be 10 times smaller than for natural dunes. For about half the beach pavilions, variations in dune height and volume are significantly correlated to the location of the beach pavilion. Here the growth behind the buildings is lower than in the surrounding area which might have consequences for long-term resilience against future climate changes.

The utility of historical image repositories is often limited due to the lack of geo-referencing. A good example is the TriMetrogon Aerial (TMA) archive, a collection of historical aerial images of Antarctica between 1940 and 2000. These images are difficult to use, as their geolocation is only approximately, with location errors in the order of tens of km. This study addresses this challenge by developing an automated geo-referencing workflow that leverages recent advancements in machine-learning-based tie-point matching. We use the algorithm LightGlue, to establish tie-points between geo-referenced Sentinel-2 satellite imagery and historical non-geo-referenced aerial images. To aid the process, we use already known approximate positions of the historical images. Due to the sub-optimal and inhomogeneous quality of the aerial images, only a portion of the images can be geo-referenced directly by matching. For the remaining images, we employed alternative means of geo-referencing, again based on tie-point matching. Out of a subset of 4,459 images located inside the research area, 3,393 images could be geo-referenced, a percentage of 76%. Reasons for the geo-referencing failing are insufficient contrast or an approximate position too far away from the real position. The workflow can easily be applied to historical images from other archives, to enhance the usability of historical image repositories for scientific research.

Open source Global Digital Elevation Models (GDEMs) serve as an important base for studies in geosciences. However, these models contain vertical errors due to various reasons. In this study, data from two Satellite LiDAR altimetry systems, GEDI and ICESat-2, were used to improve the vertical accuracy of GDEMs. Three different machine learning methods, namely an Artificial Neural Network (ANN), Extreme Gradient Boosting (XGBoost), and a Convolutional Neural Network (CNN), were employed to improve existing DEM data with satellite LiDAR data. The methodology was tested in five areas with varying characteristics. Ground control data were selected from high accuracy DEMs generated from Airborne LiDAR and GNSS data. The use of ANN method improved the vertical accuracy of SRTM data from 6.45 to 3.72 m in Test area-4. Similarly, the CNN method demonstrated an improvement in the vertical accuracy of bare ground SRTM data increasing from 3.4 to 0.6 m in Test area-4. In Test area-5, the ANN method improved the vertical accuracy of SRTM data with slopes between 30 and 60%, increasing from 3.8 to 0.5 m. Notably, the results underscore the successful improvement of GDEMs across all test areas.

Inspired by the attractive features of least-squares theory in many practical applications, this contribution introduces least-squares-based deep learning (LSBDL). Least-squares theory connects explanatory variables to predicted variables, called observations, through a linear(ized) model in which the unknown parameters of this relation are estimated using the principle of least-squares. Conversely, deep learning (DL) methods establish nonlinear relationships for applications where predicted variables are unknown (nonlinear) functions of explanatory variables. This contribution presents the DL formulation based on least-squares theory in linear models. As a data-driven method, a network is trained to construct an appropriate design matrix of which its entries are estimated using two descent optimization methods: steepest descent and Gauss–Newton. In conjunction with interpretable and explainable artificial intelligence, LSBDL leverages the well-established least-squares theory for DL applications through the following three-fold objectives: (i) Quality control measures such as covariance matrix of predicted outcome can directly be determined. (ii) Available least-squares reliability theory and hypothesis testing can be established to identify mis-specification and outlying observations. (iii) Observations’ covariance matrix can be exploited to train a network with inconsistent, heterogeneous and statistically correlated data. Three examples are presented to demonstrate the theory. The first example uses LSBDL to train coordinate basis functions for a surface fitting problem. The second example applies LSBDL to time series forecasting. The third example showcases a real-world application of LSBDL to downscale groundwater storage anomaly data. LSBDL offers opportunities in many fields of geoscience, aviation, time series analysis, data assimilation and data fusion of multiple sensors.

Shoreward sand transport and dune development are increasingly influenced by the urbanization of beach-dune systems in the Netherlands. Three topographic datasets, on various spatio-temporal scales, are used to study the effect of standalone buildings on long term local dune development. On the smallest scale, terrestrial laser scans are used to study the geomorphological effects of two sea containers on the beach. On the intermediate scale, the geomorphological effects of a beach pavilion on the local dune development are studied with a 2-year topographic dataset of (bi) monthly permanent laser scans. Finally, 15 yearly airborne lidar scans of the beach-dune system in Noordwijk are used to evaluate the effect of multiple beach pavilions on dune growth variations. The small-scale experiment shows that horseshoe-shaped deposition patterns developed on the leeside of the containers. These depositions follow daily wind changes and leave deposits corresponding to the residual wind direction over the whole measuring period. Similar patterns are found around the larger beach pavilion, but anthropogenic activities like bulldozing and beach shaping make the determination of the effect on dune development harder to discern. Evaluation of the longer-term dataset reveals large variations in dune height and volume around beach pavilions. Dune height/volume increases vary between 1 and 8 m in height and 0–200 m3 in volume. A variability analysis shows that the length scale of alongshore variability in dune height/volume of urbanized dunes can be 10 times smaller than for natural dunes. For about half the beach pavilions, variations in dune height and volume are significantly correlated to the location of beach pavilions but correlation to particular beach pavilion properties is yet inconclusive.

Detailed 3D information on vulnerable archaeological sites can document cultural heritage and contribute to its preservation. The Late Bronze Age Mycenaean cemetery of Aidonia, Greece, is a representative case of a vulnerable site. Tomb looting has occurred sporadically since the 1970s, when the Greek government was made aware of the site. Anthropogenic activities and natural denudation may affect the loss of structural integrity of tombs. In this contribution, terrestrial laser scanning and geosciences are combined to document the vulnerable cemetery through the generation of a tomb catalogue. The emphasis is on techniques applied to point clouds to extract architectural elements. The catalogue consists of 208 architectural and geological measurements, 112 qualitative observations, maps, and point clouds images displaying the architecture of 16 tombs. The tombs are mainly orientated northeast-southwest and northwest-southeast, and their average total length is 13 m. The average volume of chambers with preserved roofs is 46 m3.

The paper presents a new dry-dock method for assessing the deformation of submarine hulls using TLS point cloud data and the point cloud spatial expansion method (PCSE). The advantage of the proposed approach is the high-resolution deformation analysis that can be conducted in the case of both the availability and a lack of technical documentation on the submarine hull. The geometry assessment involves two-plane hull symmetry in longitudinal sections of a tested Kobben-class submarine located in Gdynia, Poland. The features of PCSE introduce additional geometrical parameters that are not available in the original point cloud method. The procedure for local fitting of a plane into the expansion eliminates the problem of varying densities of the hull point cloud. Accuracies of several millimetres are achieved and are applicable to multi-temporal monitoring of the deformations of submarine hulls. The assessment of similar deformations is not possible in the original point cloud method, due to the unknown parameters of the orientation and curvature of the convex cylindrical hull surface. The PCSE-based parameterisation presented here enables the creation of alternative quasi-planar point cloud projections to extract new spatial information on the object. The results of this study were verified using theoretical values derived from design data on the Kobben-class submarine, and demonstrated the effectiveness of our method in terms of detecting deformations even without design references. The proposed methodology is uniform, and can be adapted to other symmetrical structures.

The TriMetrogon Aerial (TMA) archive is an archive of historical images of Antarctica taken by the US Navy between 1940 and 2000 with analogue cameras. The analysis of such historic data can give a view of Antarctica's glaciers predating modern satellite imagery and provide unique insights into the long-term impact of changing climate conditions with essential validation data for climate modelling. However, the lack of semantic information for these images presents a challenge for large-scale computer-driven analysis. Such information can be added to the data using semantic segmentation, but traditional algorithms fail on these scanned historical grayscale images, due to varying image quality, lack of colour information and artefacts in the images. To address this, we present a deep-learning-based U-net workflow. Our approach includes creating training data by pre-processing and labelling the raw images. Furthermore, different versions of the U-net are trained to optimize its hyperparameters and augmentation methods. With the optimal hyper-parameters and augmentation methods, a final model has been trained for a use-case to segment 118 images covering Adelaide Island. We tested our approach by segmenting challenging historical images using a U-net model with just 80 training images, achieving an accuracy of 73% for 20 validation images. While no test data is available for our use case, a visual examination of the segmented images shows that our method performs effectively. The comparison of the hyper-parameters and augmentation methods provides directions for training other U-net-based models so that the presented workflow can be used to segment other archives with historical imagery. Additionally, the labelled training data and the segmented images of the test are publicly available at https://github.com/fdahle/antarctic_segmentation.

Landslides are one of the major weather-related geohazards. To assess their potential impact and design mitigation solutions, a detailed understanding of the slope processes is required. Landslide modelling is typically based on data-rich geomechanical models. Recently, machine learning has shown promising results in modelling a variety of processes. Furthermore, slope conditions are now also monitored from space, in wide-area repeat surveys from satellites. In the present study we tested if use of machine learning, combined with readily available remote sensing data, allows us to build a deformation nowcasting model. A successful landslide deformation nowcast, based on remote sensing data and machine learning, would demonstrate effective understanding of the slope processes, even in the absence of physical modelling. We tested our methodology on the Vögelsberg, a deep-seated landslide near Innsbruck, Austria. Our results show that the formulation of such a machine learning system is not as straightforward as often hoped for. The primary issue is the freedom of the model compared to the number of acceleration events in the time series available for training, as well as inherent limitations of the standard quality metrics such as the mean squared error. Satellite remote sensing has the potential to provide longer time series, over wide areas. However, although longer time series of deformation and slope conditions are clearly beneficial for machine-learning-based analyses, the present study shows the importance of the training data quality but also that this technique is mostly applicable to the well-monitored, more dynamic deforming landslides.

Church towers are key cultural heritage. In theory, towers are vertical, while facade elements are symmetrically positioned around the tower axis. However, during service of a structure, building and lifetime conditions cause deviations, with associated risks. Laser scanning point clouds can be used to assess the structural state but a universal approach was missing. The proposed algorithm first estimates the tower inclination, and tests which multi-axis representation best represents the course of the tower. Next, point cloud spatial expansion recovers relative distances and deviations of facade elements. The resulting procedure was applied to assess two Dutch medieval towers including the Old Church in Delft and the St. Bavo Church in Haarlem, respectively. As results of analysis, significant asymmetry was found with a 1.4° deviation of the multi-modal axis of the St. Bavo Church tower together with variations of 0.1%–1.5% for facade slopes, while 0.1°–3.1° radial deviations were found in the position of the turrets of the Old Church tower.

Sandy beaches are subject to changes due to multiple factors, that are both natural (e.g. storms) and anthropogenic. Great efforts are being made to monitor these ecosystems and understand their dynamics in order to assure their conservation. The identification of anthropogenic changes and its differentiation from natural ones is an important task for coastal monitoring. In this study, we present a methodology for the detection of anthropogenic changes in a coastal ecosystem by automatically detecting active bulldozers in continuous beach video data. PCA is used to highlight changes in consecutive images due to moving objects. Next, the YOLO object detection algorithm is used to identify the bulldozers in the change images. YOLO was specifically trained for the task, obtaining a precision of 0.94 and a recall of 0.81. An automatic tool was developed, and the process was carried out on two months of video data, consisting of approximately 19 000 images. The resulting information was compared with changes derived from 3D data obtained from a permanent laser scanner. The correlation among the results of the two methodologies was computed. For a validation area and daily time frame a correlation of 0.88 was obtained between the number of detected bulldozers and the area affected by changes in height larger than 0.3 m.

Global Ecosystem Dynamics Investigate (GEDI) is a spaceborne laser altimeter system used for earth observation in many areas such as forest canopy, water level and terrain height estimation. GEDI data is affected by atmospheric effects due to the sensor used while observing. In this study, we propose a 7-step, multi-variable strategy for determining the elevation of the terrain with GEDI. These steps involve both different geoid models, GEDI ancillary data, and topographic features. We evaluated the effect of each step using high quality DEM data obtained by Airborne LiDAR over the central part of Puerto Rico, where building areas and forests are dominant, while the terrain has an average slope of 24%. The GEDI data of the test area consists of 3 different orbits (O06225, O07933, O08061) with different solar elevation and cloudiness rates. While the raw data of orbit O06225, obtained during a solar elevation of 8.4 and cloudy conditions, has a Root Mean Square Error (RMSE) of 418.67 m., the RMSE is reduced to 4.59 m. after applying all seven filtering steps. The raw data of orbit O07933, obtained with a solar elevation of 50.5 during cloud free conditions, has a RMSE of 10.04 m., and is reduced to a similar value of 4.8 m. as a result of the filtering steps. On the other hand, orbit O08061 was obtained with little clouds during a near-dawn solar elevation of -0.7. Its raw RMSE of 50,34 m could only be reduced to 12.41 m. by the proposed filtering procedure. It is concluded that although there are many outliers in data acquired during cloudy conditions, the accuracy of the data remaining after applying our filtering strategy can be as high as the accuracy obtained during cloud free conditions. Better results than 5 m were obtained according to the RMSE in areas with low solar elevation. In addition, it is observed that accuracy decreases strongly when the solar elevation is close to 0. Overall, it is concluded that appropriate filtering is required when determining terrain height with GEDI data.