Applied Geomatics最新文献

Climate change, the expansion of impervious areas, and poorly planned infrastructures have significantly impacted Cyprus, resulting in severe extreme events, such as floods and droughts. Constructing dams in strategically chosen locations is crucial for effective water management, addressing flood control and water scarcity. The Pedieos River basin, characterized by the longest river with the highest flow capacity, is the most critical basin due to its high population density, agricultural and economic activities, and recurring floods. To tackle these challenges, this study aims to create a dam suitability map and identify potential sites for flood mitigation and water storage, utilizing remotely sensed datasets, Geographic Information Systems (GIS), the Analytic Hierarchy Process (AHP), and hydrological models. Initially, the AHP was employed with determined major factors, and a preliminary dam suitability map was produced. The map’s accuracy was subsequently evaluated using existing dam locations, revealing that 93.2% of these dams fell within the moderate to very high suitability zones. Furthermore, the map underwent refinement by applying a Boolean approach that considered six environmental and socioeconomic criteria. This refinement process led to the proposal of eight multi-purpose dams in the Kyrenia and Troodos Mountain ranges. These dams varied in size and capacity, with Dam 3 being the smallest (433,179 m³) and Dam 7 the largest (4,367,512 m³). Lastly, hydrological modeling using HEC-HMS evaluated flood retaining capacities, showing that the dams can effectively handle storms up to a 500-year return period, except for Dam 2, which is limited to a 50-year return period.

Advances in satellite spatial analysis enhance Land Use/Land Cover (LU/LC) assessment accuracy. Over the past four decades, Deep Learning Algorithms have seen substantial progress, especially in the comparison between Pixel-Based (PB) and Object-Based (OB) methods. These developments have greatly enhanced the reliability of satellite data for both scientific research and practical applications. This study investigates LC mapping in regions affected by mining activities, utilizing Earth observation data to explore potential benefits and insights. By conducting a comprehensive review of existing literature, the research aims to clarify the strengths and limitations of using spatial features for accurate LU/LC classification. The study offers an in-depth analysis of PB and OB techniques in satellite image classification, using samples from a false-colour composite image to train and validate Deep Convolutional Neural Networks (DCNNs) and Deep Neural Networks (DNNs) models. OB samples, comprising 6,000 carefully selected 6 × 6-pixel image samples representing various LU types, were compared to PB samples that utilized all pixels within the image samples. The comparison revealed that OB-DCNNs classification achieved a superior accuracy of 97.5%, compared to 91.5% for PB-DNNs classification. These findings highlight the enhanced effectiveness of OB classification for precise LU/LC assessment from satellite imagery.

Currently, Indonesia widely uses Global Navigation Satellite System (GNSS) technology to monitor land subsidence, both continuously and periodically. The main issue with continuous GNSS monitoring is the high cost of the receivers used. To address this challenge, a low-cost GNSS capable of continuous operation for land subsidence monitoring is necessary. This study aims to develop and evaluate a Low-Cost GNSS system for monitoring land subsidence in the Bandung Basin. The research methodology involves prototyping a low-cost GNSS system using a self-assembled U-Blox Chip-GNSS with a microstrip antenna. Observational strategies include radial and network methods over eight months. Analysis and discussion encompass prototyping aspects, observational data quality control, data processing, determination of subsidence rates, and comparison with results from Interferometric Synthetic Aperture Radar (InSAR) methods. The research findings indicate that all observation points experienced subsidence ranging from 9 to 16 cm per year. Radial and network methods show consistent trends with differences ranging between 1 to 2 cm. The GNSS results, which demonstrate patterns and magnitudes similar to those of InSAR data, validate the findings of this research.

This study introduces a novel approach that combines satellite Synthetic Aperture Radar (SAR) measurements with an Artificial Neural Network (ANN) to monitor long-term glacier surface velocities. We focus on the Bara Shigri Glacier in the western Himalayas – a region with limited prior velocity observations – over a six-year period (2017–2023). Using 24 seasonal and 6 annual Sentinel-1 SAR image pairs, we derived glacier velocity maps via sub-pixel offset tracking, and we integrated meteorological variables (2-m air temperature, surface skin temperature, and 0–7 cm soil temperature from ECMWF reanalysis) as inputs to an ANN model. We observed a peak seasonal mean velocity of 8.64 m season⁻¹ (during the April–July 2019 interval) and Interannually, we observed a notable decline in mean velocity from 2017 to 2021, followed by a partial recovery, reflecting the glacier’s dynamic response to climate forcing. Results demonstrate significant seasonal velocity variations, with summer flows approximately 40% faster than winter velocities. While a single train–test split suggested high apparent accuracy (R² = 0.97), 6-fold cross-validation gave weaker generalisation, highlighting overfitting risks given the small dataset. Accordingly, the ANN is presented as a proof-of-concept linking temperature and velocity, rather than a robust predictive tool. This work demonstrates the efficacy of combining SAR remote sensing with machine learning for glacier monitoring, offering a new framework to assess glacier dynamics under changing climatic conditions. Aligned with SDG 13 (Climate Action), this proof-of-concept highlights a potentially scalable pathway for operational glacier monitoring that could support adaptation planning once validated more broadly.

Modern Road Traffic Monitoring (RTM) systems rely on advanced and precise technologies. Unmanned Aerial Vehicles (UAVs) coupled with state-of-the-art computer vision methods offer great utility in intelligent traffic monitoring and road safety analysis. However, the precision of these cutting-edge technologies is still under debate due to technical complexities, such as inaccurate road-user localization resulting in overestimated bounding dimensions, which could hinder their effectiveness in real-world scenarios. This research introduces a geolocalization method combining a feature-matching algorithm, SIFT, for automatic georeferencing of UAV frames with deep learning-based object detection and segmentation models. The study focuses on finding the most precise solution for vehicle geolocalization, preserving the vehicle shape and dimensions. The study explores three different configurations of YOLO object detectors: a standard YOLOv8 model, a hybrid model that integrates YOLOv8 with the Segment Anything Model (SAM), and a YOLOv8 variant that employs Oriented Bounding Boxes (OBB). The evaluation of results is focused on the dimensional accuracy, internal variabilities, impact of altitude variations, vehicle tilt or rotation, and inference speed of each method. Experimental results reveal that the YOLOv8 coupled with SAM and the YOLOv8-OBB exhibit comparable precision and excel in accurately localizing road users while preserving their dimensions. This can be instrumental in a practically feasible vision-based RTM solution. In terms of speed-to-error ratio, OBB-enabled object detectors present the most practical option, allowing for near-real-time solutions in key road safety workflows, such as conflict analysis.

Least squares matching (LSM) is a well-known area-based method for matching between two perspective images. It uses a 2D geometrical transformation to match distorted regions captured from different perspectives. This idea is used here to find a global transformation, contrary to the traditional LSM with its local matching nature, between 3D LiDAR data and 2D high-resolution satellite imagery (HRSI). The proposed method exploits the whole of the overlapped regions between the LiDAR data and the HRSI to find a 3D transformation to relate LiDAR and HRSI. To do so, the radiometric similarity of the HRSI and LiDAR data has also been enhanced by adding shadows and topographic effects to the LiDAR intensity data to act as a proper entity in matching with the HRSI. The results indicate that accurate 3D registration to one-pixel precision can be obtained, on average, even when parameters from approximate 2D transformations are used as initial values.

Floriculture is a vital agricultural sector that significantly contributes to the local and regional economies. Proper identification and mapping of floricultural resources are key to effective planning and crop management, especially in communities whose livelihoods rely on this industry, promoting sustainable growth and stability. This study focused on mapping various flower and crop types in the Debra and Panskura blocks of West Bengal (India) using remote sensing based satellite products and machine learning techniques. Accurate mapping of crop types using conventional satellite datasets remains constrained by inherent limitations, including persistent cloud cover, coarse spatial resolution and related factors. To address this issue, this study utilizes high-resolution multispectral (Sentinel-2) and microwave (Sentinel-1) data collected between November 2022 and April 2023 to improve crop and flower type identification. Various combinations of Sentinel-1 and Sentinel-2 datasets, along with spectral indices Normalized Difference Vegetation Index (NDVI) and Land Surface Water Index (LSWI) were analyzed using two machine learning algorithms—Random Forest (RF) and Support Vector Machine (SVM). Classification performance was assessed based on overall accuracy, kappa coefficient, and standard deviation. Results showed that integrating multiple datasets significantly enhanced accuracy, exceeding 80%, compared to using individual datasets. Among the models evaluated, RF demonstrated superior performance over SVM in accurately identifying different flower and crops. The findings demonstrate that combining remote sensing data with machine learning can reliably distinguish different crops and flower types, providing valuable insights for agricultural management, ecological modeling, and large-scale crop mapping efforts.

At first glance, the grasslands of the Pampa biome may appear somewhat homogeneous. Nevertheless, they contain significant physical, ecological, and cultural wealth that are intricately connected in a complex relationship. However, since the 1950s, their spatial configuration has undergone significant changes due to the introduction of new land-use practices. Escalating this situation is the low rate of environmental protection and conservation in the biome. In this context, it is crucial to identify the different landscapes that compose the Pampa biome based on their physical and ecological characteristics. This article aims to determine whether the Pampa is homogeneous in its spatial extent by presenting the landscape units at the level of geocomplexes within the Pampa biome. These units were defined using the geosystem approach. Initially, a hierarchical framework of the Pampa biome landscape was established according to the taxo-chorological proposal. To identify the geocomplexes, multiresolution segmentation (GEOBIA) was applied based on topographic attributes to generate structurally similar objects. Physical-ecological information was then integrated into these objects. A total of 135 geocomplexes were identified, reflecting the biome’s complexity and diversity. The substantial number of these units demonstrates that the Pampa biome is not homogeneous but rather exhibits a rich array of attributes that distinguish the various geocomplexes. The landscape units provide a foundation for environmental and territorial planning. These plans must reflect the diverse realities that coexist within the Pampa biome to ensure that planning and protection measures are as effective as possible.

The present study proposes the deep learning-enabled U-Net model for crop classification required in crop area calculation, crop status monitoring, and food production. The traditional approaches are time-consuming, costly, and insufficient. Therefore, the Sentinel-2 images were initially pre-processed in this study, and spectral features were extracted from time-series data using the Normalized Difference Vegetation Index (NDVI) method. Then, the composite of Sentinel-2 bands and NDVI output was generated for data stacking. Lastly, classification algorithms such as random forest, artificial neural network (ANN), and U-Net models were implemented on staked data based on training samples. In this case, we used cloud-based open-source Google Earth Engine (GEE) and Google Colab programming for crop classification using Sentinel-2 L2A (S2) images. The results depict that the U-Net model is superior in terms of overall accuracy (94.8%) than the random forest (91.2%) and ANN (93%), with a kappa statistic of 89%, 85.7%, and 87.9%, respectively, for crop classification. The present study is suitable for farmland and crop status monitoring for future food security and decision-making.

The Zenith Total Delay (ZTD) is one of the main sources of error associated with the neutrosphere in Global Navigation Satellite System (GNSS). This error can be estimated through GNSS processing and determined via in situ measurements, such as those obtained from radiosondes and other techniques. The zenith delay, caused by the neutrosphere, is influenced by local atmospheric conditions. In Brazil, owing to its proximity to the equator and the presence of the Amazon rainforest, the atmospheric behavior is distinctive and highly variable. This makes the country particularly affected by this effect, which is an important focus of related studies. In this study, ZTD maps were generated for the Brazilian territory via data from 31 radiosonde stations and 31 GNSS stations, covering a period of nine years (2014–2022). The observations used were at 0 and 12 UTC (Coordinated Universal Time). The analysis involved comparing ZTD values obtained from radiosondes and the GNSS and evaluating the data quality via both methods. Kriging techniques were employed to generate delay maps, both for a general scenario and for different seasons. The quality of the maps was validated via the cross-validation technique, which achieved an approximately 98% coefficient of determination and a root-mean-square error of 1.62 cm.