Kaan Kalkan has a total of 38 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Open Access

    Assessing the Impact of Spatial Resolution and Hyperparameters on Automatic Agricultural Parcel Delineation Using the Segment Anything Model With Multi‐Resolution and Super‐Resolved Satellite Imagery

    by Fatih Fehmi Şimşek, Melih Altay, Kaan Kalkan, Mehmet Cengiz Arslanoğlu

    Accurate and automatic determination of the boundaries of agricultural parcels is essential for planning sustainable agricultural policies, managing agricultural production, tracking products and enabling precision farming applications. In recent years, pre‐trained visual base models with zero‐shot segmentation capabilities, such as the Segment Anything Model (SAM), have introduced a new paradigm to the field. Integrating such models with satellite images of varying characteristics opens up new possibilities in terms of both accuracy and operational efficiency. While traditional satellite systems such as Sentinel‐2, PlanetScope, and Pleiades provide data at specific resolution ranges, super‐resolution (SR) like S2DR3 and Sen2SR, which have been developed in recent years, can enhance the spatial detail of existing data for analysis. In this study, we tested satellite images with different spatial resolutions for the automatic extraction of agricultural parcel boundaries using the SAM model. We also evaluated the performance and contribution of super‐resolution datasets like S2DR3 and Sen2SR. Additionally, the impact of the model's hyperparameter configurations on segmentation accuracy was analyzed, and pre‐ and post‐processing steps were applied to ensure that only agricultural parcel areas were obtained. The findings revealed that resolution directly impacts segmentation performance, though not in a linear manner. Although Sen2SR images have a higher spatial resolution than PlanetScope data, the former produced weaker results, particularly with regard to geometric error metrics (GOSE: 0.318, GUSE: 0.373). Although Pleiades images achieved the highest numerical accuracy values (intersection over union (IoU): 0.942; Dice coefficient: 0.971), they were found to be limited in practical applications due to issues such as over segmentation, processing time, and cost. In contrast, S2DR3 data provided balanced segmentation results, both visually and numerically (IoU: 0.920; DICE: 0.958), and was found to be the most operationally applicable solution due to its relatively low computational demand and short processing time. Hyperparameter analyses showed that accuracy increased by up to 40% in low‐resolution Sentinel‐2 images. However, improvements beyond the default configuration were limited in high‐resolution datasets (points_per_side: 64; crop_n_layers: 2). Considering the trade‐off between processing time and accuracy, the default configurations (points_per_side = 32 and crop_n_layers = 0) were found to provide sufficient accuracy for many scenarios. In conclusion, this study demonstrates that powerful base models such as SAM can be effectively applied to satellite images enhanced with super‐resolution techniques, and that synthetic data such as S2DR3 offer the most operationally feasible solutions when criteria such as cost, hardware, processing time, and accuracy are considered. The study provides practical recommendations for practitioners regarding the resolution–hyperparameter relationship and makes practical contributions to large‐scale or resource‐constrained agricultural monitoring scenarios.


  • Open Access

    Assessment of the Performance of Sentinel-2A MSI and Landsat 9 OLI Images in Land Cover/Use Classification by Comparing Machine Learning Algorithms: A Case Study of Soma District, Türkiye

    by Sajaweddin Sadid, Kaan Kalkan

    The combination of rapid, uncontrolled population growth and economic and industrial development has significantly accelerated land use/land cover (LULC) changes. Assessing these changes is one of the most effective ways to understand and manage land transformation. The advancement of remote sensing technology and increased accessibility to satellite data have made it more feasible to produce accurate and up-to-date LULC maps through the development of classifier algorithms. This has enabled better assessment and management of ecosystem and land use changes. The main objective of this study is to evaluate the performance of four machine learning algorithms—RF, SVM, CART, and GTB—using Sentinel-2 and Landsat 9 satellite images for the Soma district of Türkiye, on the Google Earth Engine (GEE) platform. In the study, a kernel function was applied to the SVM algorithm. Downloaded satellite images were visually inspected, and Google Earth Pro images were utilized to create training and test samples. Sentinel-2 and Landsat 9 images were classified using these training data and machine learning algorithms on the GEE platform. In the evaluation of the results, an error matrix was generated for the classified images, using the test samples for validation. The evaluation showed that the overall accuracy of the SVM algorithm, using the kernel function, was 92.6% for Sentinel-2 and 87% for Landsat 9, placing it third in terms of accuracy. The GTB algorithm provided the highest overall accuracy, with 94.4% for Sentinel-2 and 88.3% for Landsat 9. The RF algorithm achieved 93.2% accuracy for Sentinel-2 and 87% for Landsat 9, matching the accuracy of SVM for Landsat 9. CART demonstrated the lowest performance, with 86.4% accuracy for Sentinel-2 and 91.4% for Landsat 9. Additionally, Sentinel-2 imagery performed better than Landsat 9 across all algorithms due to its higher spatial resolution and spectral characteristics. This study provides valuable insights for local and provincial planners, authorities, and decision-makers regarding proper land management and the production of reliable LULC maps, especially in mining regions.


  • Open Access

    Hava Aracı Hareketliliği ve Havalimanlarındaki Hava Kalitesi Arasındaki İlişkinin İncelenmesi: Aerosol İndeksi ve Avrupa Örneği

    by Eray Kaçar, Kaan Kalkan

    Bu çalışma, Avrupa'nın en yoğun altı havalimanında (İstanbul (LTFM), Amsterdam Schiphol (EHAM), Paris Charles de Gaulle (LFPG), Londra Heathrow (EGLL), Frankfurt (EDDF) ve Madrid Barajas (LEMD)) Ocak 2023 - Kasım 2024 arasındaki hava trafiği ile atmosferik aerosol indeksi (AAI) arasındaki ilişkiyi incelemektedir. Sentinel-5P uydusundan alınan AAI verileri kullanılarak, her havalimanı için hava trafiği ve AAI değerleri, zamansal değişimleri ve aralarındaki korelasyonlar analiz edilmiştir. Bu çalışma özelinde, hava trafiği ile AAI arasında genel ve belirgin bir ilişki saptanmamıştır. Bununla birlikte, havalimanı bazında yapılan incelemeler, bu ilişkinin her bir havalimanında farklı derecelerde ve bazı durumlarda düzensiz bir şekilde ortaya çıktığını göstermiştir. Madrid Barajas ve Paris Charles de Gaulle'de ilişki zayıf ve düzensizken, Amsterdam Schiphol, Londra Heathrow ve Frankfurt'ta belirgin bir negatif, İstanbul'da ise zayıf bir pozitif ilişki gözlenmiştir. Endüstriyel faaliyetler, kentsel ısınma ve diğer ulaşım kaynakları gibi faktörlerin AAI değerlerine etkisi çalışma kapsamı dışında bırakılmıştır. Sonuç olarak, hava trafiği ve AAI ilişkisi havalimanlarına göre değişmekte ve Sentinel-5P verileri bu tür analizlerde değerli bir araç sunmaktadır.


  • Analysis of the Threat of Forest Fires to Ancient Cities by GIS and Remote Sensing Methods

    by Ezgi Tükel, Kaan Kalkan

    Forest fires have been more common in recent years and caused extensive damage. Not only settlements and natural life but also historical places and ancient cities are at risk of forest fires. This study discussed forest fires in Turkey in the summer of 2021. Forest fire risk classifications were determined using Landsat-8 images. The Normalized Burn Ratio (NBR) and Differenced Normalized Burn Ratio (dNBR) indices were used to assess the area impacted by fire and to create fire risk classes. Furthermore, the burned and unburned areas in different zones from the Amos ancient city in the Marmaris were calculated using remote sensing methods. Thus, areas that should be protected from the fire were determined in future studies for fire risk areas.


  • Classification-based and rule-based methods for cloud detection in high resolution satellite imagery

    by Mehmet Efendioğlu, Savaç Özkan, Caner Demirpolat, Mustafa Teke, Kaan Kalkan

    Cloud cover ratio in electro-optical satellite images is a critical factor for the usability of the images. First step to assess the usability of satellite images is cloud detection for the estimation of a cloud coverage from the analyses and/or removing the clouds from images. Satellite imagery providers supply these cloud ratio and maps for their satellite imagery. In this work, classification-based and rule-based methods are compared for cloud detection in high-resolution satellite imagery. In particular, test images consist of different scenarios such as regular, stereo and cloudy imagery. From the experiments, it has been observed that deep learning-based methods could achieve significant cloud detection performances.

  • Ground Truth generation from automatic recognition of ground Control Points

    by Onur Aydın, Kaan Kalkan, Levent Özparlak

    Production of Ground Truth (GT) data for remotely sensed images is an important step for evaluation of accuracy assessment of different analyzes such as classification, geo-locationing and object detection. Geo-locationing systems for images taken from different platforms such as ground and laboratory tripod are not widespread and useful. Lack of peer pixel coordinates of image pixels is an absence for creating GT from imagery. Ground Control Points (GCP) which have regular shape and texture are used for creation of pixel coordinates from images taken from terrestrial photogrammetry and indoor photogrammetry applications. Projects related to geomatics and architecture are using these GPSs in many studies. In this study, world coordinates of GCPs and pixels coordinates of GCPs are matched with each other for reproduction of target coordinates in imagery. This study provides a fast approach for creation of ground truth data for remotely sensed images.

  • Vegetation discrimination analysis from Göktürk-2 imagery

    by Kaan Kalkan, Bayram Giray Filiz, Orhun Ozer, Mustafa Teke

    Remote Sensing is multi-disciplinary science, which provides important coverage, mapping and classification of land cover features, such as vegetation, soil, water and forests. Vegetation is important as it provides a basic foundation for all living beings. Classifying vegetation is so valuable that people can determine vegetation distribution and occurrence. Remote sensing can also be used to detect and prevent the spread of damaged and stressed plants. Satellite imageries are capable of performing these type of vegetation analysis. On 18 December 2012, the first domestic high resolution reconnaissance and surveillance satellite of Turkey, Göktürk-2 was launched and started to meet the high resolution image requirements of Turkish Armed Forces and also has become a data source for scientific activities such as control of forestland, tracking illegal construction, rapid assessment of damage after natural disasters, determination of agricultural boundaries and geographical data gathering [1]. The purpose of this study is to test vegetation discrimination performance of the first domestic high resolution reconnaissance and surveillance satellite of Turkey Göktürk-2 imagery and compare results with Landsat-8 imagery.