Enhancing tree species composition mapping using Sentinel-2 and multi-seasonal deep learning fusion

Abstract. Different spatial resolutions satellite imagery with global almost daily revisit time provide valuable information about the earth surface in a short time. Based on the remote sensing methods satellite imagery can have different applications like environmental development, urban monitoring, etc. For accurate vegetation detection and monitoring, especially in urban areas, spectral characteristics, as well as the spatial resolution of satellite imagery is important. In this research, 10-m and 20-m Sentinel-2 and 3.7-m PlanetScope satellite imagery were used. Although in nowadays research Sentinel-2 satellite imagery is often used for land-cover classification or vegetation detection and monitoring, we decided to test a fusion of Sentinel-2 imagery with PlanetScope because of its higher spatial resolution. The main goal of this research is a new method for Sentinel-2 and PlanetScope imagery fusion. The fusion method validation was provided based on the land-cover classification accuracy. Three land-cover classifications were made based on the Sentinel-2, PlanetScope and fused imagery. As expected, results show better accuracy for PS and fused imagery than the Sentinel-2 imagery. PlanetScope and fused imagery have almost the same accuracy. For the vegetation monitoring testing, the Normalized Difference Vegetation Index (NDVI) from Sentinel-2 and fused imagery was calculated and mutually compared. In this research, all methods and tests, image fusion and satellite imagery classification were made in the free and open source programs. The method developed and presented in this paper can easily be applied to other sciences, such as urbanism, forestry, agronomy, ecology and geology.

Monitoring within-field crop variability at fine spatial and temporal resolution can assist farmers in making reliable decisions during their agricultural management; however, it traditionally involves a labor-intensive and time-consuming pointwise manual process. To the best of our knowledge, few studies conducted a comparison of Sentinel-2 with UAV data for crop monitoring in the context of precision agriculture. Therefore, prospects of crop monitoring for characterizing biophysical plant parameters and leaf nitrogen of wheat and barley crops were evaluated from a more practical viewpoint closer to agricultural routines. Multispectral UAV and Sentinel-2 imagery was collected over three dates in the season and compared with reference data collected at 20 sample points for plant leaf nitrogen (N), maximum plant height, mean plant height, leaf area index (LAI), and fresh biomass. Higher correlations of UAV data to the agronomic parameters were found on average than with Sentinel-2 data with a percentage increase of 6.3% for wheat and 22.2% for barley. In this regard, VIs calculated from spectral bands in the visible part performed worse for Sentinel-2 than for the UAV data. In addition, large-scale patterns, formed by the influence of an old riverbed on plant growth, were recognizable even in the Sentinel-2 imagery despite its much lower spatial resolution. Interestingly, also smaller features, such as the tramlines from controlled traffic farming (CTF), had an influence on the Sentinel-2 data and showed a systematic pattern that affected even semivariogram calculation. In conclusion, Sentinel-2 imagery is able to capture the same large-scale pattern as can be derived from the higher detailed UAV imagery; however, it is at the same time influenced by management-driven features such as tramlines, which cannot be accurately georeferenced. In consequence, agronomic parameters were better correlated with UAV than with Sentinel-2 data. Crop growers as well as data providers from remote sensing services may take advantage of this knowledge and we recommend the use of UAV data as it gives additional information about management-driven features. For future perspective, we would advise fusing UAV with Sentinel-2 imagery taken early in the season as it can integrate the effect of agricultural management in the subsequent absence of high spatial resolution data to help improve crop monitoring for the farmer and to reduce costs.

Grassland fractional vegetation cover (FVC) accurate mapping on a large scale is crucial, since degraded grasslands contribute less to provisioning services, carbon storage, water purification, erosion control and biodiversity conservation. The spatial and temporal resolution of Sentinel-2 (S2) and PlanetScope (PS) data has never been explored for grassland FVC estimation so far and will enable researchers and agencies to quantify and map timelier and more precisely grassland processes. In this paper we compare FVC estimation models developed from Landsat-8 (L8), S2 and PS imagery. The reference grassland FVC dataset was obtained on the Paganella ski runs (46.15°N, 11.01°E, Italy) applying unsupervised classification to nadir grassland RGB photographs taken from 1.35 m above the soil. Fractional Response Models between reference FVC and 18 vegetation indices (VIs) extracted from satellite imagery were fitted and analysed. Then, leave-one-out cross validation and spatiotemporal change analysis were also performed. Our study confirms the robustness of the commonly used VIs based on the difference between NIR and the red wavelength region (R2 = 0.91 for EVI using S2 imagery) and indicate that VIs based on the red-edge spectral region are the best performing for PS imagery (R2 = 0.89 for RECI). Only medium to high spatial resolution imagery (S2 and PS) precisely mapped spatial patterns at the study site, since grasslands FVC varies at a fine scale. Previously available imagery at medium to low spatial and temporal resolution (e.g., L8) may still be interesting for analysis requiring long time-series of data.

Deep learning for multi-modal classification of cloud, shadow and land cover scenes in PlanetScope and Sentinel-2 imagery

تولید مدل رقومی زمین با قدرت تفکیک و دقت ارتفاعی بالا همیشه یکی از مهم ‌ترین اهداف سنجش از دور ماهواره ‌‌ای بوده است. یکی از ارکان اصلی سنجش از دور ماهواره ‌‌ای، سنجش از دور راداری می‌ باشد. تولید مدل ارتفاعی رقومی از سطح زمین با استفاده از تداخل ‌سنجی راداری به علت ویژگی ‌های منحصر به فرد این تصاویر برای محققین جذاب است. در سال‌های اخیر پروژه ‌های فضایی بسیاری آغاز به اخذ اطلاعات از سطح کره زمین کرده ‌اند که یکی از آخرین آنها پروژه سنتینل می‌ باشد. سنتینل-1 بخش راداری پروژه سنتینل است. مدل ‌های رقومی حاصل از تداخل ‌سنجی راداری به علت وجود خطاهای متنوع از جمله خطا در اطلاعات فازاینترفروگرام دارای خطا و گاهی اوقات اشتباه بزرگ در نقاط ارتفاعی می ‌باشند. از اینرو مدل ‌های رقومی حاصل از فرآیند تداخل ‌سنجی راداری پس از تولید نیاز به بهبود دارند. در این مقاله روشی برای بهبود مدل ‌رقومی ارتفاعی به دست آمده از تصاوی رسنتینل-1 با استفاده از مدل رقومی ارتفاعی موجود SRTM.(Shuttle Radar Topography Mission)و روشی بر اساس تبدیل موجک دو بعدی، پیشنهاد می ‌شود. تصاویر مورد استفاده در این مقاله بخشی از شمال شهر تهران است. مدل ارتفاعی رقومی تولید شده با استفاده از روش پیشنهادی با مدل ارتفاعی رقومی مرجع یک متر با دقت ارتفاعی بالا مورد ارزیابی قرار می ‌گیرد. نتایج مقاله نشان می ‌دهند که روش پیشنهادی به شکل مؤثری در بهبود دقت مدل رقومی حاصل از تصاویر سنتینل-1 عمل می ‌کند. با استفاده از این روش خطای مدل رقومی ارتفاعی به میزان قابل توجهی کاهش می ‌یابد (30% الی 82%) و این بدین معنی می ‌باشد که با حفظ قدرت تفکیک مدل رقومی حاصل از تصاویر سنتینل-1 می ‌توان دقت ارتفاعی آن را به شکل محسوسی بهبود داد.

Accurate forest tree-species classification not only provides data support for forest resource management but also serves as a crucial parameter for simulating various ecological processes. However, the results of forest tree-species classification have been affected by multiple factors, such as the spectral resolution, spatial resolution, and radiometric resolution of imagery, the classification algorithms used, the sample size, and the timing of image acquisition phases. Although there are many studies on the impact of individual factors on tree-species classification, there is a lack of systematic studies quantifying the magnitude of these factors’ influences, leading to uncertainties about the relative importance of different factors. In this study, Landsat-8, Landsat-9, and Sentinel-2 imagery was used as the foundational data, and random forest (RF), gradient tree boosting (GTB), and support vector machine (SVM) algorithms were employed to classify forest tree species. High-accuracy regional forest tree-species classification was achieved by exploring the impacts of spectral resolution, spatial resolution, radiometric resolution, classification algorithms, sample size, and image time phases. The results show that, for the commonly used Landsat-8, Landsat-9, and Sentinel-2 imagery, the tree-species classification results from Landsat-9 are the best, with an overall accuracy of 74.21% and a kappa of 0.71. Among the various influencing factors, the classification algorithm, image time phases, and sample size have relatively larger impacts on tree-species classification results, each exceeding 10%, while the positive impact of radiometric resolution is the smallest, at only 3.15%. Conversely, spectral and spatial resolutions had negative effects on tree-species classification results, at −4.09% and −1.4%, respectively. Based on the 30-m spring Landsat-9 and Sentinel-2 imagery, with 300 samples for each tree-species category, the classification results using the RF algorithm were the best, with an overall accuracy of 87.07% and a kappa coefficient of 0.85. The results indicate that different factors have different impacts on forest tree-species classification results, with classification algorithms, image time phases, and sample size having the largest impacts. Higher spatial and spectral resolutions do not improve the classification accuracy. Therefore, future studies should focus on selecting appropriate classification algorithms, sample sizes, and images from seasons with greater tree differences to improve tree-species classification results.

It is important for the electricity transmission and distribution (T&D) companies to patrol their own assets frequently in a wide area. however, the cost of patrolling throughout the area is budget threatening. The work on detecting the maintenance places where the vegetation encroachment problems occurred, is labor intensive, costly, and time-consuming, sometimes inapplicable due to the poor accessibility, and is thus, only practical on relatively small areas. Satellite imagery-based monitoring is reasonable and repeatable; hence it has a potential to replace the helicopter surveillance. Sentinel-2 imagery is one of the most famous satellite imageries with completely free of charge, however, its spatial resolution is relatively lower than high-cost satellite imagery such as PlanetScope or WorldView-3. In this research, we explored the effectiveness of super resolution. The refinement of spatial resolution from 10m/pix to 3.3m/pix (x3 SR) seemed to be extremely useful to assess trigonometric risk assessment, which leveraged the number of the pixels between transmission line and vegetation, and tree height information at the vegetation pixels. We employed the deep learning based super resolution model RDN (Residual Dense Network) to upsample the Sentinel-2 images. The training data is generated from the PlanetScope imagery whose resolution is 3.7m/pix. Deep learning based super resolution is generally effective to get 2-4 times finer resolution, therefore, the PlanetScope imagery is suitable to obtain the RDN model for x3 super resolution. We evaluated the performance of vegetation segmentation performance with and without super resolution in the areas along the transmission line. The experimental results showed that the imagery with super resolution yielded better result than the result without super resolution by 9.3% in weighted F1-score.

Using sentinel-2 satellite imagery to develop microphytobenthos-based water quality indices in estuaries

Detecting changes in land use and land cover (LULC) from space has long been the main goal of satellite remote sensing (RS), yet the existing and available algorithms for cloud classification are not reliable enough to attain this goal in an automated fashion. Clouds are very strong optical signals that dominate the results of change detection if they are not removed completely from imagery. As various architectures of deep learning (DL) have been proposed and advanced quickly, their potential in perceptual tasks has been widely accepted and successfully applied to many fields. A comprehensive survey of DL in RS has been reviewed, and the RS community has been suggested to be leading researchers in DL. Based on deep residual learning, semantic image segmentation, and the concept of atrous convolution, we propose a new DL architecture, named CloudNet, with an enhanced capability of feature extraction for classifying cloud and haze from Sentinel-2 imagery, with the intention of supporting automatic change detection in LULC. To ensure the quality of the training dataset, scene classification maps of Taiwan processed by Sen2cor were visually examined and edited, resulting in a total of 12,769 sub-images with a standard size of 224 × 224 pixels, cut from the Sen2cor-corrected images and compiled in a trainset. The data augmentation technique enabled CloudNet to have stable cirrus identification capability without extensive training data. Compared to the traditional method and other DL methods, CloudNet had higher accuracy in cloud and haze classification, as well as better performance in cirrus cloud recognition. CloudNet will be incorporated into the Open Access Satellite Image Service to facilitate change detection by using Sentinel-2 imagery on a regular and automatic basis.

Atmospheric correction algorithm based on deep learning with spatial-spectral feature constraints for broadband optical satellites: Examples from the HY-1C Coastal Zone Imager

The existence of forests is crucial to the sustainability of life on earth. Automatic forest mapping is necessary to obtain accurate information about the deforestation rate, quantifying, monitoring and mapping. Such information is essential for various schemes to save forest. European satellite Sentinel-2 provide data at thirteen spectral band along with three different spatial resolution level. This satellite data is freely available having medium spatial resolution and faster revisit time, which makes it suitable choice for forest mapping. The objective of this study is automatic mapping of Deciduous and Evergreen Forest by using Sentinel-2 imagery (single date data) in district Dehradun, Uttarakhand, India. Two efficient Machine Learning (ML) approaches have been used for the classification i.e., Random Forest (RF) and k-Nearest Neighbor (k-NN). Sentinel-2 satellite spectral bands (10 m spatial resolution) namely Near-infrared (NIR), visible light band (Blue, Green and Red) are stacked for classification. In this study, overall classification accuracy attained by RF and k-NN is 81.52% (kappa value of 0.759) and 80.84% (kappa value of 0.751) respectively. Results indicates that both classifiers performed well, however, RF achieved slightly higher (+0.68%) accuracy as compared to k-NN classifier. It is found that RF obtained User Accuracy (UA) and Producer Accuracy (PA) 77.53% and 82.85% respectively for Deciduous Forest. Whereas, for Evergreen Forest UA and PA is 82.26% and 77.95% respectively. On the other hand, k-NN achieved UA of 77.33% and PA of 82.25%. For Evergreen Forest attained UA and PA of 81.81% and 76.95% respectively. Results demonstrated that Sentinel-2 multispectral satellite data is highly suitable for mapping of forests.

Multispectral optical and SAR imagery have potential to improve land cover mapping in urban areas. This is because satellite imagery such as Sentinel-1 and Sentinel-2 has high spatial and temporal resolutions. Therefore, multi-seasonal optical imagery can be used to discriminate built-up areas from cropland and bare areas, while SAR imagery can be used to detect built-up structures. In addition, spectral and texture indices derived from Sentinel-2 and Sentinel-1 imagery can be used to improve urban land cover mapping. In this chapter, Sentinel-1 and Sentinel-2 imagery as well as spectral and texture indices will be used for land cover mapping.

AMFIC Web Data Base - A Satellite System for the Monitoring and Forecasting of Atmospheric Pollution

We deal with the reconstruction of the high-resolution (HR) computed tomography (CT) image from the CT projection data (Sinogram). Spatial resolution is one of the important parameters of CT images. Spatial resolution is a measure of how close to each other two objects that can still be distinguished. The spatial resolution of CT images depends on the field of view which in turn depends on the number of projections and the number of samples per projection. One way to increase the spatial resolution of the reconstructed images is to reduce the pixel size; however, this may require hardware alteration. Another way to increase the spatial resolution is to apply super-resolution (SR) technique on the low-resolution (LR) images. The conventional method for resolution enhancement is to apply SR technique as a post-process after LR CT images reconstruction. One drawback of this conventional method is that; performing two consequence steps, reconstruction and SR, requires more parameters to be tuned. Unlike the conventional method, we propose to simultaneously estimate HR image with the reconstruction step from the CT projection data, which indeed reduce the overall processing time and the required tuned parameters. On the other hand, even if the back-projection (BP) method is attractive because of its simplicity and low computational cost, it produces sub-optimal images with respect to artifacts, resolution, and noise. On contrast, iterative image reconstruction allows to easily model constraints and to incorporate prior knowledge which leads to better image quality. Therefore, in this paper, we employ the iterative reconstruction based on regularized Kaczmarz minimization algorithm, for its fast convergence, in the resolution enhancement as a post-process as well as simultaneous reconstruction and resolution enhancement methods.

Accurate, detailed, and up-to-date tree species distribution information is essential for effective forest management and environmental research. However, existing tree species maps face limitations in resolution and update cycle, making it difficult to meet modern demands. To overcome these limitations, this study proposes a novel framework that utilizes existing medium-resolution national tree species maps as ‘weak labels’ and fuses multi-temporal Sentinel-2 and PlanetScope satellite imagery data. Specifically, a super-resolution (SR) technique, using PlanetScope imagery as a reference, was first applied to Sentinel-2 data to enhance its resolution to 2.5 m. Then, these enhanced Sentinel-2 bands were combined with PlanetScope bands to construct the final multi-spectral, multi-temporal input data. Deep learning (DL) model training data was constructed by strategically sampling information-rich pixels from the national tree species map. Applying the proposed methodology to Sobaeksan and Jirisan National Parks in South Korea, the performance of various machine learning (ML) and deep learning (DL) models was compared, including traditional ML (linear regression, random forest) and DL architectures (multilayer perceptron (MLP), spectral encoder block (SEB)—linear, and SEB-transformer). The MLP model demonstrated optimal performance, achieving over 85% overall accuracy (OA) and more than 81% accuracy in classifying spectrally similar and difficult-to-distinguish species, specifically Quercus mongolica (QM) and Quercus variabilis (QV). Furthermore, while spectral and temporal information were confirmed to contribute significantly to tree species classification, the contribution of spatial (texture) information was experimentally found to be limited at the 2.5 m resolution level. This study presents a practical method for creating high-resolution tree species maps scalable to the national level by fusing existing tree species maps with Sentinel-2 and PlanetScope imagery without requiring costly separate field surveys. Its significance lies in establishing a foundation that can contribute to various fields such as forest resource management, biodiversity conservation, and climate change research.