Fig 5 - available via license: Creative Commons Attribution 4.0 International
Content may be subject to copyright.
Comparison of the results of four classification techniques (SVM-129, NN-36, ML-PCA, ML-MNF) on a single sample plot (Alonzo et al. 2014). Legend: red-birch, orange-European beech, yellow-oak, pink-hornbeam, pale blue-European larch, green-Scots pine, dark blue-Norway spruce
Source publication
Remote sensing techniques and data are becoming increasingly popular in forest management, e.g. for change detection and health condition analysis. Tree species recognition is a fundamental issue in taking forest inventories, especially in carbon budget modelling. Hyperspectral imagery provides an accurate classification results for large areas bas...
Similar publications
It is critical to acquire the information of forest type at the tree species level due to its strong links with various quantitative and qualitative indicators in forest inventories. The efficiency of deep-learning classification models for high spatial resolution (HSR) remote sensing image has been demonstrated with the ongoing development of arti...
Citations
... To address these ecologically relevant questions terrestrial inventories should be used to complement remote sensing-based analyses (Tomppo et al., 2008). Repeated multi-and hyperspectral campaigns can further aid the analysis of compositional responses to gap formation (Hycza et al., 2018;Modzelewska et al., 2020). Furthermore, given the time scales of mountain forest dynamics, our observation period is short (12 years). ...
Natural disturbances are important drivers of forest dynamics, and canopy gaps are their fingerprints in forest ecosystems. Gaps form and persist because of the interplay of tree mortality and regeneration. They can have long‐lasting impacts on ecosystems, yet the temporal dynamics of gap formation and closure remains poorly quantified.
We analysed 11,331 canopy gaps and their changes through time across 3999 ha of unmanaged temperate mountain forests at Berchtesgaden National Park (Germany). We assessed gap formation and closure using three repeat lidar acquisitions between 2009 and 2021, analysing canopy height changes at 1 m horizontal resolution. Our objective was to determine the dominant mode of gap formation, distinguishing the creation of new gaps from the expansion of existing ones. Additionally, we studied the rate of gap closure, considering closure from tree regeneration and lateral crown expansion.
Gap formation was primarily driven by gap expansion rather than the initiation of new gaps. Gap expansion accounted for 81.3% of gap formation, although new gaps were on average twice as large as gap expansions. Only 1.4% of gaps did not expand over the 12‐year study period, and Norway spruce forests had the highest rate of gap expansion.
Overall, gap closure rate (0.74 ha 100 ha ⁻¹ year ⁻¹ ) was higher than gap formation (0.58 ha 100 ha ⁻¹ year ⁻¹ ) in our study system. Ingrowth of the regenerating tree cohort was the primary mode of gap closure, with lateral crown expansion accounting for 20% of all gap area closed. Mixed‐species stands had the highest rate of gap closure, and gaps <0.1 ha closed faster than larger gaps.
Synthesis . While canopy openings are generally small in the European Alps, we show that they keep growing over multiple years, underlining that gap expansion is an important driver of temperate forest dynamics. Canopy gaps closed faster than they were created, highlighting the resilience of European mountain forests to natural disturbances. However, as disturbances are projected to increase under climate change, this resilience might be challenged in the future, requiring a continuous monitoring of gap dynamics as an important early warning indicator of forest change.
... The light detection and ranging (LiDAR) sensor can accurately map vertical forest canopy structure, including the tree canopy, the ground below the trees, and the space between canopy and ground (Brede et al., 2017;Wieser et al., 2017). A hyperspectral camera is used for species classification (Hycza et al., 2018;Tusa et al., 2019). Multispectral cameras can help evaluate the health status of forest stands through vegetation indices (Klouček et al., 2019;Junttila et al., 2022). ...
Three-dimensional (3D) mapping and unmanned aerial vehicles (UAVs) are essential components of the future development of forestry technology. Regeneration of forest stands must be ensured according to the law in the required quality and species composition. Forest management focuses on the optimization of economic costs and quality-assured seedlings. Predicting the suitability of the plots’ environment for natural forest regeneration can contribute to better strategic planning and save time and money by reducing manual work. Although the savings may be considered negligible on small forested plots, they are significant for large cleared areas, such as those harvested after large beetle infestations or strong windstorms, which are increasingly common in European forests. We present a methodology based on spatial analysis and 3D mapping to study the microrelief and surrounding of recently cleared areas. We collected data on four plots in the spring and autumn of a single year after the harvest of four Norway spruce [Picea abies (L.) Karst.] stands near Radlice, Czechia using a multirotor Phantom 4 Pro UAV with a red, green, blue (RGB) camera. We used RGB imagery to compute microrelief data at a very high spatial resolution and the surrounding forest stands after harvesting. We used the microrelief data to estimate the amount of water accumulation and incoming solar radiation across the sites. Based on presence data of newly-established seedlings, we used linear mixed effects models to create a suitability map for each site. Model variables included topographic wetness index, solar area radiation, fencing, type of soil preparation, and distance to the nearest mature forest edge. The topographic wetness index and fencing had strong positive influence on seedling establishment, while solar radiation had a negative influence. Our proposed methodology could be used to predict spontaneous regeneration on cleared harvest areas, or it can estimate how much area is suitable for regeneration, which can lead to important investment decisions.
... The use of hyperspectral imagery (HSIs) combined with machine learning techniques exhibits the potential to distinguish different plant species [2,6,10]. The research by Badola et al. [11] leveraged hyperspectral data and a modified version of the random forest classifier aligned with Principal Component Analysis (PCA) to map tropical tree species. ...
The application of hyperspectral imagery coupled with deep learning shows vast promise in plant species discrimination. Reshaping one-dimensional (1D) leaf-level reflectance data (LLRD) into two-dimensional (2D) grayscale images as convolutional neural network (CNN) model input demonstrated marked effectiveness in plant species distinction. However, the impact of the image shape on CNN model performance remained unexplored. This study addressed this by reshaping data into fifteen distinct rectangular formats and creating nine CNN models to examine the effect of image structure. Results demonstrated that irrespective of CNN model structure, elongated narrow images yielded superior species identification results. The ‘l’-shaped images at 225 × 9 pixels outperformed other configurations based on 93.95% accuracy, 94.55% precision, and 0.94 F1 score. Furthermore, ‘l’-shaped hyperspectral images consistently produced high classification precision across species. The results suggest this image shape boosts robust predictive performance, paving the way for enhancing leaf trait estimation and proposing a practical solution for pixel-level categorization within hyperspectral imagery (HSIs).
... At the same time, many questions regarding the reliability of the classification of woody plants remain unresolved. Large-scale classification of woody plants remains a fundamental challenge in green space monitoring (Hycza et al. 2018). In several experiments, the result of the classification of objects turned out to be unsatisfactory (Xu et al. 2020). ...
Invasion of alien plant species entails great ecological, economic, and social consequences. One of the effective ways to manage invasive species may be to account for and destroy them using unmanned aerial vehicles. However, this requires learning to identify invasive species using real-time remote sensing. Recently, great hopes for solving this problem have been placed on hyperspectral cameras. In this regard, there is a need to fundamentally answer the question of the possibility of identifying plant species from spectral data, regardless of the time of data acquisition. The aim of the study was to identify four species of woody plants by the time series of spectral characteristics of their leaves, obtained using a hyperspectral camera. The study was conducted in laboratory conditions, in which the number of unaccounted for factors is much less than in the field. The objects of study were one native species Quercus robur L. and three species invasive for Europe – Fraxinus pennsylvanica Marsh., Ailanthus altissima (Mill.) Swingle, Parthenocissus inserta (A. Kern.) Fritsch. The collection of leaves for Hyperspectral Imaging (HSI) was carried out during the growing season of the plants at intervals of 7–10 days. Random Forest (RF) was chosen as the object classification method. The RF pixel-based test was carried out both for specific calendar dates (time slice) and for the time series when the RF model was trained on the data of one calendar date and tested on other calendar dates. None of the RF testing methods was able to classify all four species simultaneously with sufficient probability (more than 90%). Therefore, RF testing of combinations of two samples was used – "species" & "other three species" for all calendar dates. This approach made it possible to classify species by spectral characteristics of leaves with 100% reliability. It has been established that the spectral bands informative for RF pixel-based classification lie in the visible range of the spectrum in the range from 462 to 478, from 510 to 534, and from 566 to 646 nm.
... Hyperspectral imagery contains more information about vegetation and can be used for more accurate mapping of tree species. The essential condition is that the tree species exhibit significant differences in spectral reflectance measured across multiple spectral bands (Farreira et al., 2016;Hycza et al., 2018). The capability to succesfully classify tree species using such data has been demonstrated in equatorial forests, where classifications of seven tree species achieved accuracies ranging from 80 to 100% (Clark et al., 2005;Peerbhay et al., 2013). ...
Introduction
Mapping tree species is an important activity that provides the information necessary for sustainable forest management. Remote sensing is a effective tool that offers data at different spatial and spectral resolutions over large areas. Free and open acces Sentinel satellite imagery and Google Earth Engine, which is a powerful cloud computing platform, can be used together to map tree species.
Methods
In this study we mapped tree species at a local scale using recent Sentinel-1 (S-1) and Sentinel-2 (S-2) time-series imagery, various vegetation indices (Normalized Difference Vegetation Index - NDVI, Enhanced Vegetation Index - EVI, Green Leaf Index - GLI, and Green Normalized Difference Vegetation Index - GNDVI) and topographic features (elevation, aspect and slope). Five sets of data were used, in different combinations, together with the Random Forest classifier in order to determine seven tree species (spruce, beech, larch, fir, pine, mixed, and other broadleaves [BLs]) in the studied area.
Results and discussion
Dataset 1 was a combination of S-2 images (bands 2, 3, 4, 5, 6, 7, 8, 8a, 11 and 12), for which an overall accuracy of 76.74% was obtained. Dataset 2 comprised S-2 images and vegetation indices, leading to an overall accuracy of 78.24%. Dataset 3 included S-2 images and topographic features, which lead to an overall accuracy of 89.51%. Dataset 4 included S-2 images, vegetation indices, and topographic features, that have determined an overall accuracy of 89.36%. Dataset 5 was composed of S-2 images, S-1 images (VV and VH polarization), vegetation indices, and topographic features that lead to an overall accuracy of 89.68%. Among the five sets of data, Dataset 3 produced the most significant increase in accuracy, of 12.77%, compared to Dataset 1. Including the vegetation indices with the S-2 images (Dataset 2) gave an accuracy increase of only 1.50%. By combining the S-1 and S-2 images, vegetation indices and topographic features (Dataset 5) there was an accuracy increase of only 0.17%, compared with the S-2 images plus topographic features combination (Dataset 3). However, the input brought by the S-1 images was apparent in the increase in classification accuracy for the mixed and other BL species that were mostly found in hilly locations. Our findings confirm the potential of S-2 images, used together with other variables, for classifying tree species at the local scale.
... For example, HSI has multifaceted applications in remote sensing, including environmental monitoring [6], mineral exploration [7], agriculture [8], forestry [9], and urban planning [8]. The data obtained from hyperspectral imaging can be used to map and identify different land cover types, detect changes in vegetation, monitor water quality, and assess soil characteristics. ...
... The test scene in Figure 11 is a typical 3 × 3 × 3 toy cube with bricks of 6 different colors (white, yellow, red, blue, orange, and green), where the spectrum is binned at 14 different centers showing a clear distinction of different slices. The intensity binning is performed similarly to (9), but with limits that only extend into the bin's range, as followsĨ ...
With advancements in computer processing power and deep learning techniques, hyperspectral imaging is continually being explored for improved sensing applications in various fields. However, the high cost associated with such imaging platforms impedes their widespread use in spite of the availability of the needed processing power. In this paper, we develop a novel theoretical framework required for an open source ultra-low-cost hyperspectral imaging platform based on the line scan method suitable for remote sensing applications. Then, we demonstrate the design and fabrication of an open source platform using consumer-grade commercial off-the-shelf components that are both affordable and easily accessible to researchers and users. At the heart of the optical system is a consumer-grade spectroscope along with a basic galvanometer mirror that is widely used in laser scanning devices. The utilized pushbroom scanning method provides a very high spectral resolution of 2.8 nm, as tested against commercial spectral sensors. Since the resolution is limited by the slit width of the spectroscope, we also provide a deconvolution method for the line scan in order to improve the monochromatic spatial resolution. Finally, we provide a cost-effective testing method for the hyperspectral imaging platform where the results validate both the spectral and spatial performances of the platform.
... However, many questions regarding the reliability of the determination of tree species remain open. Large-scale recognition of tree species remains a fundamental problem in monitoring (Hycza et al., 2018). In several experiments, the identification result turned out to be unsatisfactory (Xu et al., 2020). ...
... With the rapid increase in population in our developing and consuming world, the need for forests and forest products is increasing. Therefore, whether natural forests or industrially grown forests, they all need sustainable management (Hycza et al., 2018). Species recalculations may also be needed after future forest crop planning, deforestation as a result of illegal cutting, damage to forests as a result of fires or natural disasters (Xi et al., 2021;Zagajewski et al., 2021). ...
... A systematic review was made by examining the articles written in English published in peer-reviewed journals containing technological developments and methods that have occurred in recent years. In this review study we prepared for the determination of tree species; we used recent studies that have exploited hyperspectral images, multispectral images, and their combination (Hycza et al., 2018;Rumora et al., 2020;Abbas et al., 2021;Hati et al., 2021;Yang et al., 2022). ...
... In this study, 17 articles which use only MS images (Javan et al., 2021;Xi et al., 2021;Wang et al., 2022) 13 articles which use only HS images (Franklin and Ahmed, 2018;Hycza et al., 2018;Abbas et al., 2021) and 4 articles which use fusion of both images (Hati et al., 2021;Yang et al., 2022) for the detection of tree species were reviewed as shown in Figure 1. ...
Classification of tree species provides important data in forest monitoring, sustainable forest management and planning. The recent developments in Multi Spectral (MS) and Hyper Spectral (HS) Imaging sensors in remote sensing have made the detection of tree species easier and accurate. With this systematic review study, it is aimed to understand the contribution of using the Multi Spectral and Hyper Spectral Imaging data in the detection of tree species while highlighting recent advances in the field and emphasizing important directions together with new possibilities for future inquiries. In this review, researchers and decision makers will be informed in two different subjects: First one is about the processing steps of exploiting Multi Spectral and HS images and the second one is about determining the advantages of exploiting Multi Spectral and Hyper Spectral images in the application area of detecting tree species. In this way exploiting satellite data will be facilitated. This will also provide an economical gain for using commercial Multi Spectral and Hyper Spectral Imaging data. Moreover, it should be also kept in mind that, as the number of spectral tags that will be obtained from each tree type are different, both the processing method and the classification method will change accordingly. This review, studies were grouped according to the data exploited (only Hyper Spectral images, only Multi Spectral images and their combinations), type of tree monitored and the processing method used. Then, the contribution of the image data used in the study was evaluated according to the accuracy of classification, the suitable type of tree and the classification method.
... Currently, HSI classification is used to solve many Earth remote sensing problems, such as identifying tree species [1,2], estimating crop yields [3,4], and oil spill detection [5,6]. Good HSI classification results are obtained if there is sufficient labeled data for training. ...
... The second layer RPNet features are obtained by assuming that (1) F is a new input H and performing the first-layer actions. In a similar manner, features in the lth layer can be obtained for all [ ...
... is the (l -1)th layer features, then lth layer features ( ) l F can be obtain by assuming that ( 1) l F − is a new input H and using the feature extraction process from the first layer. ...
In recent years, different deep learning frameworks were introduced for hyperspectral image (HSI) classification. However, the proposed network models have a higher model complexity, and do not provide high classification accuracy if few-shot learning is used. This paper presents an HSI classification method that combines random patches network (RPNet) and recursive filtering (RF) to obtain informative deep features. The proposed method first convolves image bands with random patches to extract multi-level deep RPNet features. Thereafter, the RPNet feature set is subjected to dimension reduction through principal component analysis (PCA), and the extracted components are filtered using the RF procedure. Finally, the HSI spectral features and the obtained RPNet–RF features are combined to classify the HSI using a support vector machine (SVM) classifier. In order to test the performance of the proposed RPNet–RF method, some experiments were performed on three widely known datasets using a few training samples for each class, and classification results were compared with those obtained by other advanced HSI classification methods adopted for small training samples. The comparison showed that the RPNet–RF classification is characterized by higher values of such evaluation metrics as overall accuracy and Kappa coefficient.
... Currently, HSI classification is used to solve many Earth remote sensing problems, such as identifying tree species [1,2], estimating crop yields [3,4], and oil spill detection [5,6]. Good HSI classification results are obtained if there is sufficient labeled data for training. ...
... is the mean vector of M in the second dimension. Finally, (1) rc k F × ∈ is reshaped to (1) ...
... is the mean vector of M in the second dimension. Finally, (1) rc k F × ∈ is reshaped to (1) ...
In recent years, different deep learning frameworks were introduced for hyperspectral image (HSI) classification. However, the proposed network models have a higher model complexity and do not provide high classification accuracy if few-shot learning is used. This paper pre-sents an HSI classification method that combines random patches network (RPNet) and re-cursive filtering (RF) to obtain informative deep features. The proposed method first convolves image bands with random patches to extract multi-level deep RPNet features. Thereafter, the RPNet feature set is subjected to dimension reduction through principal component analysis (PCA) and the extracted components are filtered using the RF procedure. Finally, HSI spectral features and the obtained RPNet-RF features are combined to classify the HSI using a support vector machine (SVM) classifier. In order to test the performance of the proposed RPNet-RF method, some experiments were performed on three widely known datasets using a few training samples for each class and classification results were compared with those obtained by other advanced HSI classification methods adopted for small training samples. The comparison showed that the RPNet-RF classification is characterized by higher values of such evaluation metrics as overall accuracy and Kappa coefficient (https://github.com/UchaevD/RPNet-RF).
