No full-text available
To read the full-text of this research,
you can request a copy directly from the authors.
Pléiades satellite images for deriving forest metrics in the Alpine region
Abstract
The landscape-human relationships on the Alps, the more populated mountain region globally, depend on tree species diversity, their canopy height and canopy gaps (soil cover). The monitoring of such forest information plays an important role in forest management planning and therefore in the definition of present and future mountain forest services. In order to gain wide scale and high-resolution forest information, very high-resolution (VHR) stereo satellite imagery has the main benefit of covering large areas with short repetition intervals.
However, the application of this technology is not fully assessed in terms of accuracy in dynamic year-around forest conditions. In this study, we investigate on four study sites in the Swiss Alps 1) the accuracy of forest metrics in the Alpine forests derived from VHR Pléiades satellite images and 2) the relation of associated errors with shadows, terrain aspect and slope, and forest characteristics. We outline a grid-based approach to derive the main forest metrics (descriptive statistics) from the canopy height models (CHMs) such as the maximum height (Hmax), height percentiles (Hp95, Hp50), the standard deviation of the height values (HStd) and canopy gaps. The Pléiades-based forest metrics are compared with those obtained by aerial image matching, a technology operationally
used for deriving this information. For the study site with aerial and satellite images acquired almost at the same time, this comparison shows that the medians of Pléiades forest metrics error are -0.25m (Hmax), 0.33m (Hp95), −0.03m (HStd) and -5.6% for the canopy gaps. The highest correlation (R2=0.74) between Pléiades and aerial canopy gaps is found for very bright areas. Conversely, in shadowed forested areas a R2 of only 0.16 is obtained. In forested areas with steep terrain (> 50°), Pléiades forest metrics show high variance for all the study areas. Concerning the canopy gaps in these areas, the correlation between Pléiades and the reference data provides a correlation value of R2=0.20, whereas R2 increases to 0.66 for gently sloped areas (10- 20°). The aspect does not provide a significant correlation with the accuracy of the Pléiades forest metrics.
However, the extended shadowed mainly on north/northwest facing slopes caused by trees or terrain shade negatively affect the performance of stereo dense image matching, and hence the forests metrics. The occurrence of strong shadows in the forested areas increases dramatically by ˜40% in the winter season due to the lower sun elevation. Furthermore, due to the leaf-off condition in the winter season dense image matching may fail to derive the canopy heights. Our results show that Pléiades CHMs could be a useful alternative to CHMs based on aerial images matching for monitoring forest metrics and canopy gaps in mountain forests if captured during leaf-on conditions.
Our study offers forest research, as well as forest management planning, the benefit of a better understanding of the performance of VHR satellite imagery used for forest inventory in mountainous regions and in similar forest environments.
... As an example, Grant D. Pearse et al. (2018) [12] compared point clouds obtained from Pleiades tri-stereo imagery to LiDAR data to predict Pinus radiata forest plot inventory attributes, such as mean height (R 2 = 0.81; RMSE = 2.1 m) and total stem volume (R 2 = 0.70; RMSE = 112.6 m 3 ha −1 ). Additionally, L. Piermattei et al. (2019) [13] used Pleiades tri-stereo image-based CHMs to derive forest metrics in the Alpine region, compared to airborne image matching. Based on the applied pixel-wise approach, the forest metrics median errors −0.25 m (H max ), 0.33 m (H p 95), −0.03m (H Std ) showed that satellite-based Pléiades CHMs could be an alternative to airborne images-derived CHMs in mountain forests. ...
... As an example, Grant D. Pearse et al. (2018) [12] compared point clouds obtained from Pleiades tri-stereo imagery to LiDAR data to predict Pinus radiata forest plot inventory attributes, such as mean height (R 2 = 0.81; RMSE = 2.1 m) and total stem volume (R 2 = 0.70; RMSE = 112.6 m 3 ha −1 ). Additionally, L. Piermattei et al. (2019) [13] used Pleiades tri-stereo image-based CHMs to derive forest metrics in the Alpine region, compared to airborne image matching. Based on the applied pixel-wise approach, the forest metrics median errors −0.25 m (H max ), 0.33 m (H p 95), −0.03m (H Std ) showed that satellite-based Pléiades CHMs could be an alternative to airborne images-derived CHMs in mountain forests. ...
... Remote Sens. 2021,13, 2941 ...
The present study assessed the large-format airborne (UltraCam) and satellite (GeoEye1 and Pleiades1B) image-based digital surface model (DSM) performance for canopy height estimation in predominantly mature, closed-canopy Latvian hemiboreal forestland. The research performed the direct comparison of calculated image-based DSM models with canopy peaks heights extracted from reference LiDAR data. The study confirmed the tendency for canopy height underestimation for all satellite-based models. The obtained accuracy of the canopy height estimation GeoEye1-based models varied as follows: for a pine (−1.49 median error, 1.52 m normalised median absolute deviation (NMAD)), spruce (−0.94 median, 1.97 m NMAD), birch (−0.26 median, 1.96 m NMAD), and black alder (−0.31 median, 1.52 m NMAD). The canopy detection rates (completeness) using GeoEye1 stereo imagery varied from 98% (pine) to >99% for spruce and deciduous tree species. This research has shown that determining the optimum base-to-height (B/H) ratio is critical for canopy height estimation efficiency and completeness using image-based DSMs. This study found that stereo imagery with a B/H ratio range of 0.2–0.3 (or convergence angle range 10–15°) is optimal for image-based DSMs in closed-canopy hemiboreal forest areas.
... The selection of all thresholds for defining shadow was based on visual assessment of the shadow fraction in the orthophotos and the distribution of values in the selected bands. This is a common practice in remote sensing studies (Shahtahmassebi et al., 2013;Waser et al., 2014a), along with visual delineation of big dark shadow areas (Piermattei et al., 2019) or shadow modelling approaches (Sarabandi et al., 2004;Polewski et al., 2015c) as shadow thresholds depend on data and differ between flights and study areas according to the site and flight conditions. ...
... VHR stereoscopic satellite imagery such as WorldView-3 (European-Space-Imaging, 2018a), WorldView-4 (European-Space-Imaging, 2018b) or Pléiades (Coeurdev and Gabriel-Robe, 2012) imagery data might be an alternative to aerial images for mapping high-resolution forest structures such as deadwood or canopy gaps, especially when the methods are intended to be used at bigger spatial scales (Pluto-Kossakowska et al., 2017;Piermattei et al., 2019). More research is required on the combined use of satellite stereoscopic data for the generation of both spectral and structural information for analyses and applications in ecology. ...
The growing demand for bio-resources, expanding and diversifying human impacts on multiple use forests, together with effects of climate change and aerial nitrification permanently alter forests and their structure with consequences for forest biodiversity. The need to integrate forest biodiversity conservation into forest management in order to halt biodiversity loss is of highest relevance.
Given the limitations in assessing and monitoring species diversity at extensive spatial scales, the development of structural indicators linking biodiversity components (e.g. indicator species) to forest structure parameters and enabling monitoring of structural conditions is a widely discussed approach. Until recently this approach has been hampered by the lack of area-wide forest structural data. The growing availability of remotely sensed information now offers the possibility to assess forest structures across different spatial scales, from single trees to the landscape level. However, this requires the development of methods and metrics to assess and describe structural gradients and quantify the links between these metrics and selected biodiversity components.
Until recently the assessment of forest structures used mostly in forest nature conservation research was based either on terrestrial sampling and forest inventory or on visual interpretation of stereo aerial images or orthophotos. Nowadays, with changes from analogue to digital aerial imagery and a growing diversity of remote sensing (RS) data varying in resolution and extent, the processing methods focus primarily on automatic computing. This allows for the processing a large amount of data with objective and reproducible data analyses methods, and for adjustable algorithm parameters depending on the aim of the study.
In my doctoral thesis I focus on the value of remote sensing data and techniques for forest ecology research, combining the methodological development of forest structure detection methods and their application in habitat modelling for forest focal species. The methodological focus of the thesis lies on the detection of two forest structures considered highly relevant for forest biodiversity: canopy gaps (Chapter I) and standing deadwood (Chapter III). Regularly updated digital stereo aerial imagery data of state surveys (subsequently referred to as aerial imagery) and the derivatives thereof produced by Image Matching (Digital Surface Models (DSMs) and orthophotos) were used and evaluated as primary input data, as they could support a cost-efficient long-term monitoring of structural conditions and their changes. An emphasis however was put on the development of algorithms that could also be fed with data of another origin, and flexibly adjusted to the different ecological thresholds required by different taxa.
In the first study (Chapter I) an automatized gap mapping method based on Canopy Height Models (CHMs) derived from DSMs from aerial imagery and a Digital Terrain Model (DTM) based on Aerial Laser Scanning (ALS) is presented. Gaps were detected and delineated in relation to height and cover of the surrounding forest using a hybrid pixel and neighborhood based hierarchic procedure for data from two public flight campaigns (2009 and 20129). Gaps were detected with high overall accuracy (OA) of 0.9 (2009) and 0.82 (2012) and a producer’s accuracy (PA) of more than 0.95 (both years), as validated by visual stereo-interpretation. Lower user’s accuracy (UA) of 0.84 (2009) and 0.73 (2012) indicated an omission error (as some gaps were not detected) that could be attributed to shadow occurrence and the height of the surrounding forest stands, with UA dropping to 0.70 (2009) and 0.52 (2012) in stands with mean vegetation heights of ≥ 8m. Open forest stands were mapped as an important interim step and side-product, as they also may be of importance for photophilic species. With an OA = 0.92 and uncertainties occurring mostly in areas of intermediate forest cover, the models for detecting this forest structure class showed high reliability. Shadow occurrence and geometric limitations of the central perspective of the aerial imagery, with resulting restrictions regarding e.g. viewing angle and image distortions towards the outer parts of an image, were recognized as the main sources of errors. To achieve a potential improvement I recommended using stereo imagery with higher overlap and resolution together with enhanced image-matching algorithms.
In Chapter II I provide a greater in-depth analysis of this topic, by explicitly addressing the limitations of aerial imagery when used as input for the detection of canopy gaps based on the method described in Chapter I. The limitations of aerial imagery become obvious when attempting to accurately map fine structures as well as areas between trees or close to the ground. To evaluate the factors affecting the mapping accuracy, gap detection maps based on data from three flight campaigns differing in image overlap and spatial and radiometric resolution were compared, each covering an approx. 1000-ha study area in the Black Forest, Germany. Gap mapping based on aerial imagery of higher spatial resolution and overlap delivered more detailed gap maps and showed higher detection accuracies. The results confirmed shadow occurrence and geometric limitations of the aerial imagery as serious issues influencing the accuracy of a CHM and consequentially the gap mapping results. Both of them can be improved by harmonizing flight times and associated solar altitude when planning flight campaigns over forested areas. Increasing the spatial resolution and overlap of the aerial imagery could considerably enhance gap detectability especially in the transition areas between high and low vegetation.
In the third study (Chapter III) I present a method for detecting standing deadwood from orthophotos and CHMs using the same aerial imagery data, with a special focus on solving the problem of misclassification between deadwood and bare ground pixels. Due to deadwood mainly occurring in extensively managed forests, as well as its frequent association with canopy openings and open and complex structured stands located in rugged terrain, bare ground is often visible nearby. Having a similar spectral signature both classes are thus prone to misclassifications. Both spectral (orthophoto) and structural (CHM) predictor variables were tested for detecting standing deadwood of more than 5 m in height. The method was calibrated in a mountain forest area encompassing strictly protected and managed forests with a significant amount of deadwood in different decay stages. In a first step, Random Forest (RF) classification was employed to assign forest pixels to one of four classes: live, declining and dead trees as well as bare ground. Two enhancing procedures, aiming at eliminating misclassifications, were then developed and compared 1) post-processing, based on morphological rules filtering out potentially misclassified deadwood pixels and isolated pixels of all classes and 2) a “deadwood-uncertainty” model quantifying and predicting the probability of a deadwood-pixel to be correctly classified based on the environmental conditions and image texture in its neighborhood. Validation of the RF model based on data partitioning delivered both UA and PA over 0.9. Independent validation on stratified random sample, however, revealed a high commission error for deadwood mainly in areas with bare ground (UA = 0.60, PA = 0.87). Both enhancement-procedures, post-processing (1) and the “uncertainty filter” (2) improved the differentiation between the two classes and led to a more balanced relation between UA and PA of deadwood (UA = 0.69 and PA = 0.79 for (1) and UA = 0.74 and PA = 0.80 for (2)), with the filtering based on the uncertainty model (2) resulting in a substantially greater improvement.
The final chapter (Chapter IV) presents a case study on employing deadwood detection in habitat selection modelling. RS data is increasingly used for generating habitat variables that describe forest structures relevant for protected forest species, as it offers the unique potential to provide high resolution information over large geographic extents. We generated RS-based variables, especially information on standing deadwood, to model habitat suitability of the Three-toed woodpecker (Picoides tridactylus) in the Bavarian Forest National Park. Combined information from ALS and color-infrared (CIR) aerial imagery delivered tree-based information, such as tree type (broadleaved, coniferous), status (living, dead), as well as several tree-related metrics (e.g. tree height, projected crown area, tree volume, diameter at breast height and basal area). Generalized Additive Models (GAM) based on the single tree polygon-data, aggregated across multiple, species-relevant scales, showed that at least 8 dying or recently dead trees (the latter indicated by an average branch length of at least 2 m) within a 100 m surrounding were necessary to support woodpecker presence. In addition, and for the first time, an adverse effect of very large deadwood amounts (more than 40 - 55 dead trees within 100 m) on woodpecker occurrence was shown, making a significant contribution to the knowledge about Three-toed woodpecker ecology. The case study illustrated the great potential of the RS data to deliver reliable and meaningful input parameters for habitat models and to derive habitat thresholds that are easily applicable in forest management.
In summary, this thesis addressing several novel aspects of using the aerial imagery data in ecology research, confirms that public aerial imagery and the data products thereof, such as orthophotos and CHMs, enable the detection of forest structures and deriving ecologically relevant variables valuable for biodiversity studies. However, the methodological studies (Chapters I, II and III) showed limitations with regard to the accuracy of the vegetation heights derived from image matching and the detectability of forest gaps and standing deadwood structures in areas between tall trees and between high and low vegetation. In my thesis, I analyze these emerging problems, propose potential solutions (e.g. two alternative approaches for detecting and correcting deadwood and bare ground misclassifications, Chapter III) and discuss future research needs that would enable successful habitat modelling and deriving meaningful thresholds for forest structures to support forest biodiversity conservation (Chapter IV) based on data from aerial imagery and ALS .
... The improved satellite-based DSMs can be further used to compute object heights [70] given a high resolution DTM, but also in many forestry applications, for example to detect tree growth [71,72]. ...
The number of high and very high resolution (VHR) optical satellite sensors, as well as the number of medium resolution satellites is continuously growing. However, not all high-resolution optical satellite imaging cameras have a sufficient and stable calibration in time. Due to their high agility in rotation, a quick change in viewing direction can lead to satellite attitude oscillation, causing image distortions and thus affecting image geometry and geo-positioning accuracy. This paper presents an approach based on re-projection of regularly distributed 3D ground points from object in image space, to detect and estimate the periodic distortions of Pléiades tri-stereo imagery caused by satellite attitude oscillations. For this, a hilly region was selected as a test site. Consequently, we describe a complete processing pipeline for computing the systematic height errors (deformations, waves) of the satellite-based digital elevation model by using a Lidar high resolution terrain model. Ground points with fixed positions, but with two elevations (actual and corrected) are then re-projected to the satellite images with the aid of the Rational Polynomial Coefficients (RPCs) provided with the imagery. Therefore, image corrections (displacements) are determined by computing the differences between the distinct positions of corresponding points in image space. Our experimental results in Allentsteig (Lower Austria) show that the systematic height errors of satellite-based elevation models cannot be compensated with an usual or even high number of Ground Control Points (GCPs) for RPC bias correction, due to insufficiently known image orientations. In comparison to a reference Lidar Digital Terrain Model (DTM), the computed elevation models show undulation effects with a maximum height difference of 0.88 m in along-track direction. With the proposed method, image distortions in-track direction with amplitudes of less than 0.15 pixels were detected. After applying the periodic distortion compensation to all three images, the systematic elevation discrepancies from the derived elevation models were successfully removed and the overall accuracy in open areas improved by 33% in the RMSE. Additionally, we show that a coarser resolution reference elevation model (AW3D30) is not feasible for improving the geometry of the Pléiades tri-stereo satellite imagery.
... The method could also be tested on VHR satellite imagery data such as WorldView-3 [102], WorldView-4 [103], or Pléiades [104], particularly with regard to large-scale applications [91,105]. More research is required on how to use satellite stereoscopic data for generating both spectral and structural information for analyses and applications in ecology. ...
Deadwood mapping is of high relevance for studies on forest biodiversity, forest disturbance, and dynamics. As deadwood predominantly occurs in forests characterized by a high structural complexity and rugged terrain, the use of remote sensing offers numerous advantages over terrestrial inventory. However, deadwood misclassifications can occur in the presence of bare ground, displaying a similar spectral signature. In this study, we tested the potential to detect standing deadwood (h > 5 m) using orthophotos (0.5 m resolution) and digital surface models (DSM) (1 m resolution), both derived from stereo aerial image matching (0.2 m resolution and 60%/30% overlap (end/side lap)). Models were calibrated in a 600 ha mountain forest area that was rich in deadwood in various stages of decay. We employed random forest (RF) classification, followed by two approaches for addressing the deadwood-bare ground misclassification issue: (1) post-processing, with a mean neighborhood filter for “deadwood”-pixels and filtering out isolated pixels and (2) a “deadwood-uncertainty” filter, quantifying the probability of a “deadwood”-pixel to be correctly classified as a function of the environmental and spectral conditions in its neighborhood. RF model validation based on data partitioning delivered high user’s (UA) and producer’s (PA) accuracies (both > 0.9). Independent validation, however, revealed a high commission error for deadwood, mainly in areas with bare ground (UA = 0.60, PA = 0.87). Post-processing (1) and the application of the uncertainty filter (2) improved the distinction between deadwood and bare ground and led to a more balanced relation between UA and PA (UA of 0.69 and 0.74, PA of 0.79 and 0.80, under (1) and (2), respectively). Deadwood-pixels showed 90% location agreement with manually delineated reference to deadwood objects. With both alternative solutions, deadwood mapping achieved reliable results and the highest accuracies were obtained with deadwood-uncertainty filter. Since the information on surface heights was crucial for correct classification, enhancing DSM quality could substantially improve the results.
... xxi For many applications highly accurate and up-to-date mapping information gathered from very high resolution (VHR) satellite stereo images is needed. Exemplary applications in the domain of remote sensing are, for instance, city modeling (Duan and Lafarge, 2016;Steinnocher, Perko, and Hofer, 2014;You et al., 2018;Bittner et al., 2018), forest assessment and biomass estimation (Persson, Wallerman, et al., 2013;Persson and Perko, 2016;Persson, 2016;Piermattei, Marty, Karel, et al., 2018;Kothencz et al., 2018;Stylianidis et al., 2019;Piermattei, Marty, Ginzler, et al., 2019), change detection (Abduelmola, 2016;Bagnardi, González, and Hooper, 2016;Warth et al., 2019), land cover and land use classification (Mora et al., 2014;Belgiu, Drǎguţ, and Strobl, 2014;Schardt et al., 2018), carbon reporting (Perko, Hirschmugl, Papst, et al., 2016), farm land monitoring (Sofia et al., 2016), glacier observation (Rieg et al., 2018;Belart et al., 2019), disaster damage mapping (Maxant et al., 2013;Durić et al., 2017), landslide mapping (Leopold et al., 2017), and mapping in general (Capaldo et al., 2012;Bernard et al., 2012;Bosch, Leichtman, et al., 2017;Himmelreich, Ladner, and Heller, 2017;Ladner, Heller, and Grillmayer, 2017;Vanderhoof and Burt, 2018). To allow semantic analysis those applications need mapping products in form of digital surface models (DSM), digital terrain models (DTM), their difference, i.e., normalized digital surface models (nDSM), and the according multi-spectral ortho-rectified image mosaics. ...
An end-to-end workflow for mapping with very high resolution satellite data is the pre-requisite for any further semantic analysis. In specific, many applications in remote sensing need the following 3D mapping products: (1) digital surface model, (2) digital terrain model, (3) normalized digital surface model, and (4) ortho-rectified image mosaic. This thesis describes all underlying principles for satellite-based 3D mapping and proposes methods that extract all those products from multi-view stereo satellite imagery in the ground sampling distance of the input data. The study is based on, but not limited to, the Pleiades satellite constellation. Beside an in-depth review of related work, the methodological part proposes solutions for each component of the end-to-end workflow. In particular, this includes optimization of sensor models represented by rational polynomials, epipolar rectification, image matching, spatial point intersection, data fusion, digital terrain model derivation, ortho rectification, and ortho mosaicing. For each step implementation details are proposed and discussed. Another aim of this thesis is a detailed assessment of the resulting output products. Thus, a variety of data sets showing different acquisition scenarios are gathered, allover comprising 24 Pleiades images. First, the accuracies of the 2D and 3D geo-location are analyzed. Second, surface and terrain models are evaluated, including a critical look on the underlying error metrics, and discussing the differences of single stereo, tri-stereo, and multi-view data sets. Overall, 3D accuracies in the range of 0.2 to 0.3 meters in planimetry and 0.2 to 0.4 meters in height are achieved w.r.t. ground control points. Retrieved surface models show normalized median absolute deviations around 0.9 meters in comparison to reference LiDAR data. Multi-view stereo outperforms single stereo in terms of accuracy and completeness of the resulting surface models.
... Mapping, in general, can be described as the process of generating 3D maps of a region of interest on the Earth's surface consisting of 3D coordinates linked with the spectral information that was observed by the underlying sensor. Exemplary applications in the domain of remote sensing are, for instance, city modeling [1][2][3], forest assessment and biomass estimation [4][5][6], change detection [7,8], land cover and land use classification [9,10], carbon reporting [11], farm land monitoring [12], glacier observation [13,14], disaster damage mapping [15,16] and mapping in general [17,18]. To allow semantic analysis those applications need mapping products in form of digital surface models (DSM), digital terrain models (DTM), their difference, that is, normalized digital surface models (nDSM) and the according multi-spectral ortho-rectified images. ...
In this work, we introduce an end-to-end workflow for very high-resolution satellite-based mapping, building the basis for important 3D mapping products: (1) digital surface model, (2) digital terrain model, (3) normalized digital surface model and (4) ortho-rectified image mosaic. In particular, we describe all underlying principles for satellite-based 3D mapping and propose methods that extract these products from multi-view stereo satellite imagery. Our workflow is demonstrated for the Pléiades satellite constellation, however, the applied building blocks are more general and thus also applicable for different setups. Besides introducing the overall end-to-end workflow, we need also to tackle single building blocks: optimization of sensor models represented by rational polynomials, epipolar rectification, image matching, spatial point intersection, data fusion, digital terrain model derivation, ortho rectification and ortho mosaicing. For each of these steps, extensions to the state-of-the-art are proposed and discussed in detail. In addition, a novel approach for terrain model generation is introduced. The second aim of the study is a detailed assessment of the resulting output products. Thus, a variety of data sets showing different acquisition scenarios are gathered, allover comprising 24 Pléiades images. First, the accuracies of the 2D and 3D geo-location are analyzed. Second, surface and terrain models are evaluated, including a critical look on the underlying error metrics and discussing the differences of single stereo, tri-stereo and multi-view data sets. Overall, 3D accuracies in the range of 0 . 2 to 0 . 3 m in planimetry and 0 . 2 to 0 . 4 m in height are achieved w.r.t. ground control points. Retrieved surface models show normalized median absolute deviations around 0 . 9 m in comparison to reference LiDAR data. Multi-view stereo outperforms single stereo in terms of accuracy and completeness of the resulting surface models.
This paper investigates the velocity-tracking problem of the gimbal system in a single gimbal control moment gyro (SGCMG) suffering from high-frequency disturbance. In order to accurately and quickly reject the high-frequency disturbance resulted from nonuniform mass distribution and manufacturing imperfection, a finite-time harmonic disturbance observer (FTHDO) is constructed according to the available information about the high-frequency disturbance. Then, a composite controller is developed to achieve simultaneous disturbance rejection and attenuation. The nominal control performance can be recovered under the high-frequency disturbance. It has been demonstrated that the velocity-tracking error can be stabilized by the proposed composite controller within finite time. Finally, numerical simulation results are presented to verify the superiority of the proposed method.
Digital stereo aerial photographs are periodically updated in many countries and offer a viable option for the regular update of information on forest variables. We compared the potential of image-based point clouds derived from three different sets of aerial photographs with airborne laser scanning (ALS) to assess plot-level forest attributes in a mountain environment. The three data types used were (A) high overlapping pan-sharpened (80/60%); (B) high overlapping panchromatic band (80/60%); and (C) standard overlapping pan-sharpened stereo aerial photographs (60/30%). We used height and density metrics at the plot level derived from image-based and ALS point clouds as the explanatory variables and Lorey’s mean height, timber volume, and mean basal area as the response variables. We obtained a RMSE = 8.83%, 29.24% and 35.12% for Lorey’s mean height, volume, and basal area using ALS data, respectively. Similarly, we obtained a RMSE = 9.96%, 31.13%, and 35.99% and RMSE = 11.28%, 31.01%, and 35.66% for Lorey’s mean height, volume and basal area using image-based point clouds derived from pan-sharpened stereo aerial photographs with 80/60% and 60/30% overlapping, respectively. For image-based point clouds derived from a panchromatic band of stereo aerial photographs (80%/60%), we obtained an RMSE = 10.04%, 31.19% and 35.86% for Lorey’s mean height, volume, and basal area, respectively. The overall findings indicated that the performance of image-based point clouds in all cases were as good as ALS. This highlights that in the presence of a highly accurate digital terrain model (DTM) from ALS, image-based point clouds offer a viable option for operational forest management in all countries where stereo aerial photographs are updated on a routine basis.
Lichen woodlands (LW) are sparse forests that cover extensive areas in remote subarctic regions where warming due to climate change is fastest. They are difficult to study in situ or with airborne remote sensing due to their remoteness. We have tested a method for measuring individual tree heights and predicting basal area at tree and plot levels using WorldView-3 stereo images. Manual stereo measurements of tree heights were performed on short trees (2–12 m) of a LW region of Canada with a residual standard error of ≈0.9 m compared to accurate field or UAV height data. The number of detected trees significantly underestimated field counts, especially in peatlands in which the visual contrast between trees and ground cover was low. The heights measured from the WorldView-3 images were used to predict the basal area at individual tree level and summed up at plot level. In the best conditions (high contrast between trees and ground cover), the relationship to field basal area had a R2 of 0.79. Accurate estimates of above ground biomass should therefore also be possible. This method could be used to calibrate an extensive remote sensing approach without in-situ measurements, e.g., by linking precise structural data to ICESAT-2 footprints.
National Forest Inventories (NFI) are key data and tools to better understand the role of forests in the global carbon budget. Traditionally inventories have been carried out as field work, which makes them laborious and expensive. In recent years, the development of various remote sensing techniques to improve the cost-efficiency of the NFIs has accelerated. The goal of this study is to determine the usability of open and free multitemporal multispectral satellite images from the European Space Agency's Sentinel-2 satellite constellation and to compare their usability in forest inventories against airborne laserscanning (ALS) and three-dimensional data obtained with high-resolution optical satellite images from WorldView-2 and Synthetic Aperture Radar (SAR) stereo data from TerraSAR-X. Ground reference consisted of field data collected over 74 boreal forest plots in Southern Finland in 2014 and 2016. Features utilizing both single- and multiple-date information were designed and tested for Sentinel-2 data. Due to high cloud cover, only four Sentinel-2 images were available for the multitemporal feature analysis of all reference plots within the monitoring window. Random Forest technique was used to find the best descriptive feature sets to model five forest inventory parameters (mean height, mean diameter at breast height, basal area, volume, above-ground biomass) from all input remote sensing data. The results confirmed that the higher spatial resolution input data correlated with more accurate forest inventory parameter predictions, which is in line with other results presented in literature. The addition of temporal information to the Sentinel-2 results showed limited variation in prediction accuracy between the single and multidate cases ranging from 0.45 to 1.5 percentage points, whereof mean height, basal area and aboveground biomass are lower for single date with relative RMSEs of 14.07%, 20.66% and 24.71% respectively. Diameter at breast height and volume are lower for multi date feature combination with relative RMSEs of 18.38% and 27.21%. The results emphasize the importance of obtaining more evenly distributed data acquisitions over the growing season to fully exploit the potential of temporal features.
Mapping tropical tree species at landscape scales to provide information for ecologists and forest managers is a new challenge for the remote sensing community. For this purpose, detection and delineation of individual tree crowns (ITCs) is a prerequisite. Here, we present a new method of automatic tree crown delineation based only on very high resolution images from WorldView-2 satellite and apply it to a region of the Atlantic rain forest with highly heterogeneous tropical canopy cover – the Santa Genebra forest reserve in Brazil. The method works in successive steps that involve pre-processing, selection of forested pixels, enhancement of borders, detection of pixels in the crown borders, correction of shade in large trees and, finally, segmentation of the tree crowns. Principally, the method uses four techniques: rolling ball algorithm and mathematical morphological operations to enhance the crown borders and ease the extraction of tree crowns; bimodal distribution parameters estimations to identify the shaded pixels in the gaps, borders, and crowns; and focal statistics for the analysis of neighbouring pixels. Crown detection is validated by comparing the delineated ITCs with a sample of ITCs delineated manually by visual interpretation. In addition, to test if the spectra of individual species are conserved in the automatic delineated crowns, we compare the accuracy of species prediction with automatic and manual delineated crowns with known species. We find that our method permits detection of up to 80% of ITCs. The seven species with over 10 crowns identified in the field were mapped with reasonable accuracy (30.5–96%) given that only WorldView-2 bands and texture features were used. Similar classification accuracies were obtained using both automatic and manual delineation, thereby confirming that species’ spectral responses are preserved in the automatic method and thus permitting the recognition of species at the landscape scale. Our method might support tropical forest applications, such as mapping species and canopy characteristics at the landscape scale.
This work focuses on the accuracy estimation of canopy height models (CHMs) derived from image matching of Pléiades stereo imagery over forested mountain areas. To determine the height above ground and hence canopy height in forest areas, we use normalised digital surface models (nDSMs), computed as the differences between external high-resolution digital terrain models (DTMs) and digital surface models (DSMs) from Pléiades image matching. With the overall goal of testing the operational feasibility of Pléiades images for forest monitoring over mountain areas, two questions guide this work whose answers can help in identifying the optimal acquisition planning to derive CHMs. Specifically, we want to assess (1) the benefit of using tri-stereo images instead of stereo pairs, and (2) the impact of different viewing angles and topography. To answer the first question, we acquired new Pléiades data over a study site in Canton Ticino (Switzerland), and we compare the accuracies of CHMs from Pléiades tri-stereo and from each stereo pair combination. We perform the investigation on different viewing angles over a study area near Ljubljana (Slovenia), where three stereo pairs were acquired at one-day offsets. We focus the analyses on open stable and on tree covered areas. To evaluate the accuracy of Pléiades CHMs, we use CHMs from aerial image matching and airborne laser scanning as reference for the Ticino and Ljubljana study areas, respectively. For the two study areas, the statistics of the nDSMs in stable areas show median values close to the expected value of zero. The smallest standard deviation based on the median of absolute differences (σMAD) was 0.80 m for the forward-backward image pair in Ticino and 0.29 m in Ljubljana for the stereo images with the smallest absolute across-track angle (−5.3°). The differences between the highest accuracy Pléiades CHMs and their reference CHMs show a median of 0.02 m in Ticino with a σMAD of 1.90 m and in Ljubljana a median of 0.32 m with a σMAD of 3.79 m. The discrepancies between these results are most likely attributed to differences in forest structure, particularly tree height, density, and forest gaps. Furthermore, it should be taken into account that temporal vegetational changes between the Pléiades and reference data acquisitions introduce additional, spurious CHM differences. Overall, for narrow forward–backward angle of convergence (12°) and based on the used software and workflow to generate the nDSMs from Pléiades images, the results show that the differences between tri-stereo and stereo matching are rather small in terms of accuracy and completeness of the CHM/nDSMs. Therefore, a small angle of convergence does not constitute a major limiting factor. More relevant is the impact of a large across-track angle (19°), which considerably reduces the quality of Pléiades CHMs/nDSMs.
Terrestrial laser scanning (TLS) and unmanned aerial vehicles (UAVs) equipped with digital cameras have attracted much attention from the forestry community as potential tools for forest inventories and forest monitoring. This research fills a knowledge gap about the viability and dissimilarities of using these technologies for measuring the top of canopy structure in tropical forests. In an empirical study with data acquired in a Guyanese tropical forest, we assessed the differences between top of canopy models (TCMs) derived from TLS measurements and from UAV imagery, processed using structure from motion. Firstly, canopy gaps lead to differences in TCMs derived from TLS and UAVs. UAV TCMs overestimate canopy height in gap areas and often fail to represent smaller gaps altogether. Secondly, it was demonstrated that forest change caused by logging can be detected by both TLS and UAV TCMs, although it is better depicted by the TLS. Thirdly, this research shows that both TLS and UAV TCMs are sensitive to the small variations in sensor positions during data collection. TCMs rendered from UAV data acquired over the same area at different moments are more similar (RMSE 0.11–0.63 m for tree height, and 0.14–3.05 m for gap areas) than those rendered from TLS data (RMSE 0.21–1.21 m for trees, and 1.02–2.48 m for gaps). This study provides support for a more informed decision for choosing between TLS and UAV TCMs to assess top of canopy in a tropical forest by advancing our understanding on: (i) how these technologies capture the top of the canopy, (ii) why their ability to reproduce the same model varies over repeated surveying sessions and (iii) general considerations such as the area coverage, costs, fieldwork time and processing requirements needed.
The main goal of this research is to shed further light on the sensitivity of the vegetation indices to spatial changes of stand parameters. The analysis was done within mountain forests in the Sudetes and the Beskids in southern Poland. Some 1327 stands were analysed with more than 70 percent of spruce contribution in the species composition. The response of selected vegetation indices was verified in relation to the alterations of spruce participation, stand height, volume, stand density and diameter. The following indices were analysed: Normalized Difference Vegetation Index, Normalized Difference Red Edge Index, Green Normalized Difference Vegetation Index and Wide Dynamic Range Vegetation Index. Indices were calculated based on the Rapid Eye (Black Bridge) images. All the analysed stand characteristics influence the values of vegetation indices. In general: mean height, diameter at breast height, volume and spruce participation are the most negatively correlated with the indices. Density is a variable that, in general, cannot directly be used for indices correction, because it is hard to find any stable trend. NDRE is the most stable index for the analysis of stand characteristics.
Exploring the possibility to produce nation-wide forest attribute maps using stereophotogrammetry of aerial images, the national terrain model and data from the National Forest Inventory (NFI). The study areas are four image acquisition blocks in mid- and south Sweden. Regression models were developed and applied to 12.5 m × 12.5 m raster cells for each block and validation was done with an independent dataset of forest stands. Model performance was compared for eight different forest types separately and the accuracies between forest types clearly differs for both image- and LiDAR methods, but between methods the difference in accuracy is small at plot level. At stand level, the root mean square error in percent of the mean (RMSE%) were ranging: from 7.7% to 10.5% for mean height; from 12.0% to 17.8% for mean diameter; from 21.8% to 22.8% for stem volume; and from 17.7% to 21.1% for basal area. This study clearly shows that aerial images from the national image program together with field sample plots from the NFI can be used for large area forest attribute mapping. © 2017, Finnish Society of Forest Science. All rights reserved.
In this study, the potential of using very high resolution Pléiades imagery to estimate a number of common forest attributes for 10-m plots in boreal forest was examined, when a high-resolution terrain model was available. The explanatory variables were derived from three processing alternatives. Height metrics were extracted from image matching of the images acquired from different incidence angles. Spectral derivatives were derived by performing principal component analysis of the spectral bands and lastly, second order textural metrics were extracted from a gray-level co-occurrence matrix, computed with an 11 × 11 pixels moving window. The analysis took place at two Swedish test sites, Krycklan and Remningstorp, containing boreal and hemi-boreal forest. The lowest RMSE was estimated with 1.4 m (7.7%) for Lorey's mean height, 1.7 m (10%) for airborne laser scanning height percentile 90, 5.1 m 2 ·ha −1 (22%) for basal area, 66 m 3 ·ha −1 (27%) for stem volume, and 26 tons·ha −1 (26%) for above-ground biomass, respectively. It was found that the image-matched height metrics were most important in all models, and that the spectral and textural metrics contained similar information. Nevertheless, the best estimations were obtained when all three explanatory sources were used. To conclude, image-matched height metrics should be prioritised over spectral metrics when estimation of forest attributes is concerned.
WorldView-2 (WV2) satellite stereo images were used to derive a digital surface model, which together with a high-resolution digi- tal terrain model from airborne laser scanning (ALS) were used to estimate forest height. Lorey’s mean height (HL) could be esti- mated with a root mean square error of 1.5 m (8.3%) and 1.4 m (10.4%), using linear regression, at the two Swedish test sites Remningstorp (Lat. 58°30ʹN, Long. 13°40ʹE) and Krycklan (Lat. 64° 16ʹN, Long. 19°46ʹE), which contain hemi-boreal and boreal forest. The correlation coefficients were r = 0.94 and r = 0.91, respectively. The 10 m sample plots were 175 in Remningstorp and 282 in Krycklan. It was furthermore found that WV2 data are sometimes unstable for canopy top height estimations (ALS height percentile 100, p100) and that the reconstructed heights are generally located below the actual top height. The WV2 p60 was found to correlate best with ALS p70 in Remningstorp, while WV2 p95 was found to correlate best with ALS p70 in Krycklan, and it moreover reached the highest correlation for all other estimated variables, at both test sites. It was concluded that WV2 p95 height data overall represent approximately the forest height ALS p70. The overall high correlation coefficients above 0.90 at both test sites, with different forest conditions, indicate that stereo matching of WV2 satellite images is suitable for forest height mapping.
Forest inventory and management requirements are changing rapidly in the context of an increasingly complex set of economic, environmental, and social policy objectives. Advanced remote sensing technologies provide data to assist in addressing these escalating information needs and to support the subsequent development and parameterization of models for an even broader range of information needs. This special issue contains papers that use a variety of remote sensing technologies to derive forest inventory or inventory-related information. Herein, we review the potential of 4 advanced remote sensing technologies, which we posit as having the greatest potential to influence forest inventories designed to characterize forest resource information for strategic, tactical, and operational planning: airborne laser scanning (ALS), terrestrial laser scanning (TLS), digital aerial photogrammetry (DAP), and high spatial resolution (HSR)/very high spatial resolution (VHSR) satellite optical imagery. ALS, in particular, has proven to be a transformative technology, offering forest inventories the required spatial detail and accuracy across large areas and a diverse range of forest types. The coupling of DAP with ALS technologies will likely have the greatest impact on forest inventory practices in the next decade, providing capacity for a broader suite of attributes, as well as for monitoring growth over time.
Forest gaps are important structural elements in forest ecology to which various conservation-relevant, photophilic species are associated. To automatically map forest gaps and detect their changes over time, we developed a method based on Digital Surface Models (DSM) derived from stereoscopic aerial imagery and a LiDAR-based Digital Elevation Model (LiDAR DEM). Gaps were detected and delineated in relation to height and cover of the surrounding forest comparing data from two public flight campaigns (2009 and 2012) in a 1023-ha model region in the Northern Black Forest, Southwest Germany. The method was evaluated using an independent validation dataset obtained by visual stereo-interpretation. Gaps were automatically detected with an overall accuracy of 0.90 (2009) and 0.82 (2012). However, a very high producers' accuracy of more than 0.95 (both years) was counterbalanced by a user's accuracy of 0.84 (2009) and 0.73 (2012) as some gaps were not automatically detected. Accuracy was mainly dependent on the shadow occurrence and height of the surrounding forest with user's accuracies dropping to 0.70 (2009) and 0.52 (2012) in high stands (> 8 m tree height). As one important step in the workflow, the class of open forest, an important feature for many forest species, was delineated with a very good overall accuracy of 0.92 (both years) with uncertainties occurring mostly in areas with intermediate canopy cover. Presence of complete or partial shadow and geometric limitations of stereo image matching were identified as the main sources of errors in the method performance, suggesting that images with a higher overlap and resolution and ameliorated image-matching algorithms provide the greatest potential for improvement.
It is anticipated that many of the future forest mapping applications will be based on three-dimensional (3D) point clouds. A comparison study was conducted to verify the explanatory power and information contents of several 3D remote sensing data sources on the retrieval of above ground biomass (AGB), stem volume (VOL), basal area (G), basal-area weighted mean diameter (D-g) and Lorey's mean height (H-g) at the plot level, utilizing the following data: synthetic aperture radar (SAR) Interferometry, SAR radargrammetry, satellite-imagery having stereo viewing capability, airborne laser scanning (ALS) with various densities (0.8-6 pulses/m(2)) and aerial stereo imagery. Laser scanning is generally known as the primary source providing a 3D point cloud. However, photogrammetric, radargrammetric and interferometric techniques can be used to produce 3D point clouds from space- and air-borne stereo images. Such an image-based point cloud could be utilized in a similar manner as ALS providing that accurate digital terrain model is available. In this study, the performance of these data sources for providing point cloud data was evaluated with 91 sample plots that were established in Evo, southern Finland within a boreal forest zone and surveyed in 2014 for this comparison. The prediction models were built using random forests technique with features derived from each data sources as independent variables and field measurements of forest attributes as response variable. The relative root mean square errors (RMSEs) varied in the ranges of 4.6% (0.97 m)-13.4% (2.83 m) for H-g, 11.7% (3.0 cm)-20.6% (5.3 cm) for D-g, 14.8% (4.0 m(2)/ha)-25.8% (6.9 m(2)/ha) for G, 15.9% (43.0 m(3)/ha)-31.2% (84.2 m(3)/ha) for VOL and 14.3% (19.2 Mg/ha)-27.5% (37.0 Mg/ha) for AGB, respectively, depending on the data used. Results indicate that ALS data achieved the most accurate estimates for all forest inventory attributes. For image-based 3D data, high-altitude aerial images and WorldView-2 satellite optical image gave similar results for H-g and D-g, which were only slightly worse than those of ALS data. As expected, spaceborne SAR data produced the worst estimates. WorldView-2 satellite data performed well, achieving accuracy comparable to the one with ALS data for G, VOL and AGB estimation. SAR interferometry data seems to contain more information for forest inventory than SAR radargrammetry and reach a better accuracy (relative RMSE decreased from 13.4% to 9.5% for H-g, 20.6% to 19.2% for D-g, 25.8% to 20.9% for G, 31.2% to 22.0% for VOL and 27.5% to 20.7% for AGB, respectively). However, the availability of interferometry data is limited. The results confirmed the high potential of all 3D remote sensing data sources for forest inventory purposes. However, the assumption of using other than ALS data is that there exist a high quality digital terrain model, in our case it was derived from ALS.
Photogrammetric point clouds (PPC) obtained by stereomatching of aerial photographs now have a resolution sufficient to discern individual trees. We have produced such PPCs of a boreal forest and delineated individual tree crowns using a segmentation algorithm applied to the canopy height model derived from the PPC and a lidar terrain model. The crowns were characterized in terms of height and species (spruce, fir, and deciduous). Species classification used the 3D shape of the single crowns and their reflectance properties. The same was performed on a lidar dataset. Results show that the quality of PPC data generally approaches that of airborne lidar. For pixel-based canopy height models, viewing geometry in aerial images, forest structure (dense vs. open canopies), and composition (deciduous vs. conifers) influenced the quality of the 3D reconstruction of PPCs relative to lidar. Nevertheless, when individual tree height distributions were analyzed, PPC-based results were very similar to those extracted from lidar. The random forest classification (RF) of individual trees performed better in the lidar case when only 3D metrics were used (83% accuracy for lidar, 79% for PPC). However, when 3D and intensity or multispectral data were used together, the accuracy of PPCs (89%) surpassed that of lidar (86%).
Digital aerial photogrammetry (DAP) is emerging as an alternate data source to airborne laser scanning (ALS) data for three-dimensional characterization of forest structure. In this study we compare point cloud metrics and plot-level model estimates derived from ALS data and an image-based point cloud generated using semi-global matching (SGM) for a complex, coastal forest in western Canada. Plot-level estimates of Lorey's mean height (H), basal area (G), and gross volume (V) were modelled using an area-based approach. Metrics and model outcomes were evaluated across a series of strata defined by slope and canopy cover, as well as by image acquisition date. We found statistically significant differences between ALS and SGM metrics for all strata for five of the eight metrics we used for model development. We also found that the similarity between metrics from the two data sources generally increased with increasing canopy cover, particularly for upper canopy metrics, whereas trends across slope classes were less consistent. Model outcomes from ALS and SGM were comparable. We found the greatest OPEN ACCESS Forests 2015, 6 3705 difference in model outcomes was for H (ΔRMSE% = 5.04%). By comparison, ΔRMSE% was 2.33% for G and 3.63% for V. We did not discern any corresponding trends in model outcomes across slope and canopy cover strata, or associated with different image acquisition dates.
Satellite imagery has proven extremely useful for repetitive timeline-based data collection, because it offers a synoptic view and enables fast processing of large quantities of data. The changes in tree crown number and land cover in a very remote watershed (area 1305 ha) in Nepal were analyzed using a QuickBird image from 2006 and an IKONOS image from 2011. A geographic object-based image analysis (GEOBIA) was carried out using the region-growing technique for tree crown detection, delineation, and change assessment, and a multiresolution technique was used for land cover mapping and change analysis. The coefficient of determination for tree crown detection and delineation was 0.97 for QuickBird and 0.99 for IKONOS, calculated using a line-intercept transect method with 10 randomly selected windows (1×1 ha). The number of tree crowns decreased from 47,121 in 2006 to 41,689 in 2011, a loss of approximately 90 trees per month on average; the area of needle-leaved forest was reduced by 140 ha (23%) over the same period. Analysis of widely available very-high-resolution satellite images using GEOBIA techniques offers a cost-effective method for detecting changes in tree crown number and land cover in remote mountain valleys; the results provide the information needed to support improved local-level planning and forest management in such areas.
Abstract
Assessing forest cover change is a key issue for any national forest inventory. This was tested in two study areas in Switzerland on the basis of stereo airborne digital sensor (ADS) images and advanced digital surface model (DSM) generation techniques based on image point clouds. In the present study, an adaptive multi-scale approach to detect forest cover change with high spatial and temporal resolution was applied to two study areas in Switzerland. The challenge of this approach is to minimize DSM height uncertainties that may affect the accuracy of the forest cover change results. The approach consisted of two steps. In the first step, a ‘change index’ parameter indicated the overall change status at a coarser scale. The tendency towards change was indicated by derivative analysis of the normalized histograms of the difference between the two canopy height models (DCHMs) in different years. In the second step, detection of forest cover change at a refined scale was based on an automatic threshold and a moving window technique. Promising results were obtained and reveal that real forest cover changes can be distinguished from non-changes with a high degree of accuracy in managed mixed forests. Results had a lower accuracy for forests located on steep alpine terrain. A major benefit of the proposed method is that the magnitude of forest cover change of any specific region can be made available within a short time as often required by forest managers or policy-makers, especially after unexpected natural disturbances.
© Institute of Chartered Foresters, 2015. All rights reserved. For Permissions, please e-mail: journals.permissions@oup.com.
The recent development of operational small unmanned aerial systems (UASs) opens the door for their extensive use in forest mapping, as both the spatial and temporal resolution of UAS imagery better suit local-scale investigation than traditional remote sensing tools. This article focuses on the use of combined photogrammetry and "Structure from Motion" approaches in order to model the forest canopy surface from low-altitude aerial images. An original workflow, using the open source and free photogrammetric toolbox, MICMAC (acronym for Multi Image Matches for Auto Correlation Methods), was set up to create a digital canopy surface model of deciduous stands. In combination with a co-registered light detection and ranging (LiDAR) digital terrain model, the elevation of vegetation was determined, and the resulting hybrid photo/LiDAR canopy height model was compared to data from a LiDAR canopy height model and from forest inventory data. Linear regressions predicting dominant height and individual height from plot metrics and crown metrics showed that the photogrammetric canopy height model was of good quality for deciduous stands. Although photogrammetric reconstruction significantly smooths the canopy surface, the use of this workflow has the potential to take full advantage of the flexible revisit period of drones in order to refresh the LiDAR canopy height model and to collect dense multitemporal canopy height series.
Surface models provide key knowledge of the 3-d structure of forests. Aerial stereo imagery acquired during routine mapping campaigns covering the whole of Switzerland (41,285 km2), offers a potential data source to calculate digital surface models (DSMs). We present an automated workflow to generate a nationwide DSM with a resolution of 1 × 1 m based on photogrammetric image matching. A canopy height model (CHM) is derived in combination with an existing digital terrain model (DTM). ADS40/ADS80 summer images from 2007 to 2012 were used for stereo matching, with ground sample distances (GSD) of 0.25 m in lowlands and 0.5 m in high mountain areas. Two different image matching strategies for DSM calculation were applied: one optimized for single features such as trees and for abrupt changes in elevation such as steep rocks, and another optimized for homogeneous areas such as meadows or glaciers. The country was divided into 165,500 blocks, which were matched independently using an automated workflow. The completeness of successfully matched points was high, 97.9%. To test the accuracy of the derived DSM, two reference data sets were used: (1) topographic survey points (n = 198) and (2) stereo measurements (n = 195,784) within the framework of the Swiss National Forest Inventory (NFI), in order to distinguish various land cover types. An overall median accuracy of 0.04 m with a normalized median absolute deviation (NMAD) of 0.32 m was found using the topographic survey points. The agreement between the stereo measurements and the values of the DSM revealed acceptable NMAD values between 1.76 and 3.94 m for forested areas. A good correlation (Pearson's r = 0.83) was found between terrestrially measured tree height (n = 3109) and the height derived from the CHM. Optimized image matching strategies, an automatic workflow and acceptable computation time mean that the presented approach is suitable for operational usage at the nationwide extent. The CHM will be used to reduce estimation errors of different forest characteristics in the Swiss NFI and has high potential for change detection assessments, since an aerial stereo imagery update is available every six years.
Information on snow depth and its spatial distribution is crucial for
numerous applications in snow and avalanche research as well as in hydrology
and ecology. Today, snow depth distributions are usually estimated using
point measurements performed by automated weather stations and observers in
the field combined with interpolation algorithms. However, these
methodologies are not able to capture the high spatial variability of the
snow depth distribution present in alpine terrain. Continuous and accurate
snow depth mapping has been successfully performed using laser scanning but
this method can only cover limited areas and is expensive. We use the
airborne ADS80 optoelectronic scanner, acquiring stereo imagery with 0.25 m
spatial resolution to derive digital surface models (DSMs) of winter and
summer terrains in the neighborhood of Davos, Switzerland. The DSMs are
generated using photogrammetric image correlation techniques based on the
multispectral nadir and backward-looking sensor data. In order to assess the accuracy of the
photogrammetric products, we compare these
products with the following independent data sets acquired simultaneously: (a)
manually measured snow depth plots; (b) differential Global Navigation
Satellite System (dGNSS) points; (c) terrestrial laser scanning (TLS); and (d)
ground-penetrating radar (GPR) data sets. We demonstrate that the method presented can be
used to map snow depth at 2 m resolution with a vertical depth
accuracy of ±30 cm (root mean square error) in the complex topography
of the Alps. The snow depth maps presented have an average accuracy that is
better than 15 % compared to the average snow depth of 2.2 m over the
entire test site.
In this paper, we demonstrate the effectiveness of digital stereo images and canopy height models (CHMs) derived from them for forest height change assessment. Top heights were derived for 199 terrestrial inventory plots from forest inventories conducted in 2008 and 2013 in a forest near Traunstein, Germany. Semi-Global Matching was applied to two sets of aerial stereo images, acquired in 2009 and 2012, respectively, to compute CHMs. Subsequently, several height percentiles were calculated from the areas in the CHMs that lay within the inventory plot locations. The maximum CHM value (hmax) had the highest correlation with the field-based canopy top heights and was selected for use in all further analysis. Periodic annual increments (PAIs) of forest height were calculated from both the remote sensing and the field data at the inventory plot locations. Scatterplots of the PAIs over top height revealed similar patterns in the results derived from the two data sets. The inventory plots were assigned to three height classes representing various forest successional stages - youth, full vigour and old age. The PAI distributions within the three height classes were significantly different from one another. Our findings suggest that CHMs derived from repeat aerial image surveys can be a viable tool to measure canopy heights and to assess forest height changes over time, even for a highly structured, mixed forest in central Europe.
Few studies have evaluated the precision of IKONOS stereo data for measuring forest canopy height. The high cost of airborne LiDAR data collection for large area studies and the present lack of a spaceborne instrument necessitate the need to explore other low cost options. The U.S. Government currently has access to a large archive of commercial high-resolution imagery, which could be quite valuable to forest structure studies. At 1 m resolution, we compared canopy height models (CHMs) and height data derived from Goddard’s airborne LiDAR Hyper-spectral and Thermal Imager (G-LiHT) with three types of IKONOS stereo derived digital surface models (DSMs) that estimate CHMs by subtracting National Elevation Data (NED) digital terrain models (DTMs). We found the following in three different forested regions of the U.S. after excluding heterogeneous and disturbed forest samples: (1) G-LiHT DTMs were highly correlated to NED DTMs with R2 > 0.98, and root mean square errors (RMSEs) < 2.96 m; (2) when using one visually identifiable ground control point (GCP) from NED, G-LiHT DSMs and IKONOS DSMs had R2 > 0.84 and RMSEs of 2.7 to 4.1 m; and (3) one GCP CHMs for two-study sites had R2 > 0.7 and RMSEs of 2.6 to 3 m where data were collected less than four years apart. Our results suggest IKONOS stereo data are a useful LiDAR alternative where high quality DTMs are available.
Airborne laser scanning (ALS) has demonstrated utility for forestry applications and has renewed interest in
other forms of remotely sensed data, especially those that capture three-dimensional (3-D) forest characteristics. One such data source results from the advanced processing of high spatial resolution digital stereo imagery (DSI) to generate 3-D point clouds. From the derived point cloud, a digital surface model and forest vertical information with similarities to ALS can be generated. A key consideration is that when developing forestry related products such as a canopy height model (CHM), a high spatial resolution digital terrain model (DTM), typically from ALS, is required to normalize DSI elevations to heights above ground. In this paper we report on our investigations into the use of DSI-derived vertical information for capturing variations in forest structure and compare these results to those acquired using ALS. An ALS-derived DTM was used to provide the spatially detailed ground surface elevations to normalize DSI-derived heights. Similar metrics were calculated from the vertical information provided by both DSI and ALS. Comparisons revealed that ALS metrics provided a more detailed characterization of the canopy surface including canopy openings. Both DSI and ALS metrics had similar levels of correlation with forest structural attributes (e.g., height, volume, and biomass). DSI-based models predicted height, diameter, basal area, stem volume, and biomass with root mean square (RMS) accuracies of 11.2%, 21.7%, 23.6%, 24.5%, and 23.7%, respectively. The respective accuracies for the ALS-based predictions were 7.8%, 19.1%, 17.8%, 17.9%, and 17.5%. Change detection between ALS-derived CHM (time 1) and DSI-derived CHM (time 2) provided change estimates that demonstrated good agreement (r _0.71) with two-date, ALS only, change outputs. For the single-layered, even-aged stands under investigation in this study, the DSI-derived vertical information is an appropriate and cost-effective data source for estimating and updating forest information. The accuracy of DSI information is based on a capability to measure the height of the upper canopy envelope with performance analogous to ALS. Forest attributes that are well captured and subsequently modeled from height metrics are best suited to estimation from DSI metrics, whereas ALS is more suitable for capturing stand density. Further investigation is required to better understand the performance of DSI-derived height products in more complex forest environments. Furthermore, the difference in variance captured between ALS and DSI-derived CHM also needs to be better understood in the context of change detection and inventory update considerations.
Current technical advances in the field of digital photogrammetry demonstrate the great potential of automatic image matching for deriving dense surface measurements of the forest canopy. In contrast to airborne laser scanning (ALS), aerial stereo images are updated more regularly by national or regional mapping agencies in several countries. Frequently, ALS-based terrain models (DTMs) are available, and thus photogrammetric canopy heights can be derived. However, currently, there is little knowledge as to how accurately forest attributes can be modeled using the aerial stereo images acquired by these official, regular aerial surveys, especially for mixed forests in central Europe. Thus, a photogrammetric point cloud derived from UltraCamX stereo images in combination with an ALS-DTM and a classification of coniferous and deciduous tree regions (based on orthoimages) was used to create a stratified estimation of timber volume and basal area in a mixed forest in Germany. Suitable models were derived at the plot level using explanatory variables from the photogrammetric point cloud (which was normalized using an ALS-DTM). The prior stratification of conifer- and deciduous-dominated field plots slightly improved the estimation accuracy. The results verify that stereo images can be an alternative to ALS data for modeling key forest attributes, even in mixed central European forests with complex structure.
Optical images of different spectral and spatial resolutions continue to provide a reliable source of data for estimating forest inventory parameters. WorldView-2 launched in October 2009 is the first commercial optical satellite to provide high spatial resolution images with eight spectral bands, some of which are new and require investigation for estimation of forest structure parameters. In this study, a WorldView-2 multispectral image has been investigated for mapping pine plantation structural parameters including stand volume, basal area, stocking, mean diameter at breast height mean DBH, and mean height of trees over a Pinus radiata plantation in New South Wales, Australia. Spectral derivatives including reflectance bands, band ratios, principal components PCs, and several vegetation indices VIs were calculated using four typical bands, including blue, green, red, and near-infrared NIR1, and all eight bands. Moreover, textural information, including 11 grey-level co-occurrence matrix GLCM indices, was extracted using four window sizes and orientations. Several models were developed using the extracted attributes separately to compare the efficiency of the models derived from the attributes of four typical bands and eight bands, as well as to compare between the capability of spectral-based and textural-based models for estimating structural parameters. The results showed that models derived from textural attributes of eight spectral bands provide the best estimates compared to those derived from four typical bands and the models derived from spectral derivatives. Moreover, the mean height and mean DBH with 8% and 13.7% error of estimation, respectively, were estimated more accurately than basal area, stand volume, and stocking, where the error of estimation is up to 30%.
A best practice guide brings together state-of-the-art approaches, methods, and data to provide non-experts more detailed information about complex topics. With this guide, our goal is to inform and enable readers interested in using airborne laser scanning (ALS; also referred to as Light Detection and Ranging [LiDAR]) data to characterize, in an operational inventory context, large forest areas in a cost-effective manner. To meet this goal, we outline an approach to using ALS data that is based on (1) theoretical and technical applicability; (2) published or established heritage; (3) parsimoniousness; and (4) clarity. The best practices presented herein are based on more than 25 years of scientific research on the application of ALS data in forest inventory.
We describe the process required to generate forest inventory attributes from ALS data from start to finish, recommending best practices for each stage, from ground sampling and statistics, through to sophisticated spatial data processing and analysis. As the collection of ground plot data for model calibration and validation is a critical component of the recommended approach, we have placed appropriate emphasis on this section of the guide. Although many readers will not have the capacity—or need—to undertake all of the stages of this process themselves, we feel it is important for all readers to have some understanding of the various stages of the process. Such an understanding is necessary to make informed decisions when determining whether ALS is an appropriate data choice for a forest management area. Moreover, a minimum level of knowledge is useful when outsourcing or establishing collaborations for data acquisition, processing, or analysis, and when evaluating deliverables. To this end, we also provide some background information on ALS.
National forest inventories (NFIs) have a long history, although their current major features date only to the early years of the twentieth century. Recent issues such as concern over the effects of acid deposition, biodiversity, forest sustainability, increased demand for forest data, international reporting requirements and climate change have led to the expansion of NFIs to include more variables, greater diversity in sampling protocols and a generally more holistic approach. This review focuses on six selected topics: (1) a brief historical review; (2) a summary of common structural features of NFIs; (3) a brief review of international reporting requirements using NFI data with an emphasis on approaches to harmonized estimation; (4) an overview of inventory estimation methods that can be enhanced with remotely sensed data; (5) an overview of nearest neighbors prediction and estimation techniques; and (6) a brief overview of several emerging issues including carbon inventories in developing countries and use of lidar data. Although general inventory principles will remain unchanged, sampling designs, plot configurations and measurement protocols will require modification before they can be applied in countries with tropical forests. Technological advances, particularly in the use of remotely sensed data, including lidar data, have led to greater inventory efficiencies, better maps and accurate estimation for small areas.
Airborne laser scanning (ALS) requires GNSS (Global Navigation Satellite System; e.g. GPS) and an IMU (Inertial Measurement Unit) for determining the dynamically changing orientation of the scanning system. Because of small but existing instabilities of the involved parts - especially the mounting calibration - a strip adjustment is necessary in most cases. In order to realize this adjustment in a rigorous way the GNSS/IMU-trajectory data is required. In some projects this data is not available to the user (any more). Derived from the rigorous model, this article presents a model for strip adjustment without GNSS/IMU-trajectory data using five parameters per strip: one 3D shift, one roll angle, and one affine yaw parameter. In an example with real data consiting of 61 strips this model was successfully applied leading to an obvious improvement of the relative accuracy from (59.3/23.4/4.5) (cm) to (7.1/7.2/2.2) (defined as RMS values in (X/Y/Z) of the differences of corresponding points derived by least squares matching in the overlapping strips). This example also clearly demonstrates the importance of the affine yaw parameter.
Estimates of canopy height (H) and fractional canopy cover (FC) derived from lidar data collected during leaf-on and leaf-off conditions are compared with field measurements from 80 forested riparian buffer plots. The purpose is to determine if existing lidar data flown in leaf-off conditions for applications such as terrain mapping can effectively estimate forested riparian buffer H and FC within a range of riparian vegetation types. Results illustrate that: 1) leaf-off and leaf-on lidar percentile estimates are similar to measured heights in all plots except those dominated by deciduous compound-leaved trees where lidar underestimates H during leaf off periods; 2) canopy height models (CHMs) underestimate H by a larger margin compared to percentile methods and are influenced by vegetation type (conifer needle, deciduous simple leaf or deciduous compound leaf) and canopy height variability, 3) lidar estimates of FC are within 10% of plot measurements during leaf-on periods, but are underestimated during leaf-off periods except in mixed and conifer plots; and 4) depth of laser pulse penetration lower in the canopy is more variable compared to top of the canopy penetration which may influence within canopy vegetation structure estimates. This study demonstrates that leaf-off lidar data can be used to estimate forested riparian buffer canopy height within diverse vegetation conditions and fractional canopy cover within mixed and conifer forests when leaf-on lidar data are not available.
We present two improvements for laser-based forest inventory. The first improvement is based on using last pulse data for tree detection. When trees overlap, the surface model between the trees corresponding to the first pulse stays high, whereas the corresponding model from the last pulse results in a drop in elevation, due to its better penetration between the trees. This drop in elevation can be used for separating trees. In a test carried out in Evo, Southern Finland, we used 292 forests plots consisting of more than 5,500 trees and airborne laser scanning (ALS) data comprised of 12.7 emitted laser pulses per m(2). With last pulse data, an improvement of 6% for individual tree detection was obtained when compared to using first pulse data. The improvement increased with an increasing number of stems per plot and with decreasing diameter breast height (DBH). The results confirm that there is also substantial information for tree detection in last pulse data. The second improvement is based on the use of individual tree-based features in addition to the statistical point height metrics in area-based prediction of forest variables. The commonly-used ALS point height metrics and individual tree-based features were fused into the non-parametric estimation of forest variables. By using only four individual tree-based features, stem volume estimation improved when compared to the use of statistical point height metrics. For DBH estimation, the point height metrics and individual tree-based features complemented each other. Predictions were validated at plot level.
This study presents an approach for semi-automated classification of tree species in different types of forests using first and second generation ADS40 and RC30 images from two study areas located in the Swiss Alps. In a first step, high-resolution canopy height models (CHMs) were generated from the ADS40 stereo-images. In a second step, multi-resolution image segmentation was applied. Based on image segments seven different tree species for study area 1 and four for study area 2 were classified by multinomial regression models using the geometric and spectral variables derived from the ADS40 and RC30 images. To deal with the large number of explanatory variables and to find redundant variables, model diagnostics and step-wise variable selection were evaluated. Classifications were ten-fold cross-validated for 517 trees that had been visited in field surveys and detected in the ADS40 images. The overall accuracies vary between 0.76 and 0.83 and Cohen's kappa values were between 0.70 and 0.73. Lower accuracies (kappa < 0.5) were obtained for small samples of species such as non-dominant tree species or less vital trees with similar spectral properties. The usage of NIR bands as explanatory variables from RC30 or from the second generation of ADS40 was found to substantially improve the classification results of the dominant tree species. The present study shows the potential and limits of classifying the most frequent tree species in different types of forests, and discusses possible applications in the Swiss National Forest Inventory.
Obtaining reliable measures of tree canopy height across large areas is a central element of forest inventory and carbon accounting. Recent years have seen an increased emphasis on the use of active sensors like Radar and airborne LiDAR (light detection and scanning) systems to estimate various 3D characteristics of canopy and crown structure that can be used as predictors of biomass. However, airborne LiDAR data are expensive to acquire, and not often readily available across large remote landscapes. In this study, we evaluated the potential of stereo imagery from commercially available Very High Resolution (VHR) satellites as an alternative for estimating canopy height variables in Australian tropical savannas, using a semi-global dense matching (SGM) image-based technique. We assessed and compared the completeness and vertical accuracy of extracted canopy height models (CHMs) from GeoEye 1 and WorldView 1 VHR satellite stereo pairs and summarised the factors influencing image matching effectiveness and quality.
Our results showed that stereo dense matching using the SGM technique severely underestimates tree presence and canopy height. The highest tree detection rates were achieved by using the near-infrared (NIR) band of GE1 (8–9%). WV1-GE1 cross-satellite (mixed) models did not improve the quality of extracted canopy heights. We consider these poor detection rates and height retrievals to result from: i) the clumping crown structure of the dominant Eucalyptus spp.; ii) their vertically oriented leaves (affecting the bidirectional reflectance distribution function); iii) image band radiometry and iv) wind induced crown movement affecting stereo-pair point matching. Our detailed analyses suggest that current commercially available VHR satellite data (0.5 m resolution) are not well suited to estimating canopy height variables, and therefore above ground biomass (AGB), in Eucalyptus dominated north Australian tropical savanna woodlands.
In Europe, assessment of the ecological function of forests is increasingly important as evidenced by Natura 2000, which has seen the establishment of protected sites that are of ecological importance. The presence of canopy gaps has proved to be a useful indicator of the forest structure and biodiversity in these areas.
Two methods – Object Based Image Analysis (OBIA) and Image Texture Based Analysis (ITBA) – were assessed to detect and quantify canopy gaps and to investigate the influence of forest type on their performance. The research was undertaken in two Natura 2000 sites in Bulgaria. To estimate the gap fractions in the field, 100 square plots (150m by 150 m) were established. Within each plot, 25 circular sub-plots were laid out in a regular pattern. Gap size within each sub-plot was estimated visually.
The methods were assessed using very high-resolution (0.13m ground resolution) aerial imagery. For the OBIA method, five parameter settings were tested and the gap fractions estimated on the imagery and compared with the field estimates. Likewise, for the ITBA method, five different settings were tested. In general, both methods overestimated canopy gap fraction relative to the field-based estimates, especially where the gap fractions were relatively small. Larger gaps were more accurately estimated by the OBIA method. Neither method performed well in broadleaved forest. Statistical comparison showed that, overall, the correlations of the estimates from the imagery with field data are only moderate and that there is no significant difference between the two methods.
Point cloud data derived from stereo satellite imagery has the potential to provide large-scale forest inventory assessment but these methods are known to include higher error than airborne laser scanning (ALS). This study compares the accuracy of forest inventory attributes estimated from high-density ALS (21.1 pulses m −2) point cloud data (PCD) and PCD derived from photogrammetric methods applied to stereo satellite imagery obtained over a Pinus radiata D. Don plantation forest in New Zealand. The statistical and textural properties of the canopy height models (CHMs) derived from each point cloud were included alongside standard PCD metrics as a means of improving the accuracy of predictions for key forest inventory attributes. For mean top height (a measure of dominant height in a stand), ALS data produced better estimates (R 2 = 0.88; RMSE = 1.7 m) than those obtained from satellite data (R 2 = 0.81; RMSE = 2.1 m). This was attributable to a general over-estimation of canopy heights in the satellite PCD. ALS models produced poor estimates of stand density (R 2 = 0.48; RMSE = 112.1 stems ha −1), as did the satellite PCD models (R 2 = 0.42; RMSE = 118.4 stems ha −1). ALS models produced accurate estimates of basal area (R 2 = 0.58; RMSE = 12 m 2 ha −1), total stem volume (R 2 = 0.72; RMSE = 107.5 m 3 ha −1), and total recoverable volume (R 2 = 0.74; RMSE = 92.9 m 3 ha −1). These values differed little from the estimates of basal area (R 2 = 0.57; RMSE = 12.2 m 2 ha −1), total stem volume (R 2 = 0.70; RMSE = 112.6 m 3 ha −1), and total recoverable volume (R 2 = 0.73; RMSE = 96 m 3 ha −1) obtained from satellite PCD models. The statistical and textural metrics computed from the CHMs were important variables in all of the models derived from both satellite and ALS PCD, nearly always outranking the standard PCD metrics in measures of importance. For the satellite PCD models, the CHM-derived metrics were nearly exclusively identified as important variables. These results clearly show that point cloud data obtained from stereo satellite imagery are useful for prediction of forest inventory attributes in intensively managed forests on steeper terrain. Furthermore, these data offer forest managers the benefit of obtaining both inventory data and high-resolution multispectral imagery from a single product.
This study had the aim of investigating the utility of image-based point cloud data for estimation of vertical canopy cover (VCC). An accurate measure of VCC based on photogrammetric matching of aerial images would aid in vegetation mapping, especially in areas where aerial imagery is acquired regularly. The test area is located in southern Sweden and was divided into four vegetation types with sparse to dense tree cover: unmanaged coniferous forest; pasture areas with deciduous tree cover; wetland; and managed coniferous forest. Aerial imagery with a ground sample distance of 0.24 m was photogrammetrically matched to produce dense image-based point cloud data. Two different image matching software solutions were used and compared: MATCH-T DSM by Trimble and SURE by nFrames. The image-based point clouds were normalized using a digital terrain model derived from airborne laser scanner (ALS) data. The canopy cover metric vegetation ratio was derived from the image-based point clouds, as well as from raster-based canopy height models (CHMs) derived from the point clouds. Regression analysis was applied with vegetation ratio derived from near nadir ALS data as the dependent variable and metrics derived from image-based point cloud data as the independent variables. Among the different vegetation types, vegetation ratio derived from the image-based point cloud data generated by using MATCH-T resulted in relative root mean square errors (rRMSE) of VCC ranging from 6.1% to 29.3%. Vegetation ratio based on point clouds from SURE resulted in rRMSEs ranging from 7.3% to 37.9%. Use of the vegetation ratio based on CHMs generated from the image-based point clouds resulted in similar, yet slightly higher values of rRMSE.
Spatially explicit information on tree species composition of managed and natural forests, plantations and urban vegetation provides valuable information for nature conservationists as well as for forest and urban managers and is frequently required over large spatial extents. Over the last four decades, advances in remote sensing technology have enabled the classification of tree species from several sensor types. While studies using remote sensing data to classify and map tree species reach back several decades, a recent review on the status, potentials, challenges and outlooks in this realm is missing. Here, we search for major trends in remote sensing techniques for tree species classification and discuss the effectiveness of different sensors and algorithms based on a literature review. This review demonstrates that the number of studies focusing on tree species classification has increased constantly over the last four decades and promising local scale approaches have been presented for several sensor types. However, there are few examples for tree species classifications over large geographic extents, and bridging the gap between current approaches and tree species inventories over large geographic extents is still one of the biggest challenges of this research field. Furthermore, we found only few studies which systematically described and examined the traits that drive the observed variance in the remote sensing signal and thereby enable or hamper species classifications. Most studies followed data-driven approaches and pursued an optimization of classification accuracy, while a concrete hypothesis or a targeted application was missing in all but a few exceptional studies. We recommend that future research efforts focus stronger on the causal understanding of why tree species classification approaches work under certain conditions or – maybe even more important-why they do not work in other cases. This might require more complex field acquisitions than those typically used in the reviewed studies. At the same time, we recommend reducing the number of purely data-driven studies and algorithm-benchmarking studies as these studies are of limited value, especially if the experimental design is limited, e.g. the tree population is not representative and only a few sensors or acquisition settings are simultaneously investigated.
In this article the performance of dense image matching (DIM) is investigated regarding its capability to yield terrain data, especially close to free-flowing water ways. Therefore, over two study areas aerial images with ground sampling distances 10 cm and 6 cm, respectively, are used for matching with the software packages Match-T (Trimble) and SURE (nframes). The matching results over areas with varying vegetation density (open grassland, loose and dense vegetation) are then compared with ALS reference data. Two parameters are investigated: (a) the terrain coverage; i.e. the percentage of the terrain covered by the matching results; and (b) the height accuracy of the matching results in the terrain class. The results show that DIM can only deliver terrain data in areas with no or very loose vegetation. Additionally, it was found that in the case of open grassland the DIM terrain heights were systematically higher by 10cm compared with the ALS terrain heights. This is caused by the fact that ALS can penetrate the vegetation to some extent whereas matching occurs on top of the grass. The very good height accuracy (as standard deviation) obtainable by DIM, which is only slightly worse than the ALS accuracy (6.5 cm vs. 4.5 cm), is encouraging. Motivated by these results new possible applications arise for the respective authorities (the German Federal Institute of Hydrology and the German Federal Wa ter and Shipping Administration), e.g. capturing dry fallen areas of free flowing rivers during documentation at low water levels. © 2016 E. Schweizerbart'sche Verlagsbuchhandlung, Stuttgart, Germany.
In this study we assessed the potential of using photogrammetric data for species-specific forest inventories. The method is based on a combination of Dirichlet and ordinary linear regression models. This approach was used to predict species proportions, main tree species, total, and species-specific volume. Structural and spectral variables were used as predictors. The models were validated using 63 independent validation stands. The results from airborne laser scanning (ALS) data combined with spectral data and photogrammetric data obtained using aerial imagery with different forward overlaps of 80% and 60% were compared. The best photogrammetry based models predicted species proportions with a relative root mean square error (RMSE) of 21.4%, classified dominant species with 79% accuracy, predicted total volume with relative RMSE of 13.4%, and predicted species-specific volume with relative RMSE of 36.6%, 46.5%, and 84.9% for spruce, pine, and deciduous species, respectively. The results were similar for the three point cloud datasets obtained from aerial imagery and ALS and the accuracies of the predictions were comparable to methods used in operational FMI. The study highlights the effectiveness of forest inventories carried out using photogrammetric data, which – differently from ALS, can include species-specific information without relying on multiple data sources.
Forest inventory and monitoring programs are needed to provide timely, spatially complete (i.e. mapped), and verifiable information to support forest management, policy formulation, and reporting obligations. Satellite images, in particular data from the Landsat Thematic Mapper and Enhanced Thematic Mapper (TM/ETM+) sensors, are often integrated with field plots from forest inventory programs, leveraging the complete spatial coverage of imagery with detailed ecological information from a sample of plots to spatially model forest conditions and resources. However, in remote and unmanaged areas such as Canada's northern forests, financial and logistic constraints can severely limit the availability of inventory plot data. Additionally, Landsat spectral information has known limitations for characterizing vertical vegetation structure and biomass; while clouds, snow, and short growing seasons can limit development of large area image mosaics that are spectrally and phenologically consistent across space and time. In this study we predict and map forest structure and aboveground biomass over 37 million ha of forestland in Saskatchewan, Canada. We utilize lidar plots—observations of forest structure collected from airborne discrete-return lidar transects acquired in 2010—as a surrogate for traditional field and photo plots.Mapped explanatory data included Tasseled Cap indices and multi-temporal change metrics derived from Landsat TM/ETM+ pixel-based image composites. Maps of forest structure and total aboveground biomass were created using a Random Forest (RF) implementation of Nearest Neighbor (NN) imputation. The imputation model had moderate to high plot-level accuracy across all forest attributes (R2 values of 0.42–0.69), as well as reasonable attribute predictions and error estimates (for example, canopy cover above 2 m on validation plots averaged 35.77%, with an RMSE of 13.45%, while unsystematic and systematic agreement coefficients (ACuns and ACsys) had values of 0.63 and 0.97 respectively). Additionally, forest attributes displayed consistent trends in relation to the time since and magnitude of wildfires, indicating model predictions captured the dominant ecological patterns and processes in these forests. Acknowledging methodological and conceptual challenges based upon the use of lidar plots from transects, this study demonstrates that using lidar plots and pixel compositing in imputation mapping can provide forest inventory and monitoring information for regions lacking ongoing or up-to-date field data collection programs.
Angle-count sampling (ACS) is an established method in forest mensuration and is implemented in different National Forest Inventories (NFI). However, due to the lack of fixed reference areas of the inventory plots, these ACS-based field data are seldom used as training data for wall-to-wall mapping applications at forest enterprise level. In this paper, we demonstrate an approach to overcome this shortcoming. For a study area in northern Bavaria, Germany, we used ACS-based NFI data for model training to generate wall-to-wall maps of growing stock for broadleaf, conifer and mixed forest stands. Both spectral and height information from the very high resolution WorldView-2 (WV2) satellite were used as auxiliary information and the non-parametric Random Forests (RF) algorithm was chosen as modeling approach. The growing stock predictions were validated using out-of-bag (OOB) samples and further verified at the plot and stand level using additional data. For validation, field plots from a Management Forest Inventory (MFI) and delineated forest stands were used. Compared to stand-level aggregations based on field plots from the MFI, our approach explained 56% of the variability in the growing stock (R2) with a relative RMSE of 15% at the stand level (n=252). As expected, the scatter was higher at the plot-level (n=3973). Nonetheless, the models still achieved acceptable performance measures (R2=0.44; RMSE=34%).
Recent development in aerial digital cameras and software facilitate the photogrammetric point cloud as a new data source in forest management planning. A total of 151 field training plots were distributed systematically within three predefined strata in a 852.6 ha study area located in the boreal forest in southeastern Norway. Stratum-specific regression models were fitted for six studied biophysical forest characteristics. The explanatory variables were various canopy height and canopy density metrics derived by means of photogrammetric matching of aerial images and small-footprint laser scanning. The ground sampling distance was 17 cm for the images and the airborne laser scanning (ALS) pulse density was 7.4 points m-2. Resampled images were assessed to mimic acquisitions at higher flying altitudes. The digital terrain model derived from the ALS data was used to represent the ground surface. The results were evaluated using 63 independent test stands. When estimating height in young forest and mature forest on poor sites, the root mean square error (RMSE) values were slightly better using data from image matching compared to ALS. However, for all other combinations of biophysical forest characteristics and strata, better results were obtained using ALS data. In general, the best results were found using the highest image resolution.
Automatic Terrain Extraction (ATE) is a key component of digital photogrammetric software. Image correlation has been widely used in ATE and has proved to be a reliable and accurate approach. This paper discusses two algorithms for ATE: (1) multiple image pairs; and (2) back matching. With standard methods of image acquisition, a point on the ground is often covered by multiple stereo image pairs. By correlating on these multiple pairs, ATE computes multiple elevations, which can then be used to detect image correlation blunders and improve elevation accuracy. In difficult terrain such as airports and streets, image correlation algorithms may generate false correlations due to lack of texture and repeated patterns. The back matching algorithm effectively detects those false correlations by checking the consistency between forward matching and backward matching. Initial test results indicate that the multiple image pair algorithm can improve elevation accuracy up to 30%. The back matching algorithm has successfully detected and removed false correlations in a difficult test stereo pair over an airport. This paper presents the theoretical foundation of these two algorithms, supported by test results.
The rapid development in aerial digital cameras in combination with the increased availability of high-resolution Digital Elevation Models (DEMs) provides a renaissance for photogrammetry in forest management planning. Tree height, stem volume, and basal area were estimated for forest stands using canopy height, density, and texture metrics derived from photogrammetric matching of digital aerial images and a high-resolution DEM. The study was conducted at a coniferous hemi-boreal site in southern Sweden. Three different data-sets of digital aerial images were used to test the effects of flight altitude and stereo overlap on an area-based estimation of forest variables. Metrics were calculated for 344 field plots (10 m radius) from point cloud data and used in regression analysis. Stand level accuracy was evaluated using leave-one-out cross validation of 24 stands. For these stands the tree height ranged from 4.8 to 26.9 m (17.8 m mean), stem volume 13.3 to 455 m3 ha−1 (250 m3 ha−1 mean), and basal area from 4.1 to 42.9 m2 ha−1 (27.1 m2 ha−1 mean) with mean stand size of 2.8 ha. The results showed small differences in estimation accuracy of forest variables between the data-sets. The data-set of digital aerial images corresponding to the standard acquisition of the Swedish National Land Survey (Lantmäteriet), showed Root Mean Square Errors (in percent of the surveyed stand mean) of 8.8% for tree height, 13.1% for stem volume and 14.9% for basal area. The results imply that photogrammetric matching of digital aerial images has significant potential for operational use in forestry.
Solar elevation is an important factor in passive, airborne data collection. The minimum solar elevation allowed in missions for topographic mapping is typically 30° from the horizon. A general hypothesis is that the new, high dynamic range, digital large-format photogrammetric sensors allow for high data quality, even with lower solar elevations, which would improve the feasibility and cost-efficiency of photogrammetric technology in various applications. Objectives of this study were to investigate theoretically and empirically the impacts of solar elevation in modern photogrammetric processes. Two cutting-edge aspects of novel photogrammetric technology were considered: point cloud creation by automatic image matching and reflectance calibration of image data. For the empirical study, we used image data collected by a large-format photogrammetric camera, Intergraph DMC, with low (25–28°) and medium (44–48°) solar elevations from 2, 3 and 4 km heights. We did not detect negative influences of decreasing solar elevation during our general evaluations: an analysis of image histograms showed that the ranges of digital numbers could be tuned to similar levels with exposure settings, and evaluations of density and the accuracy of point clouds did not show any reduction of quality. We carried out detailed evaluations in forests, roads and fields. Our results did not indicate deterioration of the quality in sun-illuminated areas with decreasing solar elevation. In shadowed areas, we observed that the variation of image signal was reduced in comparison to sun-illuminated areas and emphasized the issue of complication of reflectance calibration. Artefacts appeared in automatically generated point clouds in areas shadowed by trees, which we observed in flat objects as up to 3 times increased random height variation and decreased success in measuring the terrain surface. Our results also showed that the overall performance of point cloud generation was high. Typically, point clouds could be derived even from a single stereo model with the point density corresponding to the GSD, but some expected and unexpected failures also appeared. The height accuracy was dependent on the object properties and the intersection geometry; the height accuracy was 0.5–2 times GSD at well defined objects. Our conclusions were that in the future it is of increasing importance to quantify the sensitivity of different methods on the radiometric properties of the image data. It is also important to develop interpretation methods that are not sensitive to shadows, in order to enable optimal use of photogrammetric technology in normal to rapid response applications.
Experiences from Nordic countries and Canada have shown that the retrieval of the stem volume and mean tree height of a tree or at stand level from laser scanner data performs as well as, or better than, photogrammetric methods, and better than other remote sensing methods. This paper reviews the methods of small‐footprint airborne laser scanning for extracting forest inventory data, mainly in the boreal forest zone. The methods are divided into the following categories: extraction of terrain and canopy height model; feature extraction approaches (canopy height distribution and individual‐tree‐based techniques, techniques based on the synergetic use of aerial images and lidar, and other new approaches); tree species classification and forest growth using laser scanner; and the use of intensity and waveform data in forest information extraction. Despite this, the focus is on methods, some review of quality obtained, especially in the boreal forest area, is included. Several recommendations for future research are given to foster the methodology development.
Airborne laser scanning (ALS) is currently one of the most promising remote sensing techniques for quantitative retrieval of forest parameters. While ALS has reached an operational status for mapping of boreal forests, its large area application over mountainous environments is lacking behind. This is because alpine forests often have high horizontal and vertical structural diversity and are situated in steep terrain. Also, ALS data acquisition and processing is more demanding over mountainous areas than over relatively flat regions.In this study we have used state-of-the-art ALS technology and software packages to map canopy heights and to estimate tree heights for a 128 km2 region in the western part of the Austrian Alps. Spruce and fir are the dominant tree species. Rather than employing data and methods tuned for a particular task and for a small study area, we solely use data and methods which already serve other operational applications. Thus, it is ensured that the results obtained in this study are of practical relevance.For the validation of the ALS derived canopy heights we have used 22 000 ground control points and field-measured forest inventory data from 103 sample plots, which are operationally used by the local forest administration. The validation of the digital terrain model (DTM) with the ground control points shows that over non-forested terrain DTM errors increase from 10 cm for relatively flat terrain (local slope < 10°) to over 50 cm for local slopes greater than about 60°. The validations of the ALS derived single-tree heights and Lorey's mean heights show good correlations using both, three dimensional first pulse points (R2 = 0.73–0.84) and a grid-based canopy height model (R2 = 0.68–0.87). Overall, the results demonstrate that airborne laser scanning has now reached the maturity to be used for mapping canopy heights of complex alpine forests throughout large areas.