Pléiades satellite images for deriving forest metrics in the Alpine region

Highlights

• Pléiades CHMs could be useful alternatives to those from aerial image matching for monitoring forest metrics in mountain forests.

• Pléaides images must be acquired in the summer season to reduce shadows on the ground cast by tree and the terrain itself.

• Forests must provide enough foliage (i.e. leaf-on condition) for deriving Pléaides CHMs.

• Pléiades canopy gap detection works best in very bright areas.

• Pléiades forest metrics show high variance in steep terrain (>50°).

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 (H_max), height percentiles (Hp95, Hp50), the standard deviation of the height values (H_Std) 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 -0.25 m (H_max), 0.33 m (Hp95), −0.03 m (H_Std) 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.

Introduction

The forested area in the realm of the Alpine Convention amounts to roughly 87,000 km², which is around 46% of the Alpine Convention’s territory (Alpconv, 2015). Mountain forests in the Alpine region provide a wide range of ecosystem services (or functions), especially protective, social, and economic ones. These include other forest services such as biodiversity, fresh and clean air water supply, or attractive landscape that are finally relevant for other cultural and tourism-related economic services (Cetara and Mannoni, 2013). A protective function for human infrastructures against natural hazards like flooding, debris flows, landslides, avalanches, and rockfalls, is one of the key services of forests in the Alps (Cetara and Mannoni, 2013). For instance, about half of the forested areas in Switzerland have some protective function, 30% of Austria’s forests have a protective function against natural hazards, and 63% and 42% of the forested areas in Bavaria (Germany) have one against soil erosion, and avalanches, respectively (Alpcon, 2018). To maximise the protective function of forests, forest management should ensure mixed tree species composition, and multi-layered forest vertical structure formed by trees of different ages to guarantee canopy continuity, and thus permanent soil cover (Alpcon, 2018). Canopy cover is an important structural attribute associated with the variance in canopy gaps (McElhinny et al., 2005). To create and maintain such conditions, tree felling and regeneration are necessary. Besides forest management, information about the canopy height, gaps distributions, and the main tree species (i.e. broadleaf and coniferous trees) are essential for nature and biodiversity conservation.

On the national level, much of this information comes from National Forest Inventories (NFI), which however rely on plot-based field sampling. Therefore, these measurements are limited with regard to spatial coverage, and they cannot capture the heterogeneity of forest structure across the landscape with a high spatial resolution (Zald et al., 2016), especially in remote and/or unmanaged forests. As complementary data source to sample-based field measurements, remotely sensed data is widely acknowledged as essential to derive the most important forest variables at local, regional (landscape), and national levels. Current remote-sensing systems to measure three-dimensional (3D) forest structure attributes over large scales include active sensors, such as synthetic aperture radar (SAR) or light detection and ranging (LiDAR), and passive optical sensors like high-resolution digital aerial imagery, and very high-resolution satellite images (VHR). Besides the costs, the selection of data source mainly depends on the forest parameters to be derived from the data (Koch, 2013), and on the required spatial and temporal resolution.

Since the last decade, airborne laser scanning (ALS) is the primary data source operationally (Hollaus et al., 2006) used in Scandinavia (McRoberts et al., 2010; Hyyppä et al., 2012; Latifi et al., 2015) for acquiring 3D information of forests. At regional scales, ALS is the most accurate remote sensing technique available, as demonstrated in several research studies (Hyyppä et al., 2008; Breidenbach et al., 2010; Yu et al., 2015). In fact, ALS has the capability to penetrate through vegetation, providing both a digital surface model (DSM) and a digital terrain model (DTM) (Ressl et al., 2016), which is needed to derive the vegetation height above the ground (i.e. the normalized DSM: nDSM), computed as their difference. Still, data from passive, optical sensors have gained steadily importance in the forest community since they have one-third to one-half of the costs of ALS data (White et al., 2013; Bühler et al., 2015) and they are better suited for tree species classification (Waser et al., 2011; St-Onge et al., 2015; Fassnacht et al., 2016; Zald et al., 2016; Puliti et al., 2017). Thanks to the recent innovations in the field of digital photogrammetry, a large number of studies demonstrated the great potential of dense image matching algorithms for deriving high-resolution DSMs from aerial images with levels of accuracy close to those from ALS data (Bohlin et al., 2012; Müller et al., 2014; Stepper et al., 2014; Gobakken et al., 2015; White et al., 2015; Ginzler and Hobi, 2015; Puliti et al., 2017; Nyamgeroh et al., 2018). Consequently, aerial photogrammetry in combination with a pre-existing DTM is considered as an alternative to ALS for the regular update of canopy height models (Vastaranta et al., 2013; Ullah et al., 2015, 2019) to such an extent that, in a forest inventory context and monitoring program, subsequent re-measurements are already done operationally with aerial images on the national level. Ginzler and Hobi (2015) demonstrated the successful generation of a nationwide DSM and canopy height models (CHM, i.e. nDSM within forests) in Switzerland by image matching of digital aerial imagery. Since aerial DSMs represent only the top most surface of forest canopies, White et al. (2013) stated that 3D information extracted from images contains less information about the vertical distribution of vegetation within the forest canopy than ALS data. Independent of using photogrammetry or laser scanning, the long acquisition times of airborne remote sensing present a major challenge to its use for forest monitoring at the landscape scale. In fact, full coverage of a whole country cannot be achieved in a single airborne inventory, and multiphase inventories present a logistic challenge (Koch, 2013). For instance, in Switzerland, one-third of the country is acquired by aerial images every year according to a predefined subdivision. Furthermore, the aerial image acquisitions are normally planned in large image blocks within time spans of a few days, and they depend on stable weather conditions. Consequently, as addressed by Bohlin et al. (2017), when using aerial images over large areas to map forest attributes, the images are captured at different times of the day and days of the year with respective variations of the lighting conditions. Hence, shadows may be cast differently for different flight strips over the same area (Honkavaara et al., 2012), which affects the estimation accuracy of the mapped attributes. In comparison with airborne remote sensing, VHR stereo satellite imagery, which can now acquire images of up to 30 cm spatial resolution, has the benefit of wider area coverage and spatially/temporally more homogeneous image content with short repetition intervals (Immitzer et al., 2016; Persson, 2016; Persson and Perko, 2016). Additionally, the higher correlation between the spectral data of VHR satellite images and vegetation parameters make this technology a valuable solution for deriving forest information and update perspective at the national or sub-national level, especially in remote areas like mountain forests (Uddin et al., 2015). For instance, the use of spectral vegetation indices, which usually involve differences between near infrared and other bands, can be used to detect the vegetation condition and consequently fire risk (Woźniak, 2011) and insect attack, specifically in mountain forests that are permanently exposed to stress related to climate conditions, water availability, fungi and insects (Modzelewska et al., 2017).

Most recently, VHR satellite images are of growing importance for improving the details of land cover and land use maps and for the assessment of forest parameters like crown detection, delineation, and tree species classification (Shamsoddini et al., 2013). White et al. (2016) provide a comprehensive review of the potential of high spatial resolution satellite imagery for forest inventory. However, their operational use for deriving forest parameters remains rather uncommon because little is known yet about the feasibility of calculating relevant structural forest data from VHR satellite-based stereo data, particularly in the context of difficult topographic conditions and heterogeneous forests. An additional aspect to consider is that VHR satellite imagery products vary in their spatial and spectral resolutions, geographic and temporal coverage, and security/priority regulations, which are directly reflected by their price. For example, not all areas of the globe have equal access to VHR data, and very fine-scale satellite data with high spatial, spectral and temporal resolution are rather expensive. Those aspects can hamper their consistent application in forestry, in an operational context and in research. Furthermore, the few present studies differ in site conditions, tree species, sensor resolutions, and image acquisition parameters. The most successful investigations are related to boreal and coniferous forests with dense canopy cover (Neigh et al., 2014; Persson, 2016; Persson and Perko, 2016; Yu et al., 2015; Wittke et al., 2019), whereas for Australian tropical savanna, Goldbergs et al. (2019) claim that currently commercially available VHR satellite data (0.5 m resolution) are not well suited to estimating canopy height variables. We assert that forestry research, as well as forest management and monitoring programs, would benefit from a better understanding of the performance of VHR satellite imagery used for forest inventory in mountainous regions.

In this respect, the objective of this work is to provide an assessment of some forest metrics of mountain forests in the Alpine region obtained from VHR Pléiades stereo satellite images by investigating the impact of topography characteristics of shadow, aspect, and slope. For deriving forest height information, there are currently only a few studies available, and particularly few for heterogeneous forests with a complex vertical and horizontal structure. Recently, Pearse et al. (2018) demonstrated that point cloud data obtained from Pléiades stereo satellite imagery are useful for predicting forest inventory attributes in New Zealand’s planted forests. In comparison to that work, we focus our investigations on alpine mountain forests, and we derive forest attributes from the Pléiades CHM. We outlined a grid-based approach to derive the main forest metrics (descriptive statistics) such as the maximum height, the height percentiles (95%, 50%), the mean and standard deviation of the height values and the canopy gaps. Furthermore, we investigate the reliability of Pléiades time series data for change detection. We compare the Pléiades results with those obtained from aerial image matching, a technology already used in an operational inventory context. For each study area, we use existing forest masks and ALS-derived DTMs for deriving the CHM.