The synergy between multiple optical sensors response, related to data acquisition, processing and data analysis, it is very important to be realized for a better understanding on the dynamic response of vegetation in relation to climatic changes. At this moment, there are several ESA ongoing missions, such as Sentinel 2 and 3, Earth Explorer FLEX and other Copernicus missions, that have to provide a complete overview on our planet photosynthetic activity. The new COST Action, CA17134 – “Optical synergies for spatiotemporal sensing of scalable ecophysiological traits”, aims to support these missions’ activities, one of its primary tasks being to develop the best practice procedures and protocols for calibration and validation of multiple types of sensors, on both, spatial and temporal scales. These protocols should be applied by research groups from entire Europe, providing their measurements traceability chain, ensuring the quality of data and reduced measurements uncertainties. In this context, an important starting point is to have a good knowledge on used instrumentation (from both: radiometric and spectral point of view).
The purpose of this paper is to present some practical recommendations for accurate spectral characterization and calibration of six passive optical sensors. These optical systems have different spectral capabilities and are provided by different producers. Thus, the spectral capabilities for tree fibre optic mini-spectrometers - type S2000 produced by Ocean Optics, two high resolution optical fibre spectrometers with thermo-electric cooling - one spectrometer provided by Ocean Optics (QE65000) and one spectrometer provided by Avantes (AvaSpec-ULS2048LTEC SensLine), and one spectroradiometer with multiple spectral configurations from Gooch&Housego (OL750) have been investigated in order to develop a complete and clear procedure for spectral calibration best practice. Two distinct calibration methods have been applied and will be presented here together with a complete list of possible sources of uncertainties. The sources of error are derived from measurement configuration, setup geometry and selected calibration methodology.

Area B - Board 197

We present the first results from the ESA CryoVex/KAREN land ice campaign on the EGIG line of the Greenland Ice Sheet, conducted in 2016 and 2017. The campaign was aimed at the EGIG-line to document the different glacial facies along this well studied line of way-points, from the ablation zone at the ice sheet margin, to the dry snow zone at the ice sheet summit. Shallow ice cores, snow pits, and in situ ku-band radar data was collected at 9 sites on the EGIG line, and this data was used in conjunction with airborne data collected along the same transect. Depth to power profiles at sites T9, T12 and T19 on the EGIG line show good agreement between the in-situ radar and airborne Ku-band data, despite differences in the bandwidth and therefore vertical resolution of each sensor. In particular, multiple layers in both the ASIRAS and pRES data, and an extremely strong power return, is found at approximately 5 meters depth. Our results show that this highly reflective layer may be the 2012 melt layer, and also suggests that in the dry snow zone the EGIG line volume scattering from up to 6 meters depth should be expected in airborne ku-band data. Ice cores were drilled to a maximum depth of approximately 6 meters at all 9 field sites to obtain snow stratigraphy and density measurements. We compared a density profile from RACMO (Regional Atmospheric Climate Model) with all 9 firn cores and found that there is a good agreement between the in situ and modelled densities. The density decreases along the profile as we move further from the ice sheet margin where the surface density is almost at the density of ice. At T12, a very dense layer can be seen in the firn core at a depth of 1.5 m and at a similar depth of 1.7 m a dense layer is also found in the modelled density. To investigate how the snow properties affect the elevations measured by the Ka and Ku-band airborne data, we looked at the difference between Ka and Ku-band airborne data over a section of the EGIG line. These preliminary results show that as expected the Ku-band sensor penetrates approximately 5.2 meters deeper into the snowpack than ka-band data collected along the same track.

Area B - Board 198

The Bohemian Forest right at the national border between the Czech Republic, Austria and Bavaria in Germany is a unique area of forested mountains and plateaux with relatively harsh climates and many peat bogs. Historically, the human impact in the region was comparability low and therefore led to the preservation of a unique forest ecosystem. The centre of the area is formed by two contiguous protected areas, the Bavarian Forest National Park (240 km2) and the Šumava National Park (690 km2), which are surrounded by the Bavarian Forest Natural Park (3,000 km2) and the Bohemian Forest Protected Landscape Area (1,000 km2). This area provides a perfect field of research to study natural and near natural ecosystems.
Since January 2015 the cooperation agreement “Data Pool Initiative for the Bohemian Forest Ecosystem” exists between the Bavarian Forest National Park, the Šumava National Park and a diversity of European research organizations and universities. Overall, 9 partners did sign the agreement supporting the mutual goal to contribute to the data pool with remote sensing data as well as with in-situ data including adequate metadata information respectively. The steadily increasing data pool provides input for a multitude of scientific questions concerning the two national parks, mainly related to forestry. The main objectives of the data pool initiative are to store and provide large remote sensing data sets, to offer a testing environment for new sensor developments, to serve as a laboratory for Essential Biodiversity Variables (EBV), to facilitate collaborative research and to create information for the national park management.
Within the last 4 years, 15 airborne surveys did take place with four different remote sensing systems, namely LiDAR, digital imagery, hyperspectral reflective and thermal imagery and SAR. In addition to the airborne campaigns, numerous intensive field activities did take place. The overall suite of data is complemented by satellite data, like SPOT-5, TerraSAR-X, ALOS and of course the data products of the Sentinel satellite systems. Until now, more than 40 publications covering a variety of scientific fields like forestry, ecology and animal conservation did follow from the data pool initiative.
This talk will give an overview of selected research projects and their results originated from the Data Pool Initiative for the Bohemian Forest Ecosystem.

Area B - Board 199

The extremely high species richness of tropical forests (typically more than 100 species per ha) constitutes a major hurdle to their sustainable management. Ground inventories are useful but time consuming and expensive to conduct notably because of difficulties to penetrate in deep forests. The representativeness of localised inventory plots is also of concern. The advances in remote sensing represent an opportunity to address the limitations mentioned above.
The combined use of LiDAR and hyperspectral sensors is a particularly promising avenue to characterize tree species diversity. Both sensors have complementary characteristics. LiDAR data informs on the geometric structure of objects in 3 dimensions and penetrates the canopy thus providing some information on the understory. Being active this technology is largely insensitive to acquisition parameters (solar illumination, cloud or tree shadows). Conversely, whereas hyperspectral imagery is sensitive to acquisition parameter and is restricted to imaging the canopy surface, high spectral resolution confers a strong discriminative power.
Comparisons of both sensors capabilities have shown better performance of LiDAR data for the delineation of tree crowns in heterogeneous forests [1]. On the other hand, once the crown are segmented, information on canopy reflectance allows an accurate identification of tree species. In some cases LiDAR data may also contribute to species recognition [2]. To take full advantage of the complementarity of both technologies an efficient fusion method is required. A promising methodology consists in back-projecting LiDAR point clouds data into the hyperspectral cube to avoid local geometrical distortions occurring during orthorectification [3]–[5].
Fine co-alignment of spectral and point cloud data then allows to deploy a segmentation method based on the position of the points [6]- [7], and supplemented by a spectral proximity in order to regularize the segmentation [5], [8]. Recently a similar approach applied to multispectral LiDAR proved effective in a simpler context [9]. This methodology can be generalized for colorized point clouds obtained from merging LiDAR point clouds and hyperspectral data cube.
Here we evaluate the improvement in individual tree segmentation and species recognition resulting from data fusion. Our test site is a neotropical forest track in French Guiana. The remote sensing data is composed of high density full-wave form LiDAR (Riegl LMSQ780) > 20 emitted pulse/m2, high resolution RGB-NIR imagery (10cm GSD), hyperspectral data from two HiSpex cameras providing VNIR data (160 bands 0.4–1.0 µm ) @1m GSD and SWIR data (288 bands, 1.0–2.5 µm) @ 2m GSD.  A data set comprised of >2300 tree crowns which were delineated and identified to species level was collected as ground truth. All ground and remote sensing data were acquired in 2016.


[1] M. Dalponte, H. O. Ørka, L. T. Ene, T. Gobakken, et E. Næsset, « Tree crown delineation and tree species classification in boreal forests using hyperspectral and ALS data », Remote Sens. Environ., vol. 140, p. 306‑317, janv. 2014.
[2] M. Dalponte, L. Bruzzone, et D. Gianelle, « Fusion of Hyperspectral and LIDAR Remote Sensing Data for Classification of Complex Forest Areas », IEEE Trans. Geosci. Remote Sens., vol. 46, no 5, p. 1416‑1427, mai 2008.
[3] R. Valbuena, « Integrating Airborne Laser Scanning with Data from Global Navigation Satellite Systems and Optical Sensors », in Forestry Applications of Airborne Laser Scanning, vol. 27, M. Maltamo, E. Næsset, et J. Vauhkonen, Éd. Dordrecht: Springer Netherlands, 2014, p. 63‑88.
[4] R. Valbuena, F. Mauro, F. J. Arjonilla, et J. A. Manzanera, « Comparing airborne laser scanning-imagery fusion methods based on geometric accuracy in forested areas », Remote Sens. Environ., vol. 115, no 8, p. 1942‑1954, août 2011.
[5] M. Brell, K. Segl, L. Guanter, et B. Bookhagen, « Hyperspectral and Lidar Intensity Data Fusion: A Framework for the Rigorous Correction of Illumination, Anisotropic Effects, and Cross Calibration », IEEE Trans. Geosci. Remote Sens., vol. 55, no 5, p. 2799‑2810, mai 2017.
[6] A. Ferraz, S. Saatchi, C. Mallet, et V. Meyer, « Lidar detection of individual tree size in tropical forests », Remote Sens. Environ., vol. 183, p. 318‑333, sept. 2016.
[7] A. Ferraz et al., « 3-D mapping of a multi-layered Mediterranean forest using ALS data », Remote Sens. Environ., vol. 121, p. 210‑223, juin 2012.
[8] M. Brell, C. Rogass, K. Segl, B. Bookhagen, et L. Guanter, « Improving Sensor Fusion: A Parametric Method for the Geometric Coalignment of Airborne Hyperspectral and Lidar Data », IEEE Trans. Geosci. Remote Sens., vol. 54, no 6, p. 3460‑3474, juin 2016.
[9] W. Dai, B. Yang, Z. Dong, et A. Shaker, « A new method for 3D individual tree extraction using multispectral airborne LiDAR point clouds », ISPRS J. Photogramm. Remote Sens., vol. 144, p. 400‑411, oct. 2018.

Area B - Board 200