European Facility For Airborne Research
June 28, 2022, 23:14
Airborne remote sensing for monitoring essential biodiversity variables in forest ecosystems-A
TA-013. Proposals for training courses in hyperspectral imaging applications or in-situ sampling
Professor Andrew Skidmore is Head of the Department of Natural Resources at the Faculty of Geo-Information Science and Earth Observation (ITC) of the University of Twente in the Netherlands. Originally a graduate in forestry from the Australian National University (ANU), he worked for 10 years with the Forestry Corporation of New South Wales (NSW) and undertook a PhD in remote sensing and GIS at the Australian National University. In 1991 he moved to the School of Geography, University of New South Wales, and also worked as member and later Director of the Centre for Remote Sensing and GIS. Vegetation mapping and monitoring have been his main ongoing research theme, while recent research includes wildlife habitat assessment in East Africa, hyperspectral remote sensing, AI techniques for handling geoinformation and accuracy assessment.
Skidmore, A.K., Pettorelli, N., Coops, N.C., Geller, G.N., Hansen, M., Lucas, R., Müncher, C.A., O'Connor, B., Paganini, M., Pereira, H.M., Schaepman, M.E., Turner, W., Wang, T.J. and Wegmann, M. (2015) Environmental science : agree on biodiversity metrics to track from space : comment. In: Nature : international weekly journal of science, 523 (2015)7561 pp. 403-405.
Ullah, S., Skidmore, A.K., Naeem, M. and Schlerf, M. (2012) An accurate retrieval of leaf water content from mid to thermal infrared spectra using continuous wavelet analysis. In: Science of the Total Environment, 437(2012), pp. 145-154.
Zhu, X., Wang, T.J., Darvishzadeh, R., Skidmore, A.K. and Niemann, K.O. (2015) 3D leaf water content mapping using terrestrial laser scanner backscatter intensity with radiometric correction. In: ISPRS Journal of Photogrammetry and Remote Sensing, 110 (2015) pp. 14-23
Wang, Zhihui., Skidmore, A.K., Darvishzadeh, R., Heiden, U., Heurich, M. and Wang, T.J. (2015) Leaf nitrogen content indirectly estimated by leaf traits derived from the PROSPECT model. In: IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 8(2015)6, pp. 3172-3182
Ali, Abebe M., Darvishzadeh, R., Skidmore, A.K., van Duren, I.C., Heiden, U. and Heurich, M. (2016) Estimating leaf functional traits by inversion of PROSPECT: Assessing leaf dry matter content and specific leaf area in mixed mountainous forest. In: International Journal of Applied Earth Observation and Geoinformation : JAG, 45 (2016)Part A pp. 66-76.
Forest management requires the use of comprehensive remote sensing data which enable monitoring biodiversity changes in response to calamities such as bark beetle infestation and other climate change induced phenomena. They also enable to predict the long-term impact of management decisions. Although the benefits of remote sensing for monitoring vegetation are well recognized, yet accurate and site specific monitoring of many essential biodiversity variables in forest ecosystems remain elusive. In this training course, the special skills required for processing the new generation of airborne and satellite hyperspectral, thermal and LIDAR data for retrieving essential biodiversity variables in forest ecosystems will be presented. In forests, bidirectional effects mainly influence hyperspectral airborne signals and directly affect the accuracy of derived variables. Simultaneous acquisition of thermal, VIS/NIR hyperspectral and LIDAR data (See RS4forestEBV-B) allow accurate retrieval of vegetation parameters (e.g., LAI, chlorophyll, SLA, nitrogen, water content, species occurrence and 3D vegetation structural attributes) which have been recognized as essential biodiversity variables by GEO-BON and are crucial in forestry and national park management practices. Several ongoing projects will support this training course including the ESA Innovator III project (RS4EBV). The participants will be trained in remote sensing algorithms and retrieval of essential biodiversity variables. The BIOKLIM project which is coordinated by Bavarian Forest National Park (BFNP), will provide data and expert knowledge on forest structure, biodiversity and management issues as well as facilitate access to the field sites, flux towers and field data collection techniques.
Partenavia - IMAA
•TASI-600 operated by CNR (TASI Partenavia IMAA)
The aircraft is able to carry the TASI-600 sensor on board and is operated by CNR. The thermal high spatial resolution TASI-600 sensor will provide high spectral resolution data at 8-11.5 μm thermal infrared region in 32 spectral channels. The TASI radiance is expressed in [μW cm-2 sr-1 nm-1] and the spatial resolution is about one meter. The sensor is among the limited hyperspectral thermal sensors which can be installed in an aircraft and is recently calibrated in December 2015.
Please note that due to unavailability of NERC facilities, we could not find an airborne laser scanner. Having such a data would indeed be an added value to the training course as it will provide accurate structural information regarding the plant traits of interest. Therefore, other available instruments which provide the most suitable options will be selected. The BFNP will acquire data for the entire study area in 2016. If a simultaneous airborne acquisition is not possible in 2017 by EUFAR aircraft due to availability and cost, it may be possible (but not entirely optimal) to use the 2016 LIDAR data in the training course.
Biophysical and biochemical vegetation parameters can characterize changes in biodiversity through changes in ecosystem structure and function. Although remote sensing, especially high spatial resolution hyperspectral imagery can be used to measure many biophysical and biochemical variables, retrieval of these parameters across different remote sensing systems to understand their dynamics remains an open challenge and the uncertainty sources and multiple approaches has to be taken in account. Commonly these data are used as stand-alone sources for retrieving the vegetation traits. However, the challenge is that in the presence of different data sources, which sources or combination of sources are most suitable for retrieving a specific variables and how these data sources can be used complementary. Therefore, two parallel proposals (RS4foestEBV- A and B) are suggested to address the aim of this training course which is to demonstrate how different remote sensing data and in-situ measurements of plant traits 1) can be used to model vegetation and 2) be linked to image data inversion in order to retrieve plant variables and map their spatial patterns. Accordingly, in this course the skills required for processing the new generation of airborne hyperspectral, thermal hyperspectral and LIDAR data for retrieving forest essential biodiversity traits will be presented. The course will highlight the added value of airborne data for forest management and biodiversity monitoring. Of special relevance is the envisaged test site (Bavarian Forest National Park) which encompasses a wide range of heterogeneous spatial patterns of temperate vegetation, where disturbances such as bark beetle calamities and storm damage substantially alter the structure of coniferous and mixed stands and cause changes to biodiversity. The simultaneous acquisitions of high spatial resolution airborne hyperspectral (See RS4forestEBV-B), thermal hyperspectral and LIDAR data allow us to better understand and monitor vegetation parameters. Furthermore, the ground data collection aims to provide the course participants with knowhow on tools (field spectroscopy, thermal spectrometry and terrestrial LIDAR) and measurement techniques to collect different vegetation variables. The training course enables the participants to achieve the following learning objectives:•To map different vegetation parameters using hyperspectral visible/NIR /thermal and LIDAR data ;•To understand the advantage of each data sources and the best combinations of them for retrieving vegetation parameters;.•To understand data processing chains;•To understand the challenge of collecting and integrating forest field data with remote sensing imagery;
The training course will be structured as 4 stand-alone but interlinked working groups as described in the attachment. Although each working group will be piloted independently in the training course, they are all scientifically related and each forms an important component of the training course project.
The overall themes of the working groups and their lead are as follows:
•Working Group 1- Hyperspectral: to map the species occurrence, biophysical, biochemical properties and plant traits of the study area with high spectral resolution imagery by empirical and radiative transfer models; Lead by Roshanak Darvishzadeh;
•Working Group 2- Thermal hyperspectral: to map the species occurrence, biophysical and biochemical properties of the study area with hyperspectral thermal imagery; Lead by Martin Schlerf; •Working Group 3- Atmospheric correction: to provide airborne reflectance comparable with proximal sensing including BRDF estimations and validation; Lead by Tiejun Wang•Working Group 4- LIDAR: to map the spatial patterns of heterogeneity and structural characteristics of the forest with laser scanner data; Lead by Marco Huerich
All course participants (Working Groups) will be familiarized with the design of the flight and will be involved in sampling design and field measurements of the plant traits which will be used for analysis of the acquired images. They will all have the opportunity to learn tools and techniques used during vegetation field data collection.
The University of Twente, Faculty of ITC, will be the coordinator and train participants in remote sensing algorithms for retrieval of essential biodiversity variables directly from acquired images. ITC has conducted several field measurements in BFNP and collected vegetation traits for the last three years. The BFNP as the key user has already established 330 sampling plots along four main straight transects covering the altitudinal and structural gradients. All environmental variables are derived from field measurements, aerial photographs, LIDAR data and climate stations. In 2016, the field measurements will be repeated in about half of these sample plots (i.e., 157) jointly with a LIDAR flight campaign covering gradients of altitude and forest structure. Moreover aerial photography is taking place every year in the park and the historical data are available. These huge source of existing data are an advantage for the training course, since the hands on data will be already accessible for the participants and can be used as reference dataset. Apart from participants of the training course, the collected data will be used by several PhD students who are studying the retrieval of different vegetation variables in the using different data sources at filed and airborne level. As there is an urgency to better characterize and understand forest ecosystem status in a time of rapid climate and landscape change, this training course offers EUFAR an opportunity to demonstrate innovative airborne science as well as having a valuable training component for PhD students and young carrier researchers.
The proposed activities require clear sky conditions (very low cloudiness and haze for hyperspectral sensors). Some (cumulus) clouds (up to 1/8) can be accepted.
The time schedule for the overflight requested in the present training course is the month of July in summer 2017 concurrent to ICARE conference which is hosted by DLR. Therefore, the ferry to the survey area will be minimized, since it is foreseen, that all aircraft will be all located at DLR during ICARE.Coincident times with the overpasses of Sentinel 2 are also desirable. The time of the flight should be as close as possible to local solar noon (i.e. from 10 AM to 2 PM local time for data acquisition) so as to minimize the effect of the anisotropy of the surface on reflectance measurements and under light winds to ensure stable flying conditions.
The area chosen for this study is the Bavarian Forest National Park which is more heterogeneous in tree species than similar areas in the region. It is located in south-eastern Germany along the border with the Czech Republic (490 3’ 19” N, 130 12’ 9” E). The park has a total area of 24,218 hectares. The study site is included in several European research projects and hands on data are available for the training course with the support of the “Data pool initiative for the Bohemian Forest Ecosystem”. Temporal airborne hyperspectral, LIDAR and aerial photography as well as high spatial resolution satellite images such as Rapid Eye, SPOT-5 and Sentinel2 acquired during the growing season of 2015 and 2016 in support of the RS4EBV project are available together with field measurements of plant traits for consecutive years.
We propose one flight in July 2017 for approximately 10 hours with respect to the distance of the base APT.
The vegetation variable measurements will be conducted at two level:
•At the leaf level: Chlorophyll measurements, leaf area, fresh biomass, dry matter content, water content, SLA, Nitrogen and stomatal conductance measurements.
•At the canopy level:Leaf Area Index (LAI), Average Leaf Angle (ALA), canopy gap fraction, tree height, crown radius, crown height, diameter at breast height, diameter at the ground, tree density/canopy density, number of trees per plot and species abundance.
Moreover, thermal field spectrometry of background materials and FLIR camera for calibration of thermal images will be conducted. In addition, Terrestrial Laser Scanner (TLS) will be used for understory structural measurements and complementary to airborne LIDAR data.
TASI-600 operated by CNR
We could not find an operating airborne for laser scanner up to now. Having such a data would indeed be an added value to the training course as it will provide accurate structural information regarding the plant traits of interest.
The scientific group will provide all the necessary equipment for field data measurements, forest inventory and ground truth collection of plants traits as well as Thermal field spectrometer (MIDAC), and Terrestrial Laser Scanner (TLS).
Parallel to the time of overflight a 2-day field campaign will take place, during which vegetation variables mentioned in section 3.1 will be collected at leaf and canopy level (please see parameter / measurement required 3.1). Thermal spectrometry measurements of background surface will be taken by MIDAC thermal spectroradiometer and FLIRT600 thermal camera in order to calibrate and validate the atmospheric correction of TASI-600 thermal sensor during pre-processing. TLS will be used in sampling plots to measure the structural vegetation properties and to complement the measurements taken by optical instruments (such as LiCor and hemispherical camera) and to validate the results obtained from LIDAR data. Vegetation parameters will be collected using different field equipment such as SPAD 502 and CCM-300 leaf chlorophyll meters, Li 3000 and AM350 leaf area meters, SC-1 Leaf Porometer for stomatal conductance measurements and digital scale for the weighing of leaves. Leaf area index and other related parameters such as the proportion of foliage and non-foliage elements will be collected using hemispherical photography, LAI-2000, LAI-2200 Plant Canopy Analyzer and TRAC. Stand characteristics (e.g. height, DBH, crown diameter) will be recorded using tapes, hectometers, Haga altimeter and other instruments. Respectively, leaf samples will be collected for subsequent laboratory analyses (LDMC, LWC and Nitrogen). Differential GPS will be used to determine accurate location of the sample plots and for geometric correction of the images as well as for validation purposes. At each plot, all species with more than 1% area coverage will be identified and sampled. The documentation of measurements in each sample plot will be accompanied by additional photographs.
TASI-600 images geocoding is performed by using the ITRES GEOCOR software, which incorporates photogrammetric bundle adjustment solutions to produce orthorectified images using precision IMU/GPS and terrain height data. More in detail, the geo-correction of the TASI-600 images is performed in a sequence of 4 steps based on retrieving from the raw file the attitude and GPS data, synchronizing it with respect to the sensor frames acquisitions, differentiating the GPS and merging all these info in the navigational file describing the precise trajectory of the TASI sensor’s center of mass. The attitude processing is finalized to create a pre-processed attitude data file that becomes then the attitude data input for the navigational file. The atmospheric correction of TASI depends on an earlier agreement with CNR. However, CNR is able to further process the TASI radiance images using the standard ISAC (in-scene atmospheric compensation) atmospheric correction procedure. This algorithm is well established and commonly used for in-scene atmospheric thermal data correction, because it requires only the calibrated, at-sensor radiance data to estimate the upwelling radiance and transmissivity of the atmosphere. Furthermore, the field emissivity measurements using MIDAC, will be used as reference measurements for atmospheric correction. TLS data will be used for correction of airborne LIDAR data at dense canopies where the signal penetration were weak. Classification of the images (using ENVI software) will be done to discriminate species, and to determine the areas under stress such as areas under bark beetle attack using spectral analysis.
Spectral reflectance, emissivity and returned beams of sample plots will be first extracted from the images using an in-house developed program and further investigated using different statistical analysis in MATLAB/ R programs to understand their relations with collected parameters. Several key plant biochemical and biophysical characteristics which are counted as key indicators of plant physiological status and health ( LAI, chlorophyll content, water content, canopy architecture and density) will be considered in INFORM radiative transfer model (RTM) (within MATLAB) and statistical models to explain their relations with spectral reflectance, emissivity, and discrete return and full-waveform LIDAR. In other words, it will be investigated which parameter is retrieved accurately using data from different parts of the electromagnetic spectrum and how these data can be used complementary. For this physical (3D RTM, INFORM) and empirical approaches (vegetation indices, multivariate techniques (PLS, PCA), machine learning algorithms (ANN)) will be utilized separately for different data sets and a number of statistical issues concerning validation (sampling size effect, role of cross validation, biasedness) will be additionally discussed. The results obtained from different datasets will be compared and validated with the field measurements and maps of the plant traits will be generated using ENVI, MATLAB and R programs. As the bridging of these datasets (obtained from RS4foestEBV-A and B) is the challenging investigation for this training course. At the end of the training course each working group (students) will present their findings and the milestones in their domain and provide recommendation for further work.
The applicants are actively working in different projects (as such European Space Agency Innovator III project,‘RS4EBV’) which involve retrieval of essential biodiversity variables from hyperspectral and high resolution multispectral data . Theses research projects would fund staff time from the University Twente, Netherlands. All involved scientists and lecturers from research institutes and universities are funded by their institutions for their input time. Staffing is further guaranteed through 4 PhD scholars funded by the EU and the Dutch government through grants awarded to Skidmore, Darvishzadeh and Wang. All computer facilities and field equipment necessary for data processing and field campaign are available within the University of Twente (Faculty of ITC). ITC has also a well-equipped laboratory which will facilities the biochemical measurements.
July 2017- Parallel to ICARE conference.
This training course through the RS4forestEBV projects would help to build up an expertise not only in the field of hyperspectral VIS/NIR, thermal remote sensing, but also LIDAR among researchers, PhD as well as postdocs. The 4 PhD students involved in the project would certainly benefit from these image data for their research and analysis.
At least one scientific publication which integrates the outcomes of the 4 working groups is anticipated to be another output of the training course.
John Hearne (email@example.com)Raymond Kokaly (firstname.lastname@example.org)Terry Dawson (email@example.com) Katarzyna Dabrowska-Zielinska (firstname.lastname@example.org)
The proposed training course will be coupled to current granted projects by European Space Agency projects which facilitate funding of personnel and working costs.
From your colleagues