Progress report: Status in January 2008

Leaf chlorophyll content of Norway spruce canopy retrieved from CHRIS/PROBA satellite hyperspectral image data

A physical based retrieval of leaf chlorophyll content (Cab) of young Norway (N.) spruce forest stands from hyperspectral airborne AISA Eagle and satellite CHRIS/PROBA (C/P) nadir images has been developed within the ESA/PECS project “Spectral-spatial scaling from leaf to canopy level using spectro-directional approaches in support of the GMES Sentinel 2: ‘Superspectral’ mission“, carried out on the Institute of Systems Biology and Ecology (ISBE, Academy of Sciences of the Czech Republic). The approach, proposed for this purpose, is using the physical radiative transfer modelling to relate statistically a chlorophyll sensitive vegetation index to the leaf chlorophyll content.

Input data

AISA Eagle images
The airborne AISA Eagle hyperspectral image data of very high spatial resolution (0.4 m) were acquired during the HYPERTREES campaign in September 2006. Data were radiometrically, geometrically, and atmospherically corrected into the geo-coded top-of-canopy reflectance. The quality of radiometric and atmospheric corrections was verified by means of in-situ measured ground reflectance targets – seven natural and three artificial surfaces. Mean value of reflectance deviation from reference values at 551 nm was equal to 1.6% and to 1.8% at the wavelength of 850 nm, respectively. The example spectral profiles of two selected ground reference targets derived from the AISA image data and in-situ ground measurements are depicted in Figure 1. Pre-processed AISA Eagle images are currently being processed by the Cab retrieving algorithm to reveal the map of very high spatial resolution leaf Cab content. The retrieval method has been lately improved by means of new inversion approaches: (i) an artificial neural network (ANN), and (ii) an up-dated variant of chlorophyll estimating optical index called the Area under curve Normalized to maximal Chlorophyll absorption Band dept between 650-720 nm (ANCB650-720). This Cab map, once cross-validated against the ground truth, will serve as the reference for quality assessment of the chlorophyll estimates obtained from the satellite C/P image data.

CHRIS/PROBA (C/P) images

LiDAR measurement of forest structure

The ground-based LiDAR scanner Ilris-3D (Optech Inc., Canada), coupled with the digital camera Canon EOS 350D, was used to scan the open edges of two mature Norway spruce forest stands. The Ilris-3D device emits up to 2500 laser pulses per second at a wavelength of 1500 nm within a field of view of 40°x40°. The first pulse data of front standing trees were acquired from two scanning posts located about 60 m from the first forest stand and one scanning post located about 40 m from edge of the second forest stand. Finally, one set of scans of first and last pulse within the entire canopy hemisphere was recorded inside the first forest stand. An example of obtained LiDAR laser image data is presented in Figure 2.

The first chlorophyll retrieval from C/P image data was not aiming at highly accurate Cab prediction, but more on investigation of DART abilities to reproduce top-of-canopy bidirectional reflectance factor (BRF) of C/P images. The PROSPECT model, adjusted to simulate appropriately N. spruce needle optical properties, was coupled with DART in order to produce a database of numerous young spruce canopy bidirectional reflectance factors. This BRF simulated database is representing the basic heterogeneity of young spruce canopies captured in CHRIS/PROBA scene (i.e., existing combinations of canopy closure and canopy leaf area index) and various configurations of sun-target-sensor geometry (i.e., representative combinations of slope and aspect). Eight spectral bands of central wavelengths located in accordance with the CHRIS/PROBA bands were simulated within PROSPECT/DART. In total 1080 nadir images were simulated and per band BRF mean values were automatically extracted and stored in the BRF database.

The chlorophyll sensitive vegetation index ANCB670-720 (Area under curve Normalized to maximal Chlorophyll absorption Band depth between 670-720 nm) was computed for all DART BRF simulations and also pixel-wise from BRF values of selected C/P scene. All ANCB670-720 values were sorted into the six Aspect/Slope sub-classes (based on criterions of Table 1) and six functions expressing statistical dependency of ANCB670-720 on Cab were fitted per sub-class. The obtained statistical equations were applied on ANCB670-720 values computed from the C/P image within six Aspect/Slope sub-class masks, resulting in first CHRIS/PROBA map of leaf chlorophyll content estimation (see Figure 3 and 4).

Although, no independent ground truth cross-validation has been performed yet, the visual examination of the first C/P leaf chlorophyll map revealed realistic estimates of Cab values, taking place in range from about 20 to 120 mg cm-2. Comparison of DART simulated and C/P measured ANCB670-720 values (illustrated in Figure 5) is suggesting a general ability of DART to simulate sufficiently heterogeneity of young N. spruce canopy BRF as recorded in C/P image. The only case where dynamic range of C/P measured ANCB670-720 values was slightly larger than range simulated by DART is North aspect with slope from 0-15° (N/10). However, since it is the only one deviating Aspect/Slope class, it is possible that its discrepancy originates from misclassification of C/P image rather than from DART simulations.

Table 1. Definition of the terrain Aspect/Slope sub-classes used to build masks of young N. spruce forest stands from CHRIS/PROBA image.

Figure 1: Spectral properties of two selected ground reference targets: (a) clay surface, (b) artificial reference target coated by the Nextel paint. Spectra were derived from the fully pre-processed AISA image data of 0.4 m spatial resolution (solid line) and in-situ ground reflectance measured by the ASD FieldSpec-3 spectroradiometer (dashed line).

Figure 2: Example of the ground-based LiDAR data of a mature Norway spruce forest stand: (1) an open edge of observed N. spruce stand, (2) a quicklook of the forest edge LiDAR cloud point data, and (3) individual N. spruce trees separated from the forest stand.

Figure 3: Map of the leaf chlorophyll content retrieved for young N. spruce forest stand from CHRIS/PROBA nadir scene acquired on 12.09.2006. Two map detailed subsets, indicated by white rectangles, are displayed in following Figure 4.

Figure 4: Two subsets illustrating details of leaf chlorophyll content map obtained for young N. spruce forest stands from CHRIS/PROBA image acquired on 12.09.2006 (see previous Figure 3).

Figure 5: Aspect/Slope sub-class comparison of mean ANCB670-720 values computed from DART simulated images and CHRIS/PROBA (C/P) image acquired on 12.09.2006 (bars represents standard deviations of ANCB670-720 vegetation index).