Example 2.

Introduction

The second example concerns atmospheric correction of a series of images collected by the Compact High Resolution Imaging Spectrometer (CHRIS) on board the ESA PROBA satellite. CHRIS is a programmable sensor, able to operate in a number of different modes, one of which provides data in many contiguous spectral bands (hyperspectral mode) and others in which the sensor collects data in a number of separate narrow bands (superspectral modes). The CHRIS sensor was designed by Sira Technology and draws upon their earlier work on the MERIS instrument on board ENVISAT. The increased spectral capability of CHRIS (and MERIS) compared with the Landsat ETM+ raises the possibility of correcting for the effect of the atmosphere directly. Using the hyperspectral modes it is possible to measure accurately the shape and depth of absorption features due to water vapour in the atmosphere, for example. Even the superspectral modes provide a significant increase in capability, because the bands sensed are sufficiently narrow that they may be located in regions of maximum absorption, or at wavelengths known to be associated with atmospheric features.

Mode	No. of bands	Nadir pixel size (m)	Nominal swath width (km)	Band prefix
1	62	34	25	A
2	18	17	25	W
3	18	17	25	L
4	18	17	25	C
5	37	17	12.5	H

Table 1
The five main modes of operation of the CHRIS sensor. Mode 1 is hyperspectral, the others are superspectral.
(source : Cutter, 2005)

The PROBA satellite also provides an important additional capability compared with most satellite platforms. It is designed to be highly manoeverable and to have a high degree of autonomy, meaning that the CHRIS sensor may be targeted precisely on the same area of ground from up to five different viewing positions during a single overpass. This means it is possible to use the combination of CHRIS and PROBA to acquire multi-directional hyperspectral data from space. In theory, the multi-angle capability of PROBA will also provide useful additional information for atmospheric correction, as the different images are acquired through different path lengths of atmosphere. The potential of this approach has been demonstrated with other multi-angle satellite sensors, such as ATSR-2.

The PROBA-CHRIS data set

The CHRIS data were acquired over Chichester Harbour on 7th October 2004 (Figure 2). The weather conditions on this day were excellent, with very little cloud or haze and a total of five images of the site were obtained. The nominal view zenith angles of these images are +55°, +36°, nadir, -36° and -55°, where positive angles indicate that the sensor was pointing forward, along the track of the satellite. Negative view zenith angles indicate that the sensor was looking back along the satellite track. The satellite was travelling from north to south (top to bottom in Figure 2).

Figure 2
The CHRIS images of Chichester Harbour. Data have been provided by the European Space Agency, using the ESA PROBA platform and the SIRA Technology Ltd CHRIS instrument, developed with support from BNSC. © Sira Technology / ESA 2004.

The first image to be collected was that at a nominal view zenith angle of 55° (lower-left in Figure 2), and this was acquired when the satellite was well to the north of the site. Then, a short time later, the +36° view zenith angle image was collected (top-left in Figure 2). The nadir image was then acquired over the site (central image in Figure 2). Finally, the -36° and -55° view zenith angle images were collected (top-right and lower-right respectively in Figure 2).

The PROBA platform is also able to point the CHRIS sensor in the across-track direction, meaning that the sequence of five images may be collected from a orbit which does not pass directly over the target site. Thus, the actual view zenith angles will most likely differ from the nominal angles given above. This is illustrated in Table 2 which gives the nominal and actual view zenith angles for the Chichester Harbour images. The view azimuth angles are similarly affected, so before analysing a PROBA-CHRIS image it is important to inspect the header file which contains information on the actual view zenith and azimuth angles of each image. This important aspect of the PROBA CHRIS system is described further by Cutter (2005), and this paper also describes a number of other factors which must be taken into account if precise information on the acquisition geometry of each image is required. Another way to visualise the actual viewing geometry of the CHRIS images from Chichester Harbour is to plot them on a polar co-ordinate system (Figure 3).

Image designation	Nominal view zenith angle	Actual view zenith angle	Actual view azimuth angle (clockwise from north)
A2_41	+55	+48.48	14.44
A0_41	+36	+27.12	15.08
9F_41	nadir	+3.46	186.05
A1_41	-36	-32.59	193.67
A3_41	-55	-52.04	194.17

Table 2
Comparison between the nominal view zenith angles and the actual view zenith and view azimuth angles of the Chichester Harbour images.
(source : CHRIS HDF file)

 

Figure 3
Polar plot of the viewing geometry of the CHRIS data set collected from Chichester Harbour on 7th October 2004. The zenith angle is measured radially from the centre and azimuth angle relative to grid north is measured clockwise from the top. The position of the sun is shown by the symbol.

 

Atmospheric correction procedure

Initially, it was hoped that the data could be corrected using a procedure developed specifically for PROBA CHRIS by Luis Guanter of the University of Valencia. Guanter's method combines data from generic targets on the ground (vegetation, bare surfaces etc.) with a radiative transfer model, and achieves good atmospheric correction without the need for simultaneous spectral ground data. However, several factors specific to the image acquisition prevented this. Firstly, the signal levels were relatively low, a consequence of the low Sun angle at the latitude of the UK in autumn. This problem was particularly acute for images collected from 36° and 55° view angles. Secondly, Guanter's method was developed for use with CHRIS mode 1 (hyperspectral) data in which detailed spectral data are available from the regions of water absorption around 820 nm and 940 nm. CHRIS mode 3 has a number of bands in this region but they are not optimally located for the measurement of atmospheric water content (see Table 3).

Figure 4
CHRIS mode 3 bands in relation to a typical vegetation spectrum.
(source : Luis Guanter, 2005).

Band	Min	Max	Centre	Bandwidth
L1	438	447	442	9
L2	486	495	490	9
L3	526	534	530	9
L4	546	556	551	10
L5	566	573	570	8
L6	627	636	631	9
L7	656	666	661	11
L8	666	677	672	11
L9	694	700	697	6
L10	700	706	703	6
L11	706	712	709	6
L12	738	745	742	7
L13	745	752	748	7
L14	773	788	781	15
L15	863	881	872	18
L16	891	900	895	10
L17	900	910	905	10
L18	1002	1035	1019	33

Table 3
The nominal CHRIS mode 3 bands. The actual bands sensed are a function of the sensor temperature, and are recorded in the HDF header file.

Figure 5.
CHRIS mode 3 bands in relation to typical atmospheric transmission spectrum.
(source : Luis Guanter, 2005)

Luis Guanter kindly agreed to investigate alternative approaches, and the description which follows is based on his work on the Chichester Harbour data set.

Guanter decided to use the MODTRAN-4 radiative transfer model rather than 6S, as this is more reliable at low signal levels and extreme off-nadir view and illumination angles. In the absence of any spectral ground data or information on the atmosphere at the time of acquisition, the atmospheric parameters were estimated empirically using data from the nadir view image, as follows:

Estimating the aerosol optical thickness (AOT)
This was done using the DN values of deep water pixels. AOT was varied until a value was found which minimised the values in the visible bands but did not lead to negative values in the near infra-red bands. This is based on the assumption that the water-leaving radiance in the near infra-red region is zero.

Estimating the water vapour column content
The assumption was made that ground surface reflectance was linear across the wavelength region sensed by CHRIS bands 14 to 17 (780 to 900 nm). Water vapour column content was increased until the reflectance spectra from typical surfaces no longer showed spikes and irregularities in the region around the water absorption bands.

Results

Figure 6 shows a comparison between a simulated true colour image created from the raw data and one created from the atmospherically corrected bands. Visually, there is little difference between the two, but the advantage of the atmospherically corrected data becomes clear when the values of individual pixels are plotted.

  

Figure 6..
Simulated true colour images from the original data (left) and that after atmospheric correction (right). CHRIS mode 3 bands 8, 4, 2 (R, G, B).

Figure 7 shows the data from a pixel in the middle of a uniform grass field (left image), comparing the data from the uncorrected image (top graph) with those from the atmospherically corrected image (lower graph). The characteristic shape of the green vegetation spectrum is clear from the corrected data, unlike the raw data. The correction is not perfect, however, as there is a residual spike in the reflectance spectrum in CHRIS band 16 (895 nm).

Figure 7
The result of the atmospheric correction for a pixel from a grass field.

Figure 8 shows a spectrum from a large area of bare concrete (left image), before and after the atmospheric correction has been applied. Again, a slight irregularity is evident in CHRIS band 18, and there is also a slight peak in CHRIS band 4 (551 nm) which would not be expected.

Figure 8
The result of the atmospheric correction for a pixel from an area of bare concrete.

Figure 9
The result of the atmospheric correction for a pixel from an area of water in the main channel.
(n.b. y-axis scale expanded compared with previous graphs)

Finally, Figure 9 shows a similar comparison for a water pixel. The peak in green wavelengths in this case could relate to suspended sediment in the water column. The reflectance values from the water pixel are very low, and show that any further reduction in the values in CHRIS band 16 to reduce the spike evident in the vegetation spectra would result in negative values in the water areas.

 

References

Guanter, L., Alonso, L. and Moreno, J. (2004) Atmospheric correction of CHRIS/PROBA data acquired in the SPARC campaign. Proceedings of the 2nd CHRIS/Proba Workshop, EAA/ESRIN, Frascati, Italy, 28-30 April (ESA SP-578, July 2004).

Guanter, L. et al. (in press) IEEE Transactions on Geoscience and Remote Sensing. [details to follow shortly].

Concept map

URLs (checked 1 Aug 2005)

PROBA CHRIS website : http://earth.esa.int/missions/thirdpartymission/proba.html.

Cutter, M.A. (2005) CHRIS data format. Version 4.2. http://earth.esa.int/pub/ESA_DOC/Chris_Data_Format_Issue4-2.pdf

Davidson, M. and Vuilleumer, P. (2005) Note on CHRIS acquisition procedure and image geometry. http://earth.esa.int/pub/ESA_DOC/CHRIS_acquisition-procedure_image-geometry_rev1_3.pdf

Proceedings of the Second CHRIS PROBA Workshop (2004).

Proceedings of the Third CHRIS PROBA Workshop (2005).

---

This work is licenced under a Creative Commons Licence.

© NCAVEO, 2005
Network for Calibration and Validation of Earth Observation data
School of Geography, University of Southampton
Southampton SO17 1BJ, UK