Factors affecting the difference between data from SPOT-VEGETATION and METOP-AVHRR
Else Swinnen (VITO-TAP) December 2009
The SPOT-VEGETATION (VGT) sensor was specifically designed for vegetation monitoring purposes, whereas METOP-AVHRR is mainly intended for operational meteorological monitoring. The AVHRR sensor is exactly the same as on the NOAA-series. EUMETSAT provides the “Level1b” data which contain a number of merits compared to the “Level1b” of NOAA-AVHRR (EUMETSAT, 2008). These are: all the measurements are already converted to well-calibrated radiances, the lon/lat-planes provided allow precise geolocation of the imagery and thirdly the availability of 1km resolution data for the entire globe thanks to more advanced on-board storage facilities. For more information on the processing of the METOP-AVHRR data we refer to Eerens et al. (2009) and (more up-to-date) the Products section on the website http://www.metops10.vito.be .
This document describes the differences between the two satellites/sensors and their processing and it briefly discusses how these differences can affect the resulting data. It builds further on material presented in Eerens et al. (2002), Swinnen and Veroustraete (2008) and Eerens et al. (2009). Some important characteristics of both earth observation systems are listed in the table on the next page.
Orbital characteristics: Overpass time
Both METOP and SPOT were launched in a classical, circular, near-polar, sun-synchronous orbit at an altitude of around 820 km with an orbit inclination of around 99°. The overpass times are similar, and one hour earlier for METOP compared to SPOT. The overpass time determines surface illumination conditions and hence the observed radiances by the satellite. The difference of one hour in overpass time will affect the imagery to a very limited extent.
For both sensors, the orbital behaviour is controlled and neither of the satellites experiences significant orbital drift. Orbital drift is the process in which the satellite’s overpass time is gradually shifted to later stages, and is one of the major problems with the observations of NOAA-AVHRR. For this sensor, the imagery is indeed acquired progressively later in the day with increasing satellite age, causing a systematic change in illumination conditions.
The nominal (sub-nadir) resolution of both the VGT
and AVHRR sensors is approximately 1x1 km. With a swath width of more
than 2000 km, there is global coverage with a near-daily frequency at
the equator, to a daily frequency at higher latitudes. The
differences in scanning system and optics between AVHRR and VGT
affect the off-nadir spatial resolution of the imagery. The AVHRR
sensor is a cross-track scanner, which scans the Earth in a series of
lines, perpendicularly oriented to the direction of the orbit. Each
line is scanned from one side to the other, using a rotating mirror
placed in front of a sensor. The mirror sweeps with a constant
angular velocity, resulting in the same angular resolution for every
measurement. The sensor’s instantaneous field-of-view (IFOV)
remains the same, and when sweeping away from the nadir position, the
distance to the Earth increases and so does the ground surface
resolved by the satellite. Because of the large swath width, the
Earth’s curvature adds an additional panoramic distortion to
the off-nadir pixels. This leads to large off-nadir spatial
deformations and the “bow tie effect” (Meyer, 1996). The
VGT scanning system operates with an array of 1728 detectors
positioned in one row perpendicular to the satellite orbital track
simultaneously scanning one full scan line (also called pushbroom
scanner). There is a fixed geometrical relationship between the
detector elements in each scan line and the ground resolution
measured by the sensor, accounting for a large part of off-nadir
pixel deformation. Each individual
detector measures the energy for a single ground resolution cell and
thus the size and IFOV of the detectors determine the spatial
resolution of the system. The VGT data are
acquired essentially free of distortion up to 50° off-nadir if
the Earths curvature is not taken into account (SPOT-VGT User Guide).
Characteristics of the two considered systems, with focus on the mutual differences.
|Inclination angle (degrees)||98.7||98.7|
Equator crossing time (LST)
|Stability of the platform||No orbital drift||No orbital drift|
|Blue||Ch1: 0.43 - 0.47|
|Red||Ch2: 0.61 - 0.68||Ch1: 0.58 - 0.68|
|NIR||Ch3: 0.78 - 0.89||Ch2: 0.725 - 1.0|
|SWIR||Ch4: 1.58 - 1.75||Ch3a: 1.58 - 1.64 (day time)|
|MIR||Ch3b: 3.55 - 3.93 (night time)|
|TIR1||Ch4: 10.3 - 11.3|
|TIR2||Ch5: 11.5 - 12.5|
|Calibration of shortwave channels||Onboard calibration||Vicarious calibration a posteriori (Rao & Chen, 1999)|
Inputs for atmospheric correction
|Swath width (km)||2250||2400|
|Total scan angle (degrees)||101||110.8|
|Nominal resolution (km)||1.15||1.09|
Maximum off-nadir resolution
|Attitude of satellite||Known||Known|
|Spatial resolution after resampling (VITO processing)||1°/112 along great circle||1°/112 along great circle|
|Resampling method||Cubic convolution||Nearest neighbour|
Bi-directional reflectance distribution function (BRDF)
Surface reflectance varies with illumination and viewing geometry for anisotropic surfaces, like most of the natural surfaces. The BRDF describes this dependency for each surface type corresponding with a pixel. Correction for BRDF effects was considered by Cihlar et al. (1998, 2004) as one of the most important requirements for long-term and multi sensor time series analysis with satellite imagery.
One can assume that the BRDF-effect can be of a similar magnitude for VGT as for AVHRR (up to a viewing zenith angle of 45°), but not identical due to the differences in scanning system (especially for off-nadir conditions). The difference in overpass time could also contribute to a difference in BRDF, but given the small time difference this will be very limited.
Point spread function (PSF)
An important factor influencing the spatial resolution of satellite imagery is the PSF of the optical system. The PSF defines the propagation of radiation from a point source. It is the result of the physical, optical and electronic properties of the sensor and of the atmosphere at the time of image acquisition (Ruiz and Lopez, 2002). The PSF of VGT is narrower than the one of AVHRR. For an identical nominal spatial resolution, a broad PSF implies a larger area being sensed than the one implied by the IFOV. The result of a broader PSF is an increased auto-correlation or “smearing” between pixels of an image segment. Since the PSF of VGT is narrower than that of the AVHRR’s, the smearing in a VGT image segment is smaller than that of the AVHRR.
In spite of the higher spatial auto-correlation, the AVHRR images hahve a ‘sharper’ appearance. This is because the nearest neighbour (NN) resampling is used for projection, instead of cubic convolution (CC) for VGT. The use of NN does not introduce additional spatial-autocorrelation, limiting the effect of the difference in PSF between the sensors.
Geometric accuracy is another important issue to be considered when comparing time series of remote sensing data. Mis-registration, together with the previously discussed spatial characteristics, induces “blur” in an image time series (Meyer, 1996). For VGT, a high absolute, multi-temporal and multi-band registration accuracy is obtained using an elaborate database of ground control points (GCP’s). The absolute location accuracy is 330 m as root mean square error (RMSE) (Sylvander et al., 2000). For VGT2, the stars tracker onboard SPOT5 even allows for an accurate geometric modelling without any need for GCPs.
EUMETSAT provides very accurate lat/lon-planes for the data from the METOP-AVHRR sensor. This accuracy is achievable because the attitude of the satellite is known. This is in contrast with the AVHRR data of the NOAA satellites whose attitude is unknown Clearly, this puts a limit on the attainable accuracy of the geometric correction. Location accuracy in the case of NOAA-AVHRR data depends strongly on the processing algorithms used as opposed to VGT. A locational accuracy of 1 km is achievable for NOAA-AVHRR image registration (Rosborough et al., 1994). For METOP-AVHRR, the accuracy is higher and probably of the same magnitude as for SPOT-VGT.
Spectral response functions (SRF)
The Red and NIR spectral bands are of particular
interest, because they are used to derive the NDVI, a broadly used
vegetation index. The spectral response function (SRF) of the sensor
describes which part of the electromagnetic spectrum is measured.
Though similar, the SRF’s of VGT and AVHRR show different
shapes, central wavelength locations, bandwidths and mutual degrees
of overlap, especially with respect to the transition from the
chlorophyll absorption band to the foliage reflection band (0.68-0.72
μm) (Trishchenko et al.,
2002 – see figure below). This obviously leads to differences
in NDVI among the sensors for the same surface. In general, smaller
bandwidths lead to higher NDVI values. The VGT Red channel extends
beyond the 0.7 μm limit, which has a significant impact on the Red
reflectance as well as on the NDVI. Consequently, the effect of
varying SRF’s induces radiometric errors imposed on the natural
variability in land surface attributes. The differences in
SRF-definition of the Red and NIR channels between VGT and AVHRR are
probably the largest source of deviations between the data of both
Spectral response functions for the Red and NIR bands on board of VGT1, VGT2 and AVHRR.
Trishchenko (2009) analysed this difference and provided adjustment functions to reduce them. However, it might be better to accommodate for the difference by using biophysical parameters (e.g. fAPAR), extracted from the reflectance data. Most methods (e.g. CYCLOPES − Baret et al., 2007) take sensor characteristics like the SRF into account for fAPAR extraction.
The AVHRR sensor has two thermal infrared bands. Such information is not acquired by VGT. The availability of thermal data facilitates cloud detection and leads to potentially different applications.
The radiometric performance of a sensor in the visible and near-infrared region usually degrades in orbit (e.g. Gutman, 1999). The VGT sensor, unlike the AVHRR, has an onboard calibration device for these channels, allowing accurate radiometric calibration (Henry and Meygret, 2000). For AVHRR, the estimation of calibration coefficients is performed a posteriori. Sensor response is then vicariously determined with stable terrestrial targets whose radiances can be measured or inferred. For METOP-AVHRR, the same vicarious calibration method is used as for NOAA-AVHRR to guarantee similarity between data from the same sensor family.
Due to the different spectral band definitions,
also the influence of the atmospheric perturbations will be different
for VGT and AVHRR. The broad NIR spectral band of AVHRR will suffer
more from water vapour absorption, whereas that from VGT was
specifically designed to avoid the 0.935µm water vapour
absorption band. Van Leeuwen et
al. (2006) investigated the influence of Rayleigh scattering, water absorption,
ozone absorption, and aerosol optical thickness on the NDVI of
different sensors (AVHRR, MODIS and VIIRS), and they concluded that
the data continuity of NDVI across sensors would be largely enhanced
if adequate atmospheric corrections are applied, because different
sensors are differently affected by the atmosphere. Accurate
correction largely depends on the knowledge about the atmospheric
conditions at the time of image acquisition.
The VGT and AVHRR imagery are processed with the same method (SMAC - Rahman and Dedieu, 1994) using identical atmospheric inputs, except
for aerosol optical depth. Using the same method and atmospheric
inputs for the correction does not introduce additional difference
between the data sets.
The impacts of the differences in sensor characteristics between SPOT-VGT and
Metop-AVHRR were briefly discussed, but were not quantified. Users of
both datasets should be aware that reflectance values and derived
indices (e.g. NDVI) are sensor-specific.
Cihlar, J., J.M. Chen, Z. Li, F. Huang, R. Latifovic, and R. Dixon, 1998, Can interannual land surface signal be discerned in composite AVHRR data. J. Geophys. Res., vol. 103, no. D18, p. 23163-23172
Cihlar, J., R. Latifovic, J. Chen, A. Trishchenko, Y. Du, G. Fedojesevs, and B. Guidon, 2004, Systematic corrections of AVHRR image composites for temporal studies. Remote Sens. Environ., vol. 89, no. 2, p. 217-233
Eerens H., Piccard I., Royer A. and Orlandi S., 2004, Methodology of the MARS crop yield forecasting system. Vol. 3: Remote sensing information, data processing and analysis, Eds. Royer A and Genovese G, EUR 21291 EN/3, 76 p.
Eerens, H., Baruth, B., Bydekerke, L., Deronde, B., Dries, J. Goor, E., Heyns, W., Jacobs, T., Ooms, B. Piccard, I., Royer, A., Swinnen, E. Timmermans, A., Van Roey, T., Vereecken, J. and Y. Verheijen, 2009, Ten-daily Global Composites of METOP-AVHRR, Proceedings of the ISDE Conference, Beijing, in press.
EUMETSAT, 2008, AVHRR Level1b Products Guide, EUMETSAT, Darmstadt, Germany, Document EUM/OPS-EPS/MAN/04/0029, 123 p. (http://oiswww.eumetsat.org/WEBOPS/eps-pg/AVHRR/AVHRR-PG-0TOC.htm).
Gutman, G.G., 1999, On the use of long-term global data of land reflectances and vegetation indices derived from the advanced very high radiometer. J. Geophys. Res., vol. 104, no. D6, p. 6241-6255
Henry, P. and A. Meygret, 2000, Calibration of VEGETATION cameras on-board SPOT4. Proceedings of the VEGETATION-2000 conference, Belgirate, 3-6 April, p. 23-32
Meyer, D.J., 1996, Estimating the effective spatial resolution of an AVHRR time series. Int. J. Remote Sensing, vol. 17, no. 15, p. 2971-2980
Rahman, R. and G. Dedieu, 1994, SMAC: a simplified method for the atmospheric correction of satellite measurements in the solar spectrum. Int. J. Remote Sensing, vol. 15, no. 1, p. 123-143
Rao, C.R.N. and J. Chen, 1999, Revised post-launch calibration of the visible and near-infrared channels of the Advanced Very High Resolution Radiometer (AVHRR) on the NOAA-14 spacecraft. Int. J. Remote Sensing, vol. 20, no. 18, p. 3485-3491
Rosborough, G.W., D.G. Baldwin, and W.J. Emery, 1994, Precise AVHRR image navigation. IEEE Trans. Geosci. and Remote Sens. Vol. 32, no. 3, p. 644-657
Ruiz, C.P. and F.J.A. Lopez, 2002, Restoring SPOT images using PSF-derived deconvolution filters. Int. J. Remote Sensing, vol. 23, no. 12, p. 2379-2391
Swinnen, E., F. Veroustraete , 2008, Extending the SPOT-VEGETATION NDVI time series (1998-2006) back in time with NOAA-AVHRR data (1985-1998) for Southern Africa, IEEE Transactions on Geoscience and Remote Sensing - Vol. 46 (2): 558-572
Sylvander, S., P. Henry, C. Bastien-Thiry, F. Meunier, and D. Fuster, 2000, VEGETATION geometrical image quality. In: Proceedings of the VEGETATION-2000 conference, Lago Maggiore, Italy, p. 33-44
Trishchenko, A.P., J. Cihlar, and Z. Li, 2002, Effects of spectral response function on surface reflectance and NDVI measured with moderate resolution satellite sensors. Remote Sens. Environ., vol. 81, no. 1, p. 1-18
Trishchenko, 2009, Effects of spectral response function on surface reflectance and NDVI measured with moderate resolution satellite sensors: Extension to AVHRR NOAA-17, 18 and METOP-A, Remote Sensing of Environment, vol. 113, No. 2, p. 335-341
van Leeuwen, W.J.D., B.J. Orr, S.E. March and S.M. Herrmann, 2006, Multi-sensor NDVI data continuity: uncertainties and implications for vegetation monitoring applications. Remote Sens. Environ., vol. 100, p. 67-81