3.2. Results for AVHRR-Heritage Channel at 0.6 μm (Channel 1 Visible—VIS)
Figure 4 shows the inter-comparison of normalized reflectance factors for the AVHRR and AATSR sensors AVHRR against corresponding MODIS reflectance factors. For MERIS, we compare, in
Figure 4, with original radiances as the conversion into reflectances are not straight-forward (MERIS is a Spectroradiometer with different pixel characteristics along the conical scan) and was found to introduce additional uncertainties in the results. It is clear that conditions for inter-comparisons are quite different for the sensors being in the same orbital plane as MODIS-AQUA (AVHRR) compared to those being in a perpendicular orbital plane (AATSR, MERIS). For the latter, we largely miss results from MERIS during the Southern Hemisphere summer (showing large data gaps). Data was for this study only available from the descending overpasses facing the direction of the Sun. As the SNO point in the Southern Hemisphere occurs only in the ascending ENVISAT overpasses we were left exclusively with match-ups in the Northern Hemisphere. For AATSR, data for ascending ENVISAT overpasses was available (explaining better data coverage) but, also, here we experienced some data gaps for the Southern Hemisphere summer because of frequent occurrence of too high solar zenith angles for those SNOs (being excluded by initial quality control procedures). Thus, results for AATSR and MERIS visible channels are predominantly based on the SNO points in the Northern Hemisphere whereas for infrared channels all SNO points are used.
Reflectances generally agree well,
i.e., results are organized closely along the diagonal in the scatter plots. However, it is also clear that there is a rather large spread in reflectance quota results for the time series panels resulting in low precision (high variance) and an uncertainty in the estimation of potential systematic biases or trends. As we do not want to restrict the SNO time difference criterion too much (which would reduce the number of useful targets seriously), it was considered wiser to try to restrict the number of cases with very inhomogeneous targets. In addition, cases with very low reflectance factor values should be avoided since the quota calculations will be very sensitive to small and spurious deviations. Thus, the following restrictions were applied to data:
Standard deviations within individual match-up targets were maximized to 1%.
Solar zenith angles should be less than 70°.
Normalized reflectances should be larger than 10% (AVHRR, AATSR).
In the MERIS case the restriction was guided by conditions valid exclusively for the matched MODIS target reflectances.
Results for AATSR after applying these restrictions are shown in
Figure 5. It is clear that the systematic differences are much more evident now after screening out cases with large internal (target) variability.
Table 2 summarizes results with and without the described restrictions. A small underestimation (less than 2%) is seen for AVHRR and MERIS while AATSR shows 5% higher values. Results do not change significantly after applying restrictions which indicates a high degree of robustness.
3.8. Discussion of Results for all AVHRR Heritage Channels
Overall results are summarized in
Tables 8 and
10 for all studied sensors where the quota of reflectances or brightness temperatures is compared against MODIS.
For AVHRR visible (VIS) and near infrared (NIR) channels mean deviations are estimated to approximately 1.5%. A good agreement is expected since the used AVHRR calibration [
10] is already linked to MODIS. Thus, the current results can merely serve as a confirmation of a correct implementation of this calibration method. Our results also basically confirms the validity of the method of compensating for water vapor absorption effects in AVHRR channel 2 (see
Table 3) by use of additional MODIS channels [
9]. Now assuming a correct implementation and a close agreement with MODIS reflectances, these results may give an indication of the typical deviations that might be attributed to remaining uncertainties in the spatial and temporal matching of the SNO targets. In conclusion, they should most likely not exceed 1.5%. Although not directly comparable, this agrees rather well with the random uncertainties of 1%–2% estimated in [
11] and [
13] when inter-comparing radiances over Earth surface sites.
For AVHRR MWIR, LWIR1 and LWIR2 radiances deviations are very small,
i.e., less than 0.2% (
i.e., approximately 0.5 K at 250 K). However, in practice some systematic and significant deviations are seen for very cold temperatures, especially for channel 3b (see leftmost figure in lower panels of
Figure 8). Results for channel 3b indicates that, despite applying non-linear corrections in the calibration process, these corrections are not able to remove all systematic radiance biases. Remaining deviations here are very critical because of the poor radiometric resolution in this channel [
18]. This conclusion is further strengthened by the observation that deviations (although smaller) are seen also at very high temperatures in all three infrared channels. Thus, the problem of applying non-linear corrections is not only a problem for AVHRR channel 3b but also for the other two infrared channels.
For AATSR results in
Table 9, deviations against MODIS for VIS, NIR, and SWIR channels are slightly larger than for AVHRR. Generally, deviations range from 2.9% to 5.0% with the highest values for channel 2 at 665 nm and with higher reflectances for AATSR in both channels. Results agree reasonably well with those found in [
11] and [
13] stating an overall systematic bias of 2%–3% (with the same sign as here) and a random uncertainty of about 1%–2%. It is pointed out in [
11] and [
19] that a positive deviation could be expected here (at least over land surface sites) due to remaining (non-compensated) differences in spectral response functions between the MODIS and AATSR sensors. This should mainly come from differential absorption effects (e.g., ozone) and the existing changes with wavelength of the surface reflectance (which is most serious for heterogeneous targets). As the SNO approach used here leads to the use of quite heterogeneous targets in comparison to the reference sites used in [
11] and [
13], we suspect that corresponding deviations may be larger, which is also what our results suggest.
We also see that the largest scatter in our results occurs for AATSR channel 4 at 1.6 μm (see
Figure 7), which is an indication that conditions for inter-comparison are not as good as for the other channels. As the AATSR-MODIS inter-comparisons are made at high latitudes (± 72 degrees latitude) it is clear that many of the matched targets are selected over snow- and ice-covered surfaces. Thus, targets consist of a blend of poorly reflecting ice and snow surfaces in this wavelength (although with strongly anisotropic reflection behavior) and highly reflecting clouds (also reflecting strongly anisotropic). This explains the much larger scatter in the results for both low and high reflectances and we must consider the results obtained here as quite uncertain and less significant in comparison to results for the other channels.
For LWIR1 and LWIR2 AATSR channels the agreement with MODIS is exceptionally good for all channels. Here, it is clear that the small differences in spectral response for the involved channels are not influencing the results noticeably. Nevertheless, also for AATSR we see signs of deviations at very low and very high temperatures (e.g., at very low temperatures for 11 μm and at very high temperatures for 12 μm) indicating that some problems with non-linear radiance correction might still remain.
The evaluation of MERIS radiances (
Table 10) revealed the same magnitude of deviations as for AATSR but with different sign (
i.e., lower values than MODIS instead of higher). This is also in reasonable agreement with [
13] although the differences are slightly larger here.
Finally, an interesting reference to this study is [
20], where radiances measured by the Spinning Enhanced Visible and Infrared Imager (SEVIRI) were compared to MODIS and AVHRR radiances, thus using SEVIRI (considered to be a very stable sensor) as a transfer sensor for evaluating the difference between MODIS and AVHRR.
Table 11 shows the currently retrieved results and the results from the SEVIRI transfer study for the NOAA-18 AVHRR sensor. Notice that the SEVIRI study was performed for exactly the same period as studied here. We find a very good agreement for AVHRR channel 1, while deviations are a bit larger for AVHRR channel 2 (3.4% for the SEVIRI method
vs. 1.5% for the current method). The latter result is most probably an effect of the large differences in spectral response functions for this channel for the involved sensors. These are differences which are not easily compensated for, especially not when having to take into account three different spectral responses instead of two as for the direct inter-comparisons performed here.