The uncertainty of the calibration of the laboratory blackbody targets limits the absolute accuracy of the pre-launch infrared calibrations. Weinreb et al.
(1990) give the accurancy as 
0.35K. An additional uncertainty arises when the calibration of the laboratory target is transferred to the internal blackbody;
Weinreb et al. estimate this uncertainty (1990) to be 0.1K for the AVHRRs on NOAA-10 and NOAA-11, and possibly 0.2K
for that on NOAA-6.
The calibration scheme assumes a perfectly linear response of the detectors over the full dynamic range. Such a response is not the case, however, and for very precise measurements, it is necessary to include a correction for the non-linear response of the detectors. In reality, the indium antimonide (Channel 3) detector has a linear response over the oceanic temperature range, but the non-linearities of the Channel 4 and 5 detectors require correction. In the original scheme proposed by NOAA, this correction was achieved by assigning a small negative radiance to the space view. This is unphysical but provides an approximation for the non-linear effects in the signal at terrestrial temperatures. As an alter-native solution, NOAA proposed a look-up table derived from the pre-flight calibration measurements. The table provides the non-linear correction at a set of measurement temperatures corresponding to the detected radiances (the target temperature). NOAA distributed these tables for each instrument. A more precise scheme has been proposed by Brown, Brown and Evans (1985) who analyzed the pre-launch calibration data to generate a set of tables for the non-linearity correction for the AVHRRs on NOAA-7 and NOAA-9 as functions of the target temperature and the "baseplate" temperature, i.e., the blackbody calibration temperature measured by the PRTs. This is in principle possible for all of the AVHRRs. A similar analysis for the more recent AVHRRs has resulted in the non-linearity correction tables given by Weinreb et al. (1990). There are plans to replace such tables with a simple quadratic function explicitly relating the radiance to the detector output (Weinreb 1990).
The size of the non-linearity corrections is generally less than 1K, but can reach nearly 5K at the extremes of the measurement range. The residual uncertainties in the non-linearity correction is ~0.1K (Weinreb et al. 1991).
NOAA gives the "noise equivalent temperature differences" (NEDTs) of the detectors as 0.12K (e.g., Schwalb 1978; 1982), but there appears to be good evidence
that in reality the level is somewhat lower (
m channel, the random noise was far exceeded by a periodic noise, with amplitudes ranging from
~1.5K for NOAA-6 to ~0.3K NOAA-10 for the cases he considered. This channel of some AVHRRs has exhibited significant increases in noise level with time (e.g.,
Warren 1989). Because this noise is not random, it cannot be simply reduced by pixel averaging. Warren (1989) reports significant noise energy in the interval
of 2 to 20 pixels along the scan lines and has proposed interactive corrections for its reduction. However, in many cases, this noise has rendered the Channel
3 data useless for quantitative applications.
The calibration procedure relies heavily on the measured temperature of the blackbody calibration target being well-known. To this end, four PRTs are embedded in the target to monitor its temperature. Analysis of time series of these temperatures around an orbit reveals two problems (e.g., Brown, Brown, and Evans 1985).
The first is that the PRTs do not register the same temperature, indicating the presence of spatial gradients that give rise to differences greater than 2K between the sites of the individual thermometers. It is not clear, therefore, that a simple average of the four measurements, which is the NOAA recommended approach, gives the best estimate of the temperature of that part of the calibration target used in the calibration procedure. However, there is no basis from which to propose a more appropriate combination of the four measurements.
The second problem is that the blackbody temperatures are not constant around the orbit; they change by more than 4K. The change is not a simple function because it reflects the temperature variations the satellite experiences as the sun-angle changes around the orbit and as the sun is eclipsed by the earth. Furthermore, the geometry of the instrument is such that the blackbody has good radiometric coupling with the earth. As a result, it experiences the full range of terrestrial temperatures from the ice caps or high cloud tops to the hot deserts. Again, it is not clear that a simple average of the thermometer meas-urements is the most appropriate estimate of the true calibration temperature.
The consequences of these temperature changes on the absolute accuracy of the infrared measurements are difficult to quantify. The changes are likely to magnify
the uncertainty in the transference of the calibration of the laboratory blackbody targets to the internal blackbody target (given as 0.1K by Weinreb et al.
1990), as the pre-launch calibration measurements should have been taken in stable thermal conditions. It does highlight the need to continually monitor the
blackbody target temperatures during the periods for which the AVHRR infrared measurements are being used quantitatively.
The data indicate sub-pixel mis-registration between some of the channels of at least one of the AVHRRs: the Channel 3 data from the NOAA-7 AVHRR appear to be shifted by about one quarter of a pixel with respect to the corresponding Channel 4 data (Allam 1986). This shift is likely to be a consequence of the assembly of the optical components of that particular instrument. It is not clear whether other instruments of the series have similar problems.
Another facet of pixel integrity is whether the information in a particular pixel is independent of that of its neighbors. This aspect is common to all AVHRRs. It results from the scanning geometry and temporal response characteristics of the detectors and the associated electronics. The sampling rate of each channel is such that adjacent pixels in the along-scan view overlap by about 40%; this overlap is constant across the swath. The geometrical distortion of the pixel shape increases the overlap between adjacent pixels in successive scan lines, with the greatest increase occurring toward the edges of the swath. At scan angles of about 435, the overlap is about 40% (Breaker 1990). This overlap has consequences on certain applications of the data, such as the measurement of spatial gradients at scales of the pixel resolution.
The absolute accuracy of the infrared measurements, after a best effort to minimize the calibration uncertainties, is given as 0.2K (Brown, Brown, and Evans
1985). The uncertainties introduced at the time of the pre-launch calibration may contribute a further 0.4K, resulting in a more conservative estimate of absolute
accuracy of ~0.55K (Weinreb et al. 1990).
There is no provision for the in-flight calibration of the measurements from Channels 1 and 2; the pre-launch calibration must be assumed to hold over the
lifetime of the spacecraft. The pre-launch calibration is adequate for 5% at best (Abel 1990). The in-flight degradation of the visible channels (Channel
1) on the NOAA-6, -7, and -9 spacecraft were recently estimated using long time-series measurements over the Libyan Desert and were found to be 0, 3.5% and
6.0% per year, respectively (Staylor 1990).
These uncertainty estimates are summarized in Table 2. It should be possible to use AVHRR data of this accuracy at an ARM site. Such use assumes appropriate software and access to the high-resolution digital data, including all of the information required for the calibration of the infrared channels (i.e., the HRPT [High Resolution Picture Transmission] broadcast data, or the recorded LAC [Local Area Coverage] data). For some purposes, the analogue transmission data stream, the Automatic Picture Transmission (APT) data, may be processed to adequate accuracy, even approaching that of the HRPT data (Wannamaker 1984), but this would be at the cost of spatial resolution (~4 km in place of ~1 km) and spectral resolution (2 channels in place of 5). At present, it is not clear whether the APT broadcast will continue beyond the introduction of the AVHRR/3 in 1993 (see below).
