AIRS Cold Bias Study

Figure Viewer

1 Purpose, Context and Scope.

To investigate the characteristics of the radiometric bias of AIRS at low temperatures.

It has been reported elsewhere <TBD> that AIRS scenes with deep convective clouds (DCC) drifted by as much as ~5K from launch until before 2010. Tom Pagano (JPL) has proferred the theory that the drift in DCC’s is due to a widening of the FOV over time that stopped in 2010, and is only important when you have a DCC in the FOV under interest, and very warm (tropical) FOVs surrounding it, which makes the DCC scene look hotter since it is seeing some of the hot scenes surrounding it.

LLS has reported that the for DCCs scenes versus the drift [bias] in AIRS went down very quickly with latitude.

The SNOs between AIRS and CrIS or AIRS and IASI, are available for detailed bias study, they are however distributed heavily to the high latitudes (AIRS:IASI exclusively near 75 deg) and are available from 2007 for IASI and 2012 for CrIS, so would not capture early AIRS mission drifts.

One of the biggest challenges in such analysis as this is to be certain that artifacts of the samples are distinguished from those of the sensors, even with SNOs are that are so closely matched in space and time.

2 Methods and Data.

The parent script is: /home/chepplew/projects/sno/airs_cris/ac_sno_trend_subset_year.m which uses child scripts: /home/cheppelw/gitLib/asl_sno/source/load_sno_airs_cris_asl_mat.m and stats_sno_airs_cris_asl_mat.m in same directory.

The data are in the usual locations: /home/chepplew/data/sno/airs_cris/ASL/...

For this report the low resolution (NSR) CrIS CCAST or the MtOP-A IASI observations are matched to the AIRS L1C as SNOs. Most days are available but the NSR CCAST was mostly lost when the lustre server failed in 2018, so gaps can not currently be recovered.

In brief, the SNO data are loaded into memory and various subsets are constructed and from which bias versus scene temperature can be determined and the population distributions analysed. Subsets include: day, night (determined by solar zenith angle); north hemishpere, south hemisphere, tropical, and those with positive and negative delay times of the AIRS and CrIS observations in each SNO pair. This last pair was used in light of the hot bias that occurs for scenes over land during the day, which was shown to be due in part by the mean time difference of the observations between the two sensors and the larger effective FOV of AIRS compared to CrIS.

For the purposes of this study window channels near 900, 1231, 2456 cm-1 are examined closely.

Full statistical data are saved in files at: ~chepplew/data/sno/airs_cris/stats/ac_sno_2012-18_trend_subset_year_{lw,mw,sw}*.mat

3 Results.

3.1 AIRS:CrIS 2012 to 2018

The SNPP CrIS mission overlaps AIRS from midway through 2012 to the present.

3.1.1 Population distribution count

The population counts (PDFs) for select window channels are shown in three figures (1,2 and 3) Fig.1 Fig.2 Fig.3 for various subsets. For the window channels, the main differences from day to night are the absence of the hot scenes > 301 K and the diminution of the 299 K peak, some increase in the cold tail, more espcially in the SW channel. The tropical subset is mostly of warm oceans and some cold clouds but DCCs are not particularly evident. There is some variability in total SNO samples between the years, 2012 and 2018 are partial years, due to availabilty of the CrIS CCAST NSR L1C granules.

3.1.2 Bias vs scene

The variation of the AIRS:CRIS bias with scene brightness temperature for year averages for the three window channels are shown in three figures (4,5,6 and 7). Figure 4 Fig.4 illustrates the detail for one year, with three subsets of the SNOs, to help interpret the following figures. The curve exhibits the characteristic shape I have shown on previous occasions with a similar analysis.

In each of the next figures, the mean bias is shown in the top pane as a colored line, and the standard error of the sample is shown as the black dotted. The lower panel shows the population in each scene bin for each year average.

The LW window Fig.5 has the characteristic down-turn smile shape - AIRS colder than CrIS in the hot scenes. There is little significant change between day and night.

The MW window Fig.6 channel is similar to the LW window channel with some indication of increasing bias at cold scenes.

The SW window Fig.7 channel has a much greater down-turn smile with AIRS colder than CrIS for hot and for cold scenes. This is investigated further (see below).

3.1.3 Temporal bias trend

To determine how stable is the bias between AIRS and CrIS for cold scenes, a reasonably well populated cold scene bin is selected for the given channel and the bias plotted for each year average in the sequence. These are shown in the next three figures 8,9 and 10.

The LW window channel at 899 cm-1 bias versus time is plotted in figure 8 Fig.8 with the three subsets for comparison. An estimate fo the standard error of the mean can be read directly from the previous series of plots, which for this channel is approximately 0.03 K. There is therefore no statistically significant trend of the bias at 899 cm-1.

The MW and SW window channel stability is shown in the next two figures. Fig.9 Fig.10 There is some indication of a variation over time outside the margins of the standard error. It is interesting to note that the result is consistent between the night and day subsets, even in the SW channel.

3.2 AIRS:IASI 2007 to 2018

The MetOp-A IASI mission data overlaps AIRS from May 2007 to the present. Since the SNOs occur only in a narrow latitude band near 72-deg north or south, there are no tropical samples. Subsets are compiled for the two hemispheres, for day or night and for samples with positive or negative delay time between the measurement made by AIRS and IASI for each SNO.

3.2.1 Special note on IASI noise

The IASI spectral radiance measurements are very noisy in the SW region, with an effective NEDT near 2400 to 2600 cm-1 of about 2 K to 6 K respectively for a 250 K black body. This has important consequences for finding the lowest temperature scenes as any individual observation could have a negative radiance - which is meaningless. In other words when searching for scenes that are at 215 K for example, IASI may actually be viewing a scene anywhere from 209 to 221 K. The lowest detectable scene temperature varies with wavenumber in accordance with the NEDT for that channel. This makes comparison with AIRS invalid for those low temperatures.

An example of the IASI radiance observations for the channel at 2456.50 cm-1 for all SNOs in 2017 is shown in figures 11 and 12. In figure 11 Fig.11 the raw radiance samples are shown (left panel), the empirical CDF (center panel) and their histogram (right panel) showing negative radiances and sampling discretization. Computation of statistics must be done with radiance and not brightness temperature. In figure 12 Fig.12 , the empirical CDF for the 2017 SNOs for AIRS, IASI and the translated IASItoAIRS (I2A) are shown together. For 2456.5 cm-1 channel radiance of 0.05 (0.01) corresponds to BT of 234 (212) K, resp. This demonstrates that inter-comparison is meaningless at BT cooler than about 205 K, because of the IASI signal.

Another illustration of the problem of differencing the two sensors with different signal characteristics near the zero radiance point is shown in figure 13. Fig.13

3.2.2 Population distribution counts

The population counts (PDFs) for each selected window channels are shown in three figures (14, 15 and 16) Fig.14 Fig.15 Fig.16 for various subsets. These distributions are different from the AIRS:CRIS SNOs mainly due to the absence of any hot scenes and are biased cooler. There is a notable peak near the melting point of ice. The nighttime subset are cooler than the day as expected. Also there tends to be more warm scenes in the north hemisphere than the south and more cold scenes inthe south hemisphere than the north, as expected. Notice that for the 2456 cm-1 channel IASI PDFs are highly impacted by the -ve going signal.

3.2.3 Bias vs Scene

Figures 17, 18, 19 and 20 show the bias between AIRS and IASI as a function of scene BT for varisou subsets for the three window channels. Figure 17 Fig.17 is the average bias for a single year for three subsets provided for simplicity and aid to interpreting the following three figures. You can easily see the separation of day from night within the full sample set.

Figure 18 Fig.18 for 899 cm-1 window channel, shows a quite closely similar annual bias sets without much variation with scene BT, within the bounds of the sampling range.

Figure 19 Fig.19 for 1231 channel, shows quite closely similar annual bias sets, with the day subset not varying much with scene BT the the night subset exhibiting slight positive trend with BT. The day versus night subset should not be directly affected by short wave radiation and solar input, however there may be an indirect affect or artifact of sampling over different scene types. This could be investigated further.

Figure 20 Fig.20 for 2456 cm-1, shows quite closely similar annual bias sets, with every subset showing a large divergence (AIRS warmer than IASI) at temperatures cooler than about 230 K. The earlier analysis indicated that intercomparison is not possible for scenes at 205 K, so given some margin to 230 K it is possible that there is an increasing bias between the sensors with cooler scenes, as there is some idication of this from 270 K down to 240 K as shown in these figures.

An investigation into whether there are any other sampling artifacts that could significantly contribute to these SW bias differences has been undertaken. Any dependencies with respect to delay time between samples in each SNO pair, dependencies on scene type (Greenland versus Antartica, for example) were not discovered. Therefore most of the large offset at very low temperatures is an artifact of the hight IASI noise and -ve going radiance.

3.2.4 Temporal Bias Trend

Figures 21, 22 and 23 Fig.21 Fig.22 Fig.23 show the year average bias for the three channels and the three subsets. There is no significant trend outside of the uncertainty over the period 2007 to end 2018. Mean bias (AIRS minus IASI) for 899, 1231 and 2456 cm-1 remain close to 0.1 K 0.05 K, 0.15 K respecitvely for this temeprature bin.

4 Discussion, Conclusion, Further Work

Overall, and in line with previous non-subset SNO bias calculations, the AIRS tends to run a little cold relative to NPP-CrIS and a little warm relative to MetOp-A IASI.

For the window channels examined, the AIRS:CrIS SNOs generally show a non-linear dependence on scene temperature, which is comparable to the AIRS:IASI dependence. Care must be taken to limit the temperature of cold scenes that are used to compare for the IASI, being no colder than about 220 K for the channel at 2456.5 cm-1. From the AIRS:CRIS SNO at 2456.5 cm-1 AIRS appears to be colder than CrIS for these cold scenes but warmer than IASI for the equivalent.

There is no compelling evidence for any drift between AIRS and CrIS or AIRS and IASI for the periods studied for these channels.

It is suggested to intercompare CrIS and IASI to check the cold SW non-linearity and use global random samples to investigate cold bias trends due to sampling tropical deep convective clouds. For example, it has been shown by other groups that there is a strong diurnal variation in the coldest DCCs (<210 K) compared to slightly warmer DCCS (< 230 K) espcially near ocean/land boundaries.

5 References

6 Appendix.

6.1 Analysed data used for plots:

6.1.1 AIRS:CRIS SNO in: /home/chepplew/data/sno/airs_cris/stats/

sub.un = ':'; % no subsetting
sub.nh = find(s.aLat(s.ig) >= 40); = find(s.aLat(s.ig) <= -40); = find(s.asolz(s.ig) < 90);
sub.nit = find(s.asolz(s.ig) > 90);
sub.pdt = find(s.tdiff(s.ig) >= 0);
sub.ndt = find(s.tdiff(s.ig) < 0);
sub.trn = find(s.aLat(s.ig) < 40 & s.aLat(s.ig) > -40 & s.asolz(s.ig) > 90);

CrIS NSR channels: [358:713]
CrIS NSR channels: [714:933]
CrIS NSR channels: [1147:1305]

6.1.2 AIRS:IASI SNO in /home/chepplew/data/sno/airs_iasi/stats/

sub.un = ':'; % no subsetting
sub.nh = find(s.aLat(s.ig) >= 40); = find(s.aLat(s.ig) <= -40); = find(s.asolz(s.ig) < 90);
sub.nit = find(s.asolz(s.ig) > 90);
sub.pdt = find(s.tdiff(s.ig) >= 0);
sub.ndt = find(s.tdiff(s.ig) < 0);

AIRS channels: [512:1269]
AIRS channels: [1270:1870]
AIRS channels: [2163:2645]

6.2 More on low signal bias sampling and IASI noise.

Further to the discussion above on the impact of the IASI noise in the SW band, it is instructive to demonstrate the low-level signal binnning at the longer wave channels and the symmetry, or lack thereof, when binning according to what the other sensors sees.

In figure A1 Fig.A1 for channel at 1231 cm-1, the observations for the SNOs from AIRS and for IASI are shown for three radiance bins corresponding to narrow bins centered at 204, 220, 247 K. The bias between the sensors is computed using these samples. Notice that the distributions for each sensor are similar in each bin and they are reasonably Guassian. Note that samples for each bin are found as follows: find the SNOs when AIRS measurements are in the narrow range centered on the bin; then find the SNOs when IASI measurements are in the same narrow range; Combine these two sets of indexes without duplication; then perform the statistics.

It is instructive to look more closely and see the distributions for each sensor using the indexes found from the other. This will reveal significant discrepencies between the SNO pairs. This is shown in figure A2 Fig.A2 for channel at 1231 cm-1 and a warm bin near 247 K and cold bin 204 corresponding to those shown in the preceeding figure. In this case the AIRS samples are shown using the indexes of the scenes found using IASI as the reference (blue curves) compared to IASI samples for those found using AIRS (green curves). Notice how close they are to each other and reasonably Guassian. The bias determined from these samples has a low uncertainty as shown in figures of bias vs scene.

When looking at the channel at 2456.5 cm-1 the distributions become highly non-normal for the cold scenes and the two sets of observations in each bin are discrepent for the cold scenes. This is shown in figures A3 and A4, being the counterparts of the preceeding but for the SW channel. Fig.A3 and Fig.A4 This clearly shows that signals near zero can not be used for reliable statistics. Also shown in Fig.A4 are the samples in the AIRS only bin and IASI only bin showing the width of each bin. Although IASI radiances are allowed to be negative in the L1C data stream the AIRS are not. Sampling near zero radiance becomes skewed for both sensors for different but related reasons.

comments powered by Disqus