The AIRS Level 1 Steering Committee made examination of any AIRS radiometric calibration trends the present priority for Level 1 activities during our April 2019 meeting. This document formalizes this recommendation and discusses the context for the importance of quantifying and if possible, fixing, calibration trends. This task will require a very significant effort given the large number of AIRS channels which have a variety of trend characteristics. We suggest that successful completion of this task will require planning that is properly timed to minimize the amount of data generation and computation needed for validation of trend measurements and/or improvements in trends. Moreover, these efforts should be cognizant that the AIRS climate record will be continued with the NOAA CrIS sensors, requiring careful cross-calibration between sensors.

AIRS will soon be operating for 18 years, a long enough period for significant contributions to climate change. The AIRS climate record will be continued with the NOAA CrIS sensor for the foreseeable future, and together they could provide the best measurement of atmospheric temperature, humidity, surface temperature, and cloud forcing trends available. The climate community, and the public, require rigorous measurement standards, although individual papers in the literature do not always adhere to these standards. There is a long history of satellite climate sensor calibration controversies (see cite:ttrends06 and cite:thorne2005) that use several microwave sensor channels to monitor tropospheric temperature. These highlight the significant amount of effort it takes to create climate-level trends, using only two channels without any true geophysical retrieval! Controversies regarding the analysis of these data sets continue to this day, and have even been discussed in Congressional Committees.

Hyperspectral infrared satellite radiances have several advantages compared to microwave (high vertical resolution, better water vapor sensitivity) with the disadvantage of high sensitivity to clouds. On-going work (ROSES grants) are showing that the cloud issues can be handled. With the advent of well-controlled orbits starting with the EOS satellites (for NASA) and for the JPSS (NOAA) satellites, some of the more difficult aspects of converting satellite (microwave) radiances to climate quality radiances has been alleviated. This problem has always produced uncertainty in the trends using any single satellite product.

These orbit stability improvements are highlighted in a recent Science Advances paper (cite:zou2018) that examines AQUA AMSU versus SNPP ATMS. The purpose of this paper is to prove that AMSU and ATMS are radiometrically stable because they can be inter-compared (stable orbits) and their radiance anomalies agree to ~±0.02K over six years. (AIRS can likely determine stability w/o comparisons to other sensors.)

None of the papers referenced so far are concerned with absolute radiometric calibration. They acknowledge that satellites (microwave) will have 0.2-0.3K absolute calibration differences, but given satellite overlaps the only issue addressed in detail are how to best offset one sensor to the other sensor's radiometry, which includes issues of sensor non-linearity that can make these offsets scene dependent.

The stability of the AIRS and CrIS blackbody calibrators suggests that even better climate records could be generated using the hyperspectral infrared, with increased vertical resolution and enhanced sensitivity to water vapor. For example, small amounts stratospheric sensitivity in the microwave mid-tropospheric channels has led to claims that the Earth's atmosphere is warming slower than the surface or as measured by other techniques. This is understood by the majority of the science community, but a sensor that can unequivocally separate tropospheric from stratospheric temperature trends would clearly remove the remaining nagging doubts about the level of climate change in the atmosphere.

We believe that the above discussion puts in context that it is absolutely imperative for AIRS calibration trends to be (a) rigorously measured and (b) fixed if possible, either using physics or empirical corrections. Empirical corrections are acceptable if the measurements that provide these corrections are rigorous with associated error estimates.

We may need to flow these calibration stability estimates through Level 2. A recent paper (cite:susskind2019) used AIRS Level 3 surface temperature retrievals trends to "validate" the GISSTEMP surface temperature product that is widely used by the media to communicate the level of climate change to the public based on scientific observations. This paper makes absolutely no reference to AIRS uncertainties (other than sampling) and does not contain any error estimates. I'm not quite sure why reviewers don't care about climate observation uncertainties. The AIRS Project needs to work towards making these error estimates widely available so they can be used by others, hence the need for prioritizing AIRS radiometric trends studies.

Detection, and quantification of AIRS Level 1 radiometric trends is a complex task, primarily because AIRS radiometric stability is already known to be very good, at least to the 0.01-0.02K/year level. This is based on comparisons to good SST climatologies, CO₂ and other minor gas trends, and to other instruments, primarily IASI. If one takes the nominal rate of climate change as 0.01K/year, AIRS can only make a strong case for climate trends if we can either reduce our uncertainty in AIRS stability well below 0.01K/year or somehow "correct" any AIRS calibration trends using other data sets. Given that state of other climate trend measurements, if we could establish AIRS trend uncertainties to the 0.003K/year or better, AIRS would represent one of the best (but short term) available climate trend measurements

A characteristic of AIRS is that no channel is sensitive to only one geophysical variable. For example, all window channels have from 0.3K (shortwave only) to 7K (800 cm^-1) of surface temperature darkening due to water vapor. This makes validation with truth quite complicated and I (LLS) contend that some sort of retrieval (of B(T) anomalies) is the only way to compare to other climatologies that might be as accurate, or more accurate, than AIRS. In addition, we cannot determine "AIRS" trends without looking at all "good" channels to the degree possible, since a retrieval by nature must use at least several hundred? channels. There are two main categories of trends we should explore: (a) relative trends among channels where geophysical differences have been removed, and (b) absolute trends versus "truth". We can possibly generate (b) by modifying observed B(T) trends with known geophysical trends (minor gases like CO₂, N₂O, and possibly CH₄) if we can demonstrate that these modifications are valid (with error bars).

I (LLS) have developed a methodology that retrieves geophysical trends from clear ocean scene radiances. We have recently shown that these retrievals almost perfectly return the in-situ CO₂ trends. Unfortunately, that only validates the lack of trends for a complicated set of channels that are sensitive to CO₂. However, these same retrievals (which include SST) can be used to inter-compare to SST climatologies. Since these climatologies (OISST, OSTIA, and ERSSTv5) can vary by up to 0.01K/year in trends, very careful inter-comparisons are needed. This is non-trivial, since the AIRS clear ocean data sets is not sampled uniformly, so re-sampling of each SST data set is likely needed for a full understanding of trends. It is important to note that the N₂O trend retrievals using these data indicate some drift in AIRS channels sensitive to N₂O, but far more work is needed to determine if this is real, and if the channels affected are known to have problems.

These retrievals give you itme (a) above, relative trends among channels where geophysically reasonable trends have been removed. Examination of the residuals (ours are 16 years by 40 latitude zones, with 16-day time increments) shows a number of differences in the time dependence of channels that should be similar. We propose that datasets such as this be given serious study, which is a big job since there are so many channels!

Managing the processing part of these studies is non-trivial. The steps we take at UMBC for this work include:

• Subset L1c frequency corrected clear ocean data (uniformity filter)
• Match each scene to ERA-Interim. (Need to add OISST and ERSSTSv5)
• Create 16-day B(T) averages in 40 zonal bins
• Generate B(T) anomalies (remove seasonal)
• Perform optimal-estimation retrievals on B(T) anomalies (running it hundreds of times)
• Examining the time dependence of the fit residuals to understand relative trends in channels
• Compare the CO₂ N₂O, CH₄, and SST trends from these fits to in-situ, etc.
• Try to relate differences to channel trends, using OE weight matrix

We can do all of the above steps with IASI and CrIS (and already have done some of this for shorter time periods). We recently have successfully converted the IASI ILS in these datasets to the AIRS ILS, which is very helpful. For CrIS, the only approach is to convert AIRS to the CrIS ILS (CHIRP), which allows some studies, although it average over AIRS channels.