ArticlePDF Available

# Examination of space-based bulk atmospheric temperatures used in climate research

Authors:

## Abstract and Figures

The Intergovernmental Panel on Climate Change Assessment Report 5 (IPCC AR5, 2013) discussed bulk atmospheric temperatures as indicators of climate variability and change. We examine four satellite datasets producing bulk tropospheric temperatures, based on microwave sounding units (MSUs), all updated since IPCC AR5. All datasets produce high correlations of anomalies versus independent observations from radiosondes (balloons), but differ somewhat in the metric of most interest, the linear trend beginning in 1979. The trend is an indicator of the response of the climate system to rising greenhouse gas concentrations and other forcings, and so is critical to understanding the climate. The satellite results indicate a range of near-global (+0.07 to +0.13°C decade⁻¹) and tropical (+0.08 to +0.17°C decade⁻¹) trends (1979–2016), and suggestions are presented to account for these differences. We show evidence that MSUs on National Oceanic and Atmospheric Administration’s satellites (NOAA-12 and −14, 1990–2001+) contain spurious warming, especially noticeable in three of the four satellite datasets. Comparisons with radiosonde datasets independently adjusted for inhomogeneities and Reanalyses suggest the actual tropical (20°S-20°N) trend is +0.10 ± 0.03°C decade⁻¹. This tropical result is over a factor of two less than the trend projected from the average of the IPCC climate model simulations for this same period (+0.27°C decade⁻¹).
Content may be subject to copyright.
Full Terms & Conditions of access and use can be found at
https://www.tandfonline.com/action/journalInformation?journalCode=tres20
International Journal of Remote Sensing
ISSN: 0143-1161 (Print) 1366-5901 (Online) Journal homepage: https://www.tandfonline.com/loi/tres20
Examination of space-based bulk atmospheric
temperatures used in climate research
John R. Christy, Roy W. Spencer, William D. Braswell & Robert Junod
To cite this article: John R. Christy, Roy W. Spencer, William D. Braswell & Robert Junod (2018)
Examination of space-based bulk atmospheric temperatures used in climate research, International
Journal of Remote Sensing, 39:11, 3580-3607, DOI: 10.1080/01431161.2018.1444293
UK Limited, trading as Taylor & Francis
Group.
Published online: 08 Mar 2018.
Submit your article to this journal
Article views: 5707
View related articles
View Crossmark data
Examination of space-based bulk atmospheric temperatures
used in climate research
John R. Christy, Roy W. Spencer, William D. Braswell and Robert Junod
Earth System Science Center, The University of Alabama in Huntsville, Huntsville, AL, USA
ABSTRACT
The Intergovernmental Panel on Climate Change Assessment Report
5 (IPCC AR5, 2013) discussed bulk atmospheric temperatures as
indicators of climate variability and change. We examine four satellite
datasets producing bulk tropospheric temperatures, based on micro-
wave sounding units (MSUs), all updated since IPCC AR5. All datasets
produce high correlations of anomalies versus independent observa-
tions from radiosondes (balloons), but dier somewhat in the metric
of most interest, the linear trend beginning in 1979. The trend is an
indicator of the response of the climate system to rising greenhouse
gas concentrations and other forcings, and so is critical to under-
standing the climate. The satellite results indicate a range of near-
1
) and tropical (+0.08 to +0.17°C
1
) trends (19792016), and suggestions are presented to
account for these dierences. We show evidence that MSUs on
(NOAA-12 and 14, 19902001+) contain spurious warming, espe-
cially noticeable in three of the four satellite datasets.
inhomogeneities and Reanalyses suggest the actual tropical (20°S-20°
N) trend is +0.10 ± 0.03°C decade
1
.Thistropicalresultisoverafactor
of two less than the trend projected from the average of the IPCC
climate model simulations for this same period (+0.27°C decade
1
).
ARTICLE HISTORY
Accepted 6 February 2018
1. Introduction
The study of human-induced climate change depends to a considerable extent on the
observations of the Earth system that attempt to characterize the change over time of
many aspects of the climate. In particular, accurately documenting the planetsbulk
atmospheric temperature is key because this response variable is directly tied to the
accumulation of heat in the Earth system, being in contact with the oceans (which
dominate heat accumulation) and thus is an indicator of the climate response to extra
greenhouse gases. With increasing concentrations of greenhouse gases, the rate at which
the heat accumulates becomes an indicator of the sensitivity of the climate to the forcing,
(i.e. the magnitude of temperature response vs. the magnitude of the extra forcing.)
The bulk atmosphere, specically the troposphere (the air from the surface to the
stratosphere, or about 85% by mass), is an especially informative layer because it is
CONTACT John R. Christy Christy@nsstc.uah.edu ESSC, UAH, 320 Sparkman Drive, Huntsville, AL, 35805, USA
INTERNATIONAL JOURNAL OF REMOTE SENSING, 2018
VOL. 39, NO. 11, 35803607
https://doi.org/10.1080/01431161.2018.1444293
(http://creativecommons.org/licenses/by-nc-nd/4.0/), which permits non-commercial re-use, distribution, and reproduction in any
medium, provided the original work is properly cited, and is not altered, transformed, or built upon in any way.
anticipated to show the most pronounced bulk temperature response to greenhouse
forcing (Christy and McNider 2017). In Figure 1 we show the vertical and latitudinal
atmospheric temperature trend from a computer simulation of 19792016 to demon-
strate this feature, common to climate model simulations, in which the upper air
throughout the troposphere will generally warm more rapidly than the surface for the
tropics and mid-latitudes (for other model results see Christy and McNider 2017).
Because of the mass involved in the bulk troposphere, monitoring its average tempera-
ture provides a more representative answer to the question of how much heat is
accumulating in the atmospheric climate system than would surface measurements.
There is a relatively large body of literature related to the construction of tropo-
spheric temperature datasets and how they are applied to issues of climate variability
and change (e.g. Christy and McNider 2017; Santer et al. 2017; Hartmann et al. 2013;
citations therein). The key points of uncertainty in these papers deal with the condence
one may have regarding the various construction techniques of the products, particu-
larly how they impact the resulting trend magnitudes. This is important because as new
versions of the datasets are produced, trend magnitudes have changed markedly, for
example the central estimate of the global trend of the mid-troposphere in Remote
Sensing Systems increased 60% from +0.078 to +0.125°C decade
1
, between consecu-
tive versions 3.3 and 4.0 (Mears and Wentz 2016). Lower trends would suggest relatively
modest sensitivity of the climate system to extra greenhouse gas forcing, while higher
trends would support greater sensitivity. Thus, the trend magnitudes are critical, for
example, to understanding the response of the climate system to enhanced forcing. In
this paper, our focus will be on the credibility of the datasets and associated trends,
Figure 1. Latitude Altitude cross-section of 38-year temperature trends (°C decade
1
) from the
Canadian Climate Model Run 3. The tropical tropospheric section is in the outlined box.
INTERNATIONAL JOURNAL OF REMOTE SENSING 3581
suggesting reasons for their dierences and oering a best estimate based on multiple,
independent eorts.
The satellite-monitored layer for this study is commonly referred as the mid-tropo-
sphere (T
MT
) because the peak of energy received by the satellite originates in the mid to
upper troposphere with the main weighting function (96% of the signal) going from the
surface to 70 hPa. Some stratospheric inuence occurs, where trends are negative,
especially outside the tropics, thus trends of T
MT
will be slightly lower than trends of
the troposphere alone.
In this study we shall examine the satellite, radiosonde and reanalyses datasets to
assess their similarities and dierences for the purpose of understanding how the global
bulk-atmospheric temperature has uctuated. We shall look at global comparisons of 564
stations of the Integrated Global Radiosonde Archive (IGRA) dataset as well as sub-sections
(with number of stations in parentheses) as follows: 30°S-30°N (135, to accommodate one
of the datasets), Australian radiosondes (34) and those produced for the United States by
the VIZ Manufacturing Company (US VIZ) radiosondes (32). These latter two subsets have
been studied extensively in the past, including by Christy, Spencer, and Norris (2011),
hereafter C11, and will provide continuity with those studies. The reader is encouraged to
access C11 for much of the background that will be briey discussed herein.
We shall also examine T
MT
as produced from homogenized radiosonde datasets and
Reanalyses in the physically-coherent deep tropics (20°S-20°N). From this analysis, we
shall briey examine these tropical results by comparing them with 102 climate model
simulations utilized in the Intergovernmental Panel on Climate Change Assessment
Report 5 (IPCC AR5, Flato et al. 2013) and made available through the Koninklijk
Nederlands Meteorologisch Instituut (KNMI) Climate Explorer (van Oldenbrogh 2016).
2. Satellite microwave and radiosonde observations of the bulk
atmosphere
Since late 1978, the United States included microwave sounding units (MSUs) on board
their polar-orbiting operational weather observation satellites. With the launch of the
National Oceanic and Atmospheric Administrations NOAA-15 spacecraft in 1998, a new
instrument, the Advanced MSU (AMSU) began service that has continued to be placed
into orbit by succeeding U.S. NOAA and National Aeronautics Space Administration
(NASA) missions as well as those of the European Space Agency. For the layer discussed
here (roughly the surface to 70 hPa), the radiometer on the MSU channel 2 and AMSU
channel 5 measures the intensity of emissions, which is directly proportional to tempera-
ture, from atmospheric oxygen near the 53.7 GHz band (Spencer and Christy 1990). Data
that have been utilized to generate the satellite time series have been provided from nine
spacecraft that have carried the MSU (19792001+, the University of Alabama in Huntsville
(UAH) ceases to use the last MSU in mid-2001, others through 2004) and six AMSU (1998-
present). These spacecraft have been launched roughly every two or three years and
usually have a multi-year overlap with previously-launched, on-orbit spacecraft so that
intersatellite biases may be calculated and removed, producing a single time series.
Products from four groups who generate bulk atmospheric temperatures from satellites
are available from which we are able to provide a substantial update to our previous work
(C11). All datasets have been revised and are available as monthly gridded anomalies in
3582 J. R. CHRISTY ET AL.
new versions referred to as UAHv6.0, Spencer, Christy, and Braswell 2017), Remote Sensing
Systems v4.0 (RSSv4.0, Mears and Wentz 2016), NOAAv4.0 (Zou and Wang 2011 plus
website updates) and a new dataset, University of Washington v1.0 (UWv1.0, Po-Chedly,
Thorsen, and Fu 2015). The datasets provide monthly gridded T
MT
anomalies on a 2.5
degree near-global grid (85°S-85°N), though UW processes 30°S-30°N only.
Each group builds the datasets in dierent ways to account for (a) small calibration
drifts once on orbit, (b) radiometer response uctuations due to variable solar insolation
on the instrument, (c) non-climatic temperature variations that occur when the space-
craft drifts east or west from a xed local crossing time at the equator (diurnal drift), (d)
intersatellite biases, and (e) loss of altitude due to frictional drag from the thin atmo-
sphere. The publications cited above for the various satellite datasets delve into the
details of how each group has decided to account for these inuences and should be
referred to if the reader has further interest.
The key metric that shows the greatest divergence among the satellite datasets is the
linear trend over the period examined here (19792016), ranging among the datasets to
as much as 50% from a central estimate in the tropics for example (see later). As noted in
the introduction, the trend is a critical parameter to aid in understanding how the
global, bulk atmosphere is responding to the extra forcing brought about by increasing
concentrations of greenhouse gases. As such, we shall focus more on the trend as the
metric of discussion because (a) it reveals the greatest discrepancies and (b) it is critical
to the science issue of human-induced climate change.
A new update of a major radiosonde dataset from NOAA, IGRA v.2 or IGRAwith data
through mid-2016, will be used as a source for independent comparisons (Durre and Yin
2011). Key updates to the IGRA dataset are (1) inclusion of data not found in IGRA v.1, (2)
more convenient organization and formatting, (3) inclusion of derived parameters and (4)
updates in station metadata. Most of the analysis below will utilize the period 19792015,
but in a few cases will use 19792016 recognizing that the stations used here from IGRA,
though containing the peak warm temperatures of early 2016, end in June 2016.
Unlike the satellite data that provide near-global coverage from a single instrument
per satellite, IGRA radiosonde stations are scattered around the globe with greatest
coverage over the northern midlatitude continents. Most stations release their balloons
either once or twice a day (00 and/or 12 Coordinated Time Universal or UTC). From these
daily data, microwave brightness temperatures are calculated and monthly averages are
generated for each station. Ten days of data in the month are required for a monthly
average to be calculated for each of the observation times. In many cases, only one of
the release times (00 or 12 UTC) exceeded the days-per-month threshold, so several
stations provided only one release time for their time series. If sucient data were
available from the two release times, the data were merged into a single time series after
accounting for diurnal biases and dierences in mean annual cycles. Otherwise, the time
series consisted of data from only one release time. In all comparisons below, if a month
were missing in the radiosonde data, it was set to missing in the satellite data so there
would be no aliasing of means or annual cycles between the two data sources.
IGRA stations were accepted if at least 240 monthly observations were available (of
the 450 possible in the Jan 1979 to Jun 2016 period.) Then, a nal quality check was
performed. If the monthly correlation of anomalies between the unadjusted station data
and the satellite data at the corresponding gridpoint exceeded 0.70, the station was
INTERNATIONAL JOURNAL OF REMOTE SENSING 3583
accepted for the comparison studies to follow. In general, correlations versus satellites
for stations with fairly systematic data collection, but even with changes in instrumenta-
tion, were on average around 0.90, with many above 0.95. Of the 620 stations with
240 months of data, 56 failed this correlation test due to signicantly erratic time series.
The IGRA station-by-station, (usually) twice-daily radiosonde data (temperature and
humidity) on all reporting levels are vertically interpolated to 61 prescribed pressure
levels for the calculation of the satellite temperature through a radiative transfer model
based on Ellingson and Fouquart (1991,dening the pressure levels used) and
Rosenkranz (1993) for the function code. These are then averaged into monthly values
from which a mean annual cycle and anomalies are calculated. These are equivalent to
the satellite-observed T
MT
anomalies for direct comparison with the satellite tempera-
tures at the station/grid-point (2.5°x2.5°) level.
2.1. Comparison methodology and adjustment procedure
We will show below both the direct comparison of IGRA vs. the satellite time series
without any adjustment to any dataset as well as results using adjusted IGRA data. The
non-adjusted result is the most independent of the comparisons to be shown which
means that the dierences in the time series may be due to radiosonde and/or satellite
problems. Unfortunately radiosondes have undergone events which impact the time
series, i.e. changes (improvements) in software, instrumentation, and factors such as
cord length between the balloon and the instrument box. These changes are generally
detectible as a shift in the time series (examples will be shown later). Unlike surface
temperature stations that may suer from slow temperature drifts due to gradual
changes in the immediate surroundings for which adjustment is extremely dicult,
radiosonde time series usually are composed of homogeneous segments at the end-
points of which are sudden shifts as a change is instituted (e.g. Christy et al. 2006;
Christy, Norris, and McNider 2009; Haimberger 2007; Zhang and Seidel 2011). As a result,
we are able to detecta radiosonde change by examining the time series of dierences
When a signicant shift in the dierence-time-series is detected (by the simple
statistical test of the dierence of two segments of 24-months in length on either side
of the potential shift) we then adjust the radiosonde to match the satellite at that shift
point. In the sections below we shall use a t-test value of 3.0 to detect and adjust for a
shift (C11). Each satellite will be utilized to generate the shift points according to its own
time series, thus IGRA ADJwill be specic for each satellite dataset. Satellite time series
will not be adjusted in anyway.
This adjustment procedure does not guarantee that the adjusted IGRA station time
series is more correct. It is entirely possible that a detected shift may result from an issue
with the satellite, for example when a new satellite is merged into the time series, there
may be a shift introduced, especially on a regional scale. We are assuming that most of
the shifts detected are indeed IGRA station issues but recognize that when such a shift is
applied as a correction to the radiosonde, we are forcing an improvement in the
agreement between the radiosonde and satellite in a contrived way. Because the
satellites have varying methodologies of construction (UAH is especially dierent than
the other three) this will provide more than one test of the radiosonde breaks.
3584 J. R. CHRISTY ET AL.
A unique aspect of this study is the idea that each satellite dataset may be thought of
as an independent reference datasetrelative to the radiosondes. By using identical
methods to correctradiosondes with diering satellite datasets, we will have multiple
results for which common and disparate results may be forthcoming. For example, as
will be shown later, there is a common result among all satellite datasets regarding
trends vs. radiosondes in the 1990s which gives us condence in its veracity.
2.2. Averaging
We shall present results in which the comparison data have been averaged for multi-
station means. We gather the stations into 5° latitude x 5° longitude grid boxes,
merge the anomalies of the stations (accounting for biases) and generate a single
time series based on all stations in the grid box. However, recall that each individual
station is adjusted relative to its corresponding 2.5° x 2.5° satellite gridbox time
series before being placed into the 5° x 5° gridbox. For the global and sectional
averages we then average the 5° latitude x 5° longitude time series into 5° latitude
by 30° longitude grid boxes and compute the area-weighted global/sectional mean
from there. We found that several grid boxes in populated regions contained 4 or 5
stations, so giving each of these the same weight as an isolated island station does
not properly account for spatial characteristics of a large-region average (Figure 2).
Hence the gridding process will have the advantage of providing a quantity that
greatly reduces spatial inconsistencies.
3. Results
3.1. Near-global results
In the following we will show several metrics of comparison between IGRA and the
satellite datasets with the goal of synthesizing what is discovered into an estimate of the
true bulk atmospheric trends and oer evidence-based suggestions as to why the
Figure 2. Map of radiosonde locations on a 5° x 5° degree grid with number of stations indicated by color.
INTERNATIONAL JOURNAL OF REMOTE SENSING 3585
datasets dier. We begin with the global time series of the unadjusted IGRA data and
subtract from that the equivalent satellite time series both unadjusted and with adjust-
ments to IGRA (Figure 3). There were 564 IGRA stations which met the requirements for
inclusion. In the top time series of Figure 3, IGRA UnAdj T
MT
demonstrates the signicant
inuence of El Nino/Southern Oscillation events with warm phases in 1983, 1987, 1990,
2005, 2007, 2010 and major ENSOs in 199798 and 201516. The dierences between
each satellite dataset and unadjusted and adjusted IGRA stations are shown as well.
To quantify the obvious dierences between IGRA and the satellite datasets in
Figure 3 we shall display statistical information in the following three gures. In
Figure 4 we show the standard deviation of the dierence-time series for unadjusted
and adjusted IGRA station data. It is evident that the adjustment procedure has per-
formed as anticipated and has considerably reduced the magnitude of the dierences in
all datasets. In the unadjusted and adjusted results, we see UAH produces the smallest
In an associated statistic, we show the correlations of IGRA, both monthly and
annually, versus the various satellite time series in Figure 5. Here again, as anticipated,
we nd improved correlations in the adjusted data, with some above 0.99. However,
recall that our adjustment procedure forces the IGRA stations to align with the satellite
data at places where shifts are detected, but even so, the agreement is remarkable. We
again remind the reader that we are demonstrating only relative agreement between
time series since none of the products examined here may be considered to be an
absolute reference standard. By way of approximation, using the Fisher r-to-z transforma-
tion for 100 degrees of freedom, dierences are generally statistically dierent from one
Figure 3. Top: Time series of the unadjusted near-global anomalies of the IGRA radiosonde data
(black). Thick Blue (UAH), Red (RSS) and Green (NOAA) time series are the dierences (unadjusted
IGRA minus satellite). Thin time series are the adjusted IGRA station data minus satellite time series.
3586 J. R. CHRISTY ET AL.
another for value above 0.9 if the lower value is of a magnitude smaller than the
dierence between the higher value and 1.0. So, for example, two correlations of 0.98
and 0.95 are statistically dierent since the dierence between 0.98 and 0.95 is greater
than the dierence between 1.00 and 0.98. In Figure 5, the various correlations approach
Figure 4. Standard deviation (°C) of the monthly dierence time series in Figure 3.
Figure 5. Monthly and annual correlation of global anomalies between the unadjusted and adjusted
IGRA time series and that of the satellites.
INTERNATIONAL JOURNAL OF REMOTE SENSING 3587
signicant dierence relative to UAH, but do not reach it. That UAH demonstrates higher
levels of agreement with IGRA may simply be due to unrelated but common errors in
both, though we shall oer some explanations below.
The third statistical metric, and physically most important, is the linear trend, and is
displayed in Figure 6. We include the satellite trend that is calculated from the full
(complete) satellite grid for the section (near-global here) to compare with those which
are calculated from satellite grids in which IGRA stations exist. The general result here is
that the full grid trends, which include more oceanic area, tend to be lower than the
partial IGRA grid except for NOAA whose relative trends over oceans are more positive
than that indicated by UAH and RSS (see later).
TheunadjustedIGRAtimeseriesindicate closest agreement with UAH and thus
little overall change is generated when adjusted even though the IGRA stations
much more positive than the unadjusted IGRA trend and so their adjusted IGRA
time series become more positive, though for NOAA, the adjusted trend is within
1
of UAH.
The dierence-times-series (Figure 3) reveal useful information. Common to all unad-
justed dierence time series (thicker time series) is a relative warming in the satellite
data vs. IGRA for the period 19902000. This is followed by a similar, though of lesser
magnitude (except UAH), relative satellite cooling from 2003 to around 2006. We show
the magnitudes of these dierences in Table 1 in which we have computed the
dierences between relatively stable periods. UAH data reveal the least amount of
relative warming between the rst two periods and the most relative cooling between
the second two periods in nearly all cases.
As can be seen in the Adjustedresults of Figure 3 and Table 1, the shift-point
adjustment procedure eliminates most of the two main discrepancies, though it is
applied to each radiosonde independently. Even so, the rst drift in the dierence
Figure 6. Linear trend (°C decade
1
) of the various global time series for 19792015. The Full grid
satellitetrend is the trend of the satellite data at all grids (i.e. complete coverage), whereas satellite
is the trend based on IGRA station grids only (i.e. partial coverage).
3588 J. R. CHRISTY ET AL.
time series in 1990s retains some magnitude (Table 1) in the adjusted data, especially for
the non-UAH datasets. As will be discussed later, this drift is hypothesized to be a
consequence of spurious warming in the NOAA-12 and 14 processed satellite data
and not spurious cooling in the IGRA radiosondes.
3.2. 30°S 30°N
With the 30°S-30°N latitude band we now bring in UW data into the comparison study
where for this region there are 135 stations to examine. In Figure 7 we show the low-
latitude time series as described in Figure 3 and which depicts similar features the
relative warming of satellites from about 19902000 and relative cooling 20002006.
More obvious here is the various amounts of relative cooling from the satellites in post-
2011 unadjusted IGRA data, least for RSS and most for NOAA. These dierences are
The magnitude of the correlation of the anomalies (Figure 8)ishighestforUAH,
though in the adjusted-IGRA data, UAH, RSS and UW are statistically indistinguishable.
Figure 9 displays the trends for this latitude section and as was true in the case of
global average, the IGRA-adjusted NOAA and UAH (red) values are very similar to the
1
for all satellites are less positive than those of the satellite to which they have been
tuned, which may be explained by residual relative drifts in the satellite data during
periods when there are no breakpoints to detect (C11). One additional observation to
note is, as mentioned earlier, that NOAAstrendofthefull grid(pink) is much more
positive than its trend calculated from only the IGRA grids (green) while the other
three datasets are virtually the same (within ±0.003°C decade
1
). This essentially
indicates more positive trends over the ocean than land in NOAA and is a peculiarity
In support of the information in Table 1 we show in Figure 10 the relative
dierence over the periods between satellites and homogenized radiosonde datasets
provided by dierent institutions (satellite values were calculated from grids contain-
ing radiosonde data). The result is clear that RSS, NOAA and UW experience
Table 1. Relative segment dierences (°C) in the satellite comparison with unadjusted and adjusted
IGRA data for globe and 30°S-30°N section. Periods are (1) 19791990, (2) 20002003 and (3)
20062015. Thus (2)-(1)indicates the average of 20002003 minus the average of 19791990. The
rst value for UAH is +0.116°C meaning UAH warmed +0.116°C more than IGRA UnAdj between the
two periods 19791990 and 20002003. The adjustments are unrestricted, i.e. the satellite alone
determines all breakpoints based only on the breakpoint detection scheme.
GL
(°C)
30°S-N
(°C)
GL
(°C)
30°S-N
(°C)
UAH +0.116 0.114 +0.144 0.121 +0.017 0.025 +0.031 0.005
RSS +0.248 0.111 +0.294 0.108 +0.056 0.019 +0.067 +0.010
NOAA +0.204 0.101 +0.211 0.092 +0.087 +0.003 +0.114 +0.011
UW +0.318 0.187 +0.084 0.033
INTERNATIONAL JOURNAL OF REMOTE SENSING 3589
signicant warming relative to the homogenized radiosonde datasets and reanalyses
between 1990 and 2000. This is further and substantial evidence that the satellite
time series are characterized by spurious warming in the 1990s. [We note again the
peculiarity of the oceanic warming in the NOAA dataset in that the relative
Figure 7. As in Figure 3 above but for stations in 30°S-30°N and now includes UW (purple).
Figure 8. Correlations with IGRA values as in Figure 5 above but for 30°S-30°N and now includes UW.
3590 J. R. CHRISTY ET AL.
dierences between NOAA and the homogenized radiosonde datasets appears less
than that of RSS and UW (Figure 10 left three comparisons) whereas it is greater for
the reanalyses comparison. This is so because NOAAs high warming rate over the
Figure 9. As above in Figure 6 but for 30°S-30°N.
Figure 10. Magnitude of the relative dierence between two periods for the respective satellite
datasets (colored bars) and the respective radiosonde-based datasets (i.e. positive value indicates
satellite warmed more than the radiosonde-based data between dened periods.).
INTERNATIONAL JOURNAL OF REMOTE SENSING 3591
oceans is not captured in the grid-matched homogenized radiosonde comparisons
whereas the Reanalyses are full-grid in the datasets, and thus include the oceans.]
Figure 10 also supports the idea that the homogenized radiosonde datasets (as
shown below for Australian and VIZ stations) are likely characterized by spurious warm-
ing after 2000. We see close agreement between the satellite datasets and the
Reanalyses for the dierences between the latter two periods ((20062015) minus
(20002003)) yet signicant dierences for all satellite datasets vs. the homogenized
radiosonde datasets. While not denitive, these results and those below suggest spur-
ious post-2000 warming in the radiosondes (that is largely eliminated in our adjustment
process, see Table 1 far right column.)
3.3. Australia stations
In this and the following section (3.4 U.S. VIZ) we look more closely at the comparisons
between radiosondes and satellites because there is a body of research that includes
considerable meta-data regarding the instrumentation and software of the radiosonde
observations for these two nations each of which has wide geographical coverage
(tropics to polar latitudes) with one in each hemisphere. With such information, we
will have the opportunity to make hypotheses about potential satellite problems. We
begin with Australian radiosondes for which we have clear evidence of positive shifts
due to instrumentation changes (Christy and Norris 2009, C11). These changes were not
simultaneous across the network, so we continue to utilize the single-station adjustment
procedure as used above, i.e. allowing each satellite dataset to detect the shifts. In this
comparison, we have reduced the longitudinal dimension to 10° while retaining the 5°
latitudinal spacing. There were 34 stations in the sample that met the criteria for all of
Australia-controlled stations and 18 for the region north of 30°S for UW comparisons.
For the 34-station network, all three of the correlations of the adjusted IGRA data
were between 0.98 and 0.99 for annual anomalies (not shown). Unadjusted annual
correlations were 0.915, 0.916, and 0.865 for UAH, RSS and NOAA respectively. For the
18-station network in the low latitudes (north of 30°S), the four satellite datasets
0.916 (0.972), 0.889 (0.976) and 0.847 (0.957) for UAH, RSS, NOAA and UW respectively.
Of most interest is the information on trends shown in Figure 11. As expected from
previous research (Christy and Norris 2009), the unadjusted Australian data (black and gray)
indicate more positive trends than any dataset. We note that when UAH and RSS are used
to adjust the radiosondes the resultant time series have very similar trends, within 0.01°C
1
of each other in both cases (blue and red). Adjustments applied utilizing NOAA
and UW for breakpoint detection produce very similar magnitudes with each other and are
more substantial, i.e. producing a less positive adjusted trend than when using UAH and
RSS. NOAAs satellite trend for the 34-stations is considerably less than UAH and RSS
(orange), hence when applying adjustments, will produce a less positive result.
In an attempt to understand Figure 11, we show the net accumulation of the break-
points detected by each satellite for the 25 Australia stations with > 400 months
reporting in Figure 12. The three main features are the detection of (a) a radiosonde
warming shift in the late 1980s, (b) a gradual radiosonde cooling period in the 1990s,
and (c) a radiosonde warming shift in late 2009 and early 2010 (conrmed by
3592 J. R. CHRISTY ET AL.
comparisons with the European Centre for Medium-range Weather Forecasts analyses, L.
Haimberger personal communication). UAH and RSS are highly consistent in their
detection rates and magnitudes, with their time series correlating at +0.93. NOAA is
less consistent, correlating with UAH and RSS at +0.79 and +0.73 respectively. The trends
1
) for the various time series using Australian stations only.
Figure 12. Net accumulation of breakpoint magnitudes (°C) for 25 Australia stations with > 400 months
of data. This may be interpreted as an index of the relative dierence between the raw Australian data
and the satellites, showing a relative warming vs. the satellites.
INTERNATIONAL JOURNAL OF REMOTE SENSING 3593
1
and NOAA respectively. The detection of a gradual radiosonde cooling trend in the
1990s is evidence of spurious warming in the satellites during the 1990s, as discussed
below, as there were essentially no changes in the instrumentation between 1991 and
2000. In other words, the satellites appear to detect breakpoints more or less randomly
due to their spurious warming rather than to spurious cooling in the radiosondes. That
NOAA detects fewer of these in the period appears to be related to the lower correlation
of NOAA vs. radiosondes so that the signicance level is not achieved as often to apply a
3.4. U.S. VIZ stations
A decision was made in NOAAs weather observing strategy before the 1980s to set
aside several radiosonde stations for which, as much as possible, a consistent set of
instruments and software would be maintained so as to create climate data records
suitable for long-term studies. These stations utilized the VIZ radiosonde, but as
improvements were naturally forthcoming through the years, these stations indeed
adopted new instrumentation from the same manufacturer but with as much backward
compatibility as possible (Christy and Norris 2006). Unlike the Australia network, changes
to the VIZ network were usually near-simultaneous and thus easily spotted using the
composite time series when checking for shifts (see examples in C11). As with the
Australian study we shall bin the data into 10° longitude x 5° latitude cells.
Christy and Norris (2006) identied four shifts due to instrument or software changes
between 1979 and 2004, occurring in (a) mid-1983, (b) end-1989, (c) mid-1997 and (d)
end-2001. To this list, with data to mid-2016 now, the comparison against all satellite
datasets indicate two additional, signicant shift points (a) 2007 (when the instruments
switched to Global Positioning System (GPS) tracking) and (b) late 2012 (cause unknown
as metadata not updated through this date). There is a possibility that introducing the
data from the Meteorological Operational (METOP) spacecraft launched by the European
Organization for the Exploitation of Meteorological Satellites in 2012 (METOP-B) may
have caused a spurious shift in the satellite data. However, UAH performed the test with
and without METOP-B, yet detected the shift with as much signicance as the other
satellite datasets both ways, so it is unlikely that the addition of METOP-B (i.e. a satellite
issue) caused the shift.
In this analysis we shall apply corrections only at these six shift points, the value of
which will be determined by each satellite dataset. This is a more restrictive test than
previous tests because with the earlier tests, all shift-points were accommodated with-
out any knowledge of which dataset, IGRA or satellite, might have created the error. In
this test, we are essentially certain that the IGRA data requires adjustment and we shall
not allow other shift-points to be applied, assuming they are likely due to satellite
problems. As such, this test will have the potential of identifying problems with the
satellites more directly. We shall again divide the analysis into two groupings (1) all 32
stations for UAH, RSS and NOAA, and (2) 12 stations within the 30°S-30°N band to allow
direct comparison of all satellite datasets with UW.
In Figure 13 we show the adjustments applied to the composited VIZ stations as
determined by the satellite data (for low latitudes). Generally speaking the individual
3594 J. R. CHRISTY ET AL.
shift-point values are slightly more positive for RSS and NOAA than UAH and UW, but overall
are quite consistent among the satellite datasets. It is interesting to note that the accumula-
tion of these adjustments is 0.03, +0.08, +0.03 and +0.02 for UAH, RSS, NOAA and UW
which is consistent with the less positive trend of UAH and the higher trend of RSS.
The correlations of monthly and annual anomalies are shown in Figure 14 for
adjusted (pink and red) and unadjusted (black and gray) IGRA data. In all cases, UAH
data demonstrate higher levels of agreement and UW the lowest with RSS and NOAA
producing similar values. As noted earlier, the Fisher z-to-r transformation indicates the
monthly correlations of RSS, NOAA and UW are near or just signicantly less than UAH
for all results. The trend magnitudes are shown in Figure 15. In all cases the adjusted
trend (blue) of the 32-station composite is less than the unadjusted radiosonde trend
(orange.) We note that the more positive trend values of the satellites (less so in UAH)
indicates relative trend dierences during periods without radiosonde changes, in
particular as shown below, the period of the 1990s.
We focus now on an issue raised in previous research (C11) that suggested the MSUs
during the 1990s exhibited a spurious warming drift that was unexplained and (mostly)
untreated by the satellite dataset providers. In Figure 16 we show the annual anomaly
time series of the IGRA data for low-latitude US VIZ stations with the dierence time
series (IGRA-adjusted minus satellite) below. As the mean of all of the radiosonde minus
satellite time series indicates, the largest departure from a at dierence series is the
relative warming of the satellites between 1990 and 2000. There appear to be two other
smaller dierences, a slight relative satellite cooling in the early 1980s and after 2000.
We shall return to this topic in the discussion section, but here provide statistical
results. Taking only data from VIZ radiosondes for which no instrument or software
Figure 13. Values of the shift magnitude (°C) added to the 12 tropical/subtropical IGRA VIZ radio-
sondes from the indicated time forward.
INTERNATIONAL JOURNAL OF REMOTE SENSING 3595
changes were applied (Jan 1990 to Jun 1997), we nd the satellites warm relative to
radiosondes with a signicant magnitude when comparing the 48 months before and
after this period (mean satellite shift of +0.13°C.) After including the correction for the
Figure 15. Trend magnitudes (°C decade
1
data for 19792015.
Figure 14. Correlations of monthly and annual anomalies, unadjusted and adjusted between US VIZ
stations and the four satellite datasets, Jan 1979 Jun 2016.
3596 J. R. CHRISTY ET AL.
Jul 1997 VIZ shift (due to change from VIZB to VIZ-B2 instrumentation), the mean
satellite shift for the period Jan 1990 to Dec 2000, was +0.18°C. Thus the evidence is
strong that NOAA-12 and 14s MSUs were characterized by a spurious warming drift, as
also seen in the independent set of Australian radiosonde analysis above. We note that
this feature is also present in the global and tropical time series (Figure 3,7), which
involves the full IGRA dataset with numerous radiosonde stations. We note too that for
the longer period tested, the magnitude of the relative drifts were smallest for UAH,
especially in the low latitude stations and likely due to the unique way UAH merged this
section of the time series (see later). Given these results, we estimate the actual global
(full grid, Figure 6)T
MT
trend to be +0.10 ± 0.03°C decade
1
, i.e. a value slightly less
positive than satellite datasets on average indicate.
3.5. Other tropical T
MT
datasets (20°S-20°N)
Before moving to a synthesis and analysis of the results above we include one more test
involving homogenized radiosonde and reanalyses datasets. These datasets in various
ways seek to remove temporal inhomogeneities independently from any process per-
formed in this paper. For simplicity we will consider annual anomalies of the deep tropics
(20°S-20°N) for which Figure 1 indicates is a region of signicant response to generic
forcing of the troposphere of any kind with upper air warming faster than the surface. In
terms of magnitude, when comparing surface and T
MT
warming rates since 1979, the
theory, as indicated by model simulations, reveals that the rate of warming in the T
MT
layer
is an amplication of the surface warming rate by a factor of 1.4 (Christy 2017.)
There are four homogenized radiosonde and three Reanalyses datasets described in
Christy (2017) and listed in Table 2. The radiosonde datasets represent the homogeniza-
tion processes of the associated authors in which adjustments are calculated and
applied in dierent ways to the individual station records. The reanalyses represent an
Figure 16. Time series of the composite of annual anomalies of IGRA US VIZ and the dierence (°C)
versus the four satellite datasets for the 12 low-latitude stations.
INTERNATIONAL JOURNAL OF REMOTE SENSING 3597
assimilation of all available data, including radiosonde and microwave satellite data used
here, into a global general circulation model. (We note that University of Wien (UWien)
radiosonde datasets are indirectly interdependent with European Reanalysis European
Centre for Medium Range Forecasts Reanalyses-Interim (ERA-I) reanalyses, (Haimberger
2007) The assimilation process aids in identifying and then accounting for spurious
changes in the various observing systems. It is clear that each of the radiosonde and
reanalyses are independently constructed.
We next examine the central estimate of the trend metric for these products
(Figure 17). IGRA trends are calculated through June 2016 (the end of the IGRA-2
dataset) which included the peak warming in Feb 2016 from the El Niño event. [Note
that for IGRA-Adj we include a bracket to show the range of the adjusted trends from
the four satellite products (+0.056 to +0.124°C decade
1
).] The IGRA UnAdj trend is
clearly lower than the others, though only slightly, and indicates that the net impact of
adjustment procedures is to increase the trend. This is due in part to a common
improvement to many (not all) of the radiosonde instrument packages through the
years in which shielding of the direct sunlight, which had caused excessively warm
temperatures in the upper troposphere and stratosphere in the early years in some
models, was made more eective (e.g. Sherwood and Nishant 2015). Because T
MT
has a
portion of its signal from the upper troposphere and lower stratosphere, such an
improvement would increase the trend relative to no adjustment applied.
The results of Figure 17 indicate that the trend of the products not exclusively based
1
. Indeed, the
trend calculated by generating a time series of the annual average of anomalies using
the 8 products (counting the mean of IGRA-Adj as one of the samples) is +0.103°C
1
with a standard deviation among the trends of 0.0109°C. Determining
the error bounds is not straight-forward, but though we know each product was
independently constructed, the use of common data likely removes some degrees of
freedom (d.o.f.). With that in mind we select a t-test value calculated for 5 d.o.f. providing
a condence interval of ±0.028°C decade
1
1
.
Table 2. Datasets utilized in the comparison of tropical (20°S-20°N) T
MT
statistics.
Dataset Type Source Citation
UWien Radiosonde Average of RAOBCORE and RICH
RAOBCOREv1.5 Radiosonde Haimberger, Tavolato, and Sperka 2012
RICHv1.5 Radiosonde Haimberger, Tavolato, and Sperka 2012
NOAA/RATPAC-A2 Radiosonde Free et al. 2005
UNSW Radiosonde Sherwood and Nishant 2015
ERA-I Reanalyses Dee et al. 2011
JRA-55 Reanalyses Kobayashi et al. 2015
MERRA-2 Reanalyses Bosilovich et al. 2017
UWien: University of Wien.
RAOBCORE: RAdiosonde OBservation COrrection using Reanalyses.
RATPAC: Radiosonde Atmospheric Temperature Products for Assessing Climate.
UNSW: University of New South Wales.
ERA-I: European Centre for Medium-Range Weather Forecasts, Reanalysis Interim.
JRA-55: Japanese 55-year Reanalyses.
MERRA-2: Modern-Era Retrospective-Analysis for Research and Applications v. 2.
3598 J. R. CHRISTY ET AL.
Such a range marginally captures UAH on the low side (+0.082) and UW on the high end
(+0.130). Note that the range of the IGRA-adjusted trends from the four satellites is very
similar to this. Thus, (naively) applying the statistical results here we conclude that the
1
) are signicantly dier-
ent from the T
MT
trends calculated by major international eorts. We say naively
because we cannot reject a hypothesis that the results from this diverse set of data
providers who use a diverse set of methods may all be aected by common spurious
cooling problems.
We tested the 20°S-20°N trends of each satellite time series against the average of the
for Assessing Climate or RATPAC, University of New South Wales or UNSW) and the average
of the reanalyses (ERA-I, Japanese 55-year Reanalyses or JRA-55 and Modern-Era
Retrospective analysis for Research and Applications version 2 or MERRA-2) in Table 3.
Trend dierences between radiosondes and reanalyses are due to the subsampled grid
for the homogenized radiosonde datasets. The trend of the satellite datasets may be
thought of as the central estimate of the time series around which errors are distributed.
We nd that the dierence trend for time series beginning in 1979 and ending in 2005
(covering the period of hypothesized satellite warming see Table 1)issignicantly more
positive than that of either of the composite comparisons for RSS, NOAA and UW.
We conclude this section with an important comparison of the products shown here
with the output of the IPCC AR5 climate models (from the Climate Model
Intercomparison Project 5 or CMIP-5, Flato et al. 2013) using Representative
Concentration Pathway 4.5. These simulations were forced with estimates of the actual
forcing due to volcanic aerosols, greenhouse gasses, etc. through 2006 then estimates of
these forcings thereafter, so were anticipated to reproduce fairly closely the global and
Figure 17. The central estimate of the trend magnitudes (°C decade
1
) for T
MT
, 20°S-20°N for several
datasets described in Table 2. The IGRA Adj range represents the low to high values as adjusted by
the four satellite datasets.
INTERNATIONAL JOURNAL OF REMOTE SENSING 3599
tropical-scale temperature changes to this point. There were 102 simulations available
which were grouped into 32 time series according to institution and model type.
Simulations from institutions that provided multiple simulations were averaged for a
single time series, so the results and statistics are based on a sample of 32 model time
series. The satellite temperatures were calculated from temperatures at 17 pressure
levels of model output using a static weighting function identical to that used for the
We show the results in Figure 18 for the metric of most interest, the trend over the
period 19792016. The vertical pressure-level trends form the main part of the diagram
Table 3. Trend magnitudes and 95% condence intervals of the time series of
the dierences between the satellite datasets (SAT) and the average of the
homogenized radiosonde datasets (RAOB: UWien, RATPAC-A2, UNSW) and the
average of the Reanalyses (REAN: ERA-I, JRA-55, MERRA-2). (Personal
Communication, R. McKitrick.) The period is 19792005 (capturing the period
using MSU instruments) and the geographic extent is 20°S-20°N (see Figure 10).
1
with bold values indicating the central estimate of the
satellite trend is signicantly more positive than the comparison dataset.
19792005
Trend SAT minus RAOB
1
)
Trend SAT minus REAN
1
)
UAH +0.045 ± 0.066 0.008 ± 0.039
RSS +0.116 ± 0.070 +0.066 ± 0.049
NOAA +0.111 ± 0.069 +0.091 ± 0.053
UW +0.117 ± 0.069 +0.064 ± 0.040
Figure 18. Trend magnitudes (°C decade
1
) from radiosonde and CMIP-5 climate model simulations
over the period 19792016. In the upper box are the trend magnitudes of the T
MT
layer as calculated
by the various datasets dened in Table 2 and in this paper.
3600 J. R. CHRISTY ET AL.
and in the upper box is the single value of the T
MT
trend for the various datasets
discussed above. The main result here is the high value of trends in the simulations
(average for T
MT
1
) is signicantly above the estimates from
1
).
4. Discussion
We now discuss our results and oer some suggestions to explain the dierences
among the satellite datasets and provide estimates of the bulk atmospheric temperature
trends. We begin with a key result that aects all satellite data to some extent.
4.1. Satellite warming in the 1990s
We have shown substantial evidence to support the hypothesis that the satellite
datasets experienced spurious warming during the period that began with NOAA-12
and ended with NOAA-14 (19902001+) that could not be explained by the processes
already addressed. The evidence is seen in (a) the full IGRA comparisons in both global
and low-latitude regions, (b) the more controlled analysis using the US VIZ and the
Australian radiosondes separately (both global and low-latitude) and (c) the compar-
ison with Reanalyses and independently-constructed, homogenized radiosonde data-
sets. The impact is least obvious in UAH data and most in UW (Tables 1,4)e.g. note
from Table 4 (30°S-30°N, 12 stations) the relative dierence vs. adjusted US VIZ for
20002003 minus 19791990 was; +0.08, +0.19, +0.19 and +0.25°C for UAH, RSS, NOAA
and UW respectively.
We will discuss a merging detail here to help explain the discrepancies, though the
reader is encouraged to consult the original papers. In the period of interest (1990
2000), we note that NOAA-11 (19881994) and 14 (19952001+) were p.m.orbiters in
which the spacecraft was inserted into an orbit with nominal local equatorial crossing
time (LECT) of 1330 and with a purposeful drift toward later times in the afternoon to
avoid backing into local solar noon. NOAA-10 (19871991), 12 (19911998) and 15
(19982008+) were a.m.orbiters with nominal LECT of 0730. In general, a.m. orbiters
drifted much less than p.m. orbiters. In particular, NOAA-14 drifted well away from 1330,
being at 1700 by mid-2001 and 2030 (7 hours of drift) by the end of 2004. This was a far
Table 4. As in Table 1 except using the 32 US VIZ radiosondes and xed dates of breakpoint shift
data for full 32-station network and 30°S-30°N for the 12-station network. Periods are (1) 19791990,
(2) 20002003 and (3) 20062015. The adjustments are those using the dened breakpoint events
(breakpoints due to known/likely VIZ radiosonde changes).
(2)-(1)
32 Stns
(°C)
(3)-(2)
32 Stns
(°C)
(2)-(1)
12 Stns
(°C)
(3)-(2)
12Stns
(2)-(1)
32 Stns
(°C)
(3)-(2)
32 Stns
(°C)
(2)-(1)
12 Stns
(°C)
(3)-(2)
12 Stns
(°C)
UAH +0.104 0.122 +0.105 0.138 +0.147 0.044 +0.082 0.055
RSS +0.266 0.119 +0.281 0.091 +0.250 0.055 +0.186 0.058
NOAA +0.270 0.122 +0.287 0.139 +0.262 0.029 +0.193 0.085
UW +0.293 0.143 +0.248 0.104
INTERNATIONAL JOURNAL OF REMOTE SENSING 3601
greater continuous drift than utilized in the other p.m. satellites and part of the reason
UAH truncates NOAA-14 in mid-2001.
In the construction process, once the adjustments for diurnal drift and other issues
were applied, in both UAH and RSS products, NOAA-14 revealed a trend that was about
1
more positive relative to NOAA-15, a spacecraft carrying the new and
more carefully calibrated AMSU (Mo 2009). At this point UAH applied an objective
algorithm to calculate and remove trend dierences of NOAA-11 and 14 relative to
the three a.m. orbiters, NOAA-10, 12 and 15 (Spencer, Christy, and Braswell 2017). This
was done because these residual trend dierences were very likely due to the changing
impact of solar heating on the relatively rapidly drifting p.m. instruments. With the
truncation of NOAA-14 data in 2001 and the trend adjustment based on simultaneous
comparison with NOAA-12 and NOAA-15, the NOAA-14 trend dierence in UAH data
was considerably reduced. NOAA-12 was not impacted, however, as it was assumed to
be stable. The fact that the US VIZ comparison indicates the relative warming of the
satellites begins with NOAA-12 is a strong indication that it too was characterized by a
spurious warming trend that was not accounted for in the UAH trend adjustment. In any
case, this adjustment procedure is a partial explanation for the result that relative to the
other satellite datasets in nearly all comparisons, UAH correlates highest, has the lowest
magnitude of dierences and the least dierence in trends.
On the other hand, RSS (and likely NOAA and UW in some manner) choose to retain the
relatively warm trend of NOAA-14, which they termed an unexplained mystery(Mears
and Wentz 2016). This, combined with a likely spurious warming of NOAA-12, produces
the eect of liftingthe post-NOAA-14 time series up, producing a more positive trend.
As an experiment, Mears et al. recalculated the RSS overall trend by simply
truncating NOAA-14 data after 1999 (which reduced their long-term trend by 0.02 K
1
). However, this does not address the problem that the trends of the entire
NOAA-12 and 14 time series (i.e. pre-2000) are likely too positive and thus still aect
the entire time series. Additionally, the evidence from the Australian and U.S. VIZ
comparisons support the hypothesis that RSS contains extra warming (due to NOAA-
12, 14 warming.) Overall then, this analysis suggests spurious warming in the central
1
, which is consistent with results
shown later based on other independent constructions for the tropical belt. Details of
the merging procedure of NOAA and UW are not known to this detail, but possibly
are inuenced in the same way. Dierences in the diurnal correction may be impor-
tant as well as discussed below.
There are substantial dierences in dataset construction that separates UAH from
the others that impacts trends. One is the diurnal correction. UAH determines
corrections for the drift of the spacecraft through the diurnal cycle using empirical
evidence only, i.e. intercomparing diurnally-drifting vs. non-diurnally-drifting satel-
lites to calculate the eect directly. RSS and NOAA apply climate-model-calculated
values which are evidently not suciently satisfactory (Mears and Wentz 2016;their
Figure 7) while UW uses a numerical regression technique on residual dierences
between co-orbiting satellite pairs (both of which may be drifting) which also
contains remaining biases(Po-Chedly, Thorsen, and Fu 2015;theirFigure 2.) RSS
and NOAA then apply an additional step of optimizationrequired to compensate
for the initial attempt to accommodate diurnal cycle errors. For example, the
3602 J. R. CHRISTY ET AL.
NOAA-14 satellite after diurnal correction over land from the climate model alone
in RSS data, retained a trend relative to the co-orbiting NOAA-15 of +0.34°C
1
.Theoptimizationprocess reduced this discrepancy considerably to
1
as noted above (Mears and Wentz 2016). Similar procedures
were followed in NOAA and UW. This two-step method creates some dierences
in trends relative to UAH but it is clear that the attempted adjustment through a
climate model was suciently erroneous that further adjustments were required. It
is possible that a correction applied to a decient correction may not lead to
greater accuracy.
4.2. Satellite cooling in 2000s
The relative cooling of the satellite datasets versus IGRA from 2002 to 2006 is more
dicult to explain because once adjustments are applied, including the dened breaks
of US VIZ radiosondes, the dierences largely disappear (see Tables 1,4). The main shift
in the US VIZ is known for 2002 and once adjusted, the issue is mostly resolved.
However, even after adjustment UAH retains a noticeable relative cooling trend of
unknown origin during and following this episode which is also seen in NOAA and
UW. We examined the satellites in question (NOAA-15 and AQUA) and found them to
have a high degree of agreement (AQUA required no adjustments as it was a spacecraft
restricted to a constant LECT.) The evidence from all satellite datasets suggests that the
relative radiosonde warming during this period may be due to changes in a sucient
number of radiosonde stations to aect a global average. Even so, the comparative
evidence (section 3.5) still suggests that UAH retains a spurious cooling trend after 2002
that likely introduces a relative negative inuence of a magnitude of about 0.02°C
1
on the complete time series.
4.3. Tropical trends, 20°S-20°N
The issue of tropical trends introduced in Figure 1 is especially important for under-
standing the theory of the atmospheric response to increasing concentrations of green-
house gases. With the inclusion of independently homogenized radiosonde datasets;
UWien (average of RAOBCORE and RICH), NOAA, and UNSW, and the reanalyses from
major data-construction centers; ERA-I, JMA, and MERRA-2, we have more condence in
our conclusions. Figure 16 displays the 19792016 trends for the tropics (20°S-20°N). The
IGRA-ADJ represents the trend of the mean of the IGRA stations adjusted for breakpoints
by the four satellite datasets with the error bar indicating the spread of the results.
The median trend of the non-satellite-only (radiosondes and Reanalyses) datasets is
+0.098 and including the satellite datasets is +0.103°C decade
1
. As noted earlier, one
may conclude that the tropical T
MT
trend is +0.10 ± 0.03°C decade
1
based on dierence
statistics of the non-satellite-only data. This result suggests, again, that RSS and NOAA
(in particular) are impacted by extra warming, especially in the NOAA-12 to 14 satellite
period, to the extent that their overall central-estimate of the trend exceeds that
produced in independent methods by a signicant margin. We repeat an underlying
fact here however, that these results may be viewed as strongly suggestive but not
denitive as there is no true reference with which trend-precision may be determined.
INTERNATIONAL JOURNAL OF REMOTE SENSING 3603
[The use of Australian and US VIZ radiosondes during periods without instrument
changes is perhaps as close as we are able to approach such a reference at this time,
and that analysis supports this hypothesis.]
4.4. General remarks
When examining all of the evidence presented here, i.e. the correlations, magnitude of
errors and trend comparisons, the general conclusion is that UAH data tend to agree
nized radiosonde datasets and (c) Reanalyses at a higher level, sometimes signicantly
so, than the other three. We have presented evidence however that suggests UAHs
1
less positive than actual. NOAA and UW tended to
show lower correlations and larger dierences than UAH and RSS indeed UAH and RSS
generally displayed similarly high levels of statistical agreement with the radiosonde
datasets except that UAH was usually in closer agreement regarding trends. UW tended
(e.g. Figure 13) while NOAA displayed a peculiarity in that low latitude trends over the
ocean were much more positive than the other datasets. We have presented evidence
that strongly suggests the satellite data during the 1990s contains spurious warming of
unknown origin, revealing itself least in UAH data due to its intersatellite trend-adjust-
ment process.
5. Summary
We performed this intercomparison study so as to document dierences among the four
microwave satellite temperature datasets of the bulk atmospheric layer known as T
MT
.
While all datasets indicated high levels of agreement with independent data, UAH and
RSS tended, in broad terms, to exhibit higher levels of agreement than NOAA and UW.
This conclusion does not apply however to the test-statistic of the trend where UAH
tended to agree most closely with independent datasets.
One key result here is that substantial evidence exists to show that the processed data
from NOAA-12 and 14 (operating in the 1990s) were aected by spurious warming that
impacted the four datasets, with UAH the least aected due to its unique merging process.
RSS, NOAA and UW show considerably more warming in this period than UAH and more
than the US VIZ and Australian radiosondes for the period in which the radiosonde
instrumentation did not change. Additionally the same discrepancy was found relative to
the composite of all of the radiosondes in the IGRA database, both global and low-latitude.
While not denitive, the evidence does support the hypothesis that the processed satellite
data of NOAA-12 and 14 are characterized by spurious warming, thus introducing spur-
iously positive trends in the satellite records. Comparisons with other, independently-
constructed datasets (radiosonde and reanalyses) support this hypothesis (Figure 10).
Given this result, we estimate the global T
MT
trend is +0.10 ± 0.03°C decade
1
.
The rate of observed warming since 1979 for the tropical atmospheric T
MT
layer,
which we calculate also as +0.10 ± 0.03°C decade
1
, is signicantly less than the average
of that generated by the IPCC AR5 climate model simulations. Because the model trends
are on average highly signicantly more positive and with a pattern in which their
3604 J. R. CHRISTY ET AL.
warmest feature appears in the latent-heat release region of the atmosphere, we would
hypothesize that a misrepresentation of the basic model physics of the tropical hydro-
logic cycle (i.e. water vapour, precipitation physics and cloud feedbacks) is a likely
candidate.
Acknowledgments
We thank L. Haimberger (UWien) for examining and conrming the breakpoint issue in Australian
radiosondes in 2009. We also thank R. McKitrick (U Guelph) for performing the signicance tests
given in Table 3. This research was supported by the Department of Energy (DE-SC0012638) and
the Alabama Oce of the State Climatologist. Finally we thank the editor if IJRS for carefully
considering the many issues discussed in this paper.
Disclosure statement
No potential conict of interest was reported by the authors.
Funding
This work was supported by the Alabama Oce of State Climatologist and the U.S. Department of
Energy [DE-SC0005330];
References
Bosilovich, M. G., F. R. Robertson, L. Takacs, and A. Molod. 2017.Atmospheric Water Balance and
Variability in the MERRA-2 Reanalysis.Journal of Climate 30: 11771196. doi:10.1175/jcli-d-16-
0338.1.
Christy, J. 2017.[Global Climate] Lower and Mid-Tropospheric Temperature [In State of the
Climate 2016].The Bulletin of the American Meteorological Society 98 (8): S16S17.
doi:10.1175/2017BAMSStateoftheClimate.1.
Christy, J. R., and R. T. McNider. 2017.Satellite Bulk Tropospheric Temperatures as a Metric for
Climate Sensitivity.Asia-Pac Journal Atmos Sciences 53 (4): 18. doi:10.1007/s13143=017-
0070-z.
Christy, J. R., and W. B. Norris. 2006.Satellite and VIZ-Radiosonde Intercomparisons for Diagnosis
on Non-Climatic Inuences.Journal Atmos Oc Technical 23: 11811194. doi:10.1175/
JTECH1937.1.
Christy, J. R., and W. B. Norris. 2009.Discontinuity Issues with Radiosondes and Satellite
Temperatures in the Australian Region 1979-2006.Journal Atmos Oc Technical 26: 508522.
doi:10.1175/2008JTECHA1126.1.
Christy, J. R., W. B. Norris, and R. T. McNider. 2009.Surface Temperature Variations in East Africa
and Possible Causes.Journal of Climate 22: 33423356. doi:10.1175/2008JCLI2726.1.
Christy, J. R., W. B. Norris, K. Redmond, and K. Gallo. 2006.Methodology and Results of Calculating
Central California Surface Temperature Trends: Evidence of Human-Induced Climate Change?
The Journal of Climate 19: 548563. doi:10.1175/JCLI3627.1.
Christy, J. R., R. W. Spencer, and W. B. Norris. 2011.The Role of Remote Sensing in Monitoring
Global Bulk Tropospheric Temperatures.International Journal Remote Sens 32: 671685.
doi:10.1080/01431161.2010.517803.
Dee, D. P., S. M. Uppala, A. J. Simmons, P. Berrisford, P. Poli, S. Kobayashi, U. Andrae, M. A.
Balmaseda, G. Balsamo, P. Bauer, P. Bechtold, A. C. M. Beljaars, L. van de Berg, J. Bidlot, N.
Bormann, C. Delsol, R. Dragani, M. Fuentes, A. J. Geer, L. Haimberger, S. B. Healy, H. Hersbach, E.
INTERNATIONAL JOURNAL OF REMOTE SENSING 3605
V. Hólm, L. Isaksen, P. Kållberg, M. Köhler, M. Matricardi, A. P. McNally, B. M. Monge-Sanz, J.-J.
Morcrette, B.-K. Park, C. Peubey, P. de Rosnay, C. Tavolato, J.-N. Thépaut, and F. Vitart. 2011.The
ERA-Interim Reanalysis: Conguration and Performance of the Data Assimilation System.
Quarterly Journal of the Royal Meteorological Society 137: 553597. doi:10.1002/qj.828.
Durre, I., and X. Yin, 2011.Enhancements of the Dataset of Sounding Parameters Derived from the
Integrated Global Radiosonde Archive.23rd conference on climate variability and change,
Seattle, WA, January 25. https://ams.confex.com/ams/91Annual/webprogram/Paper179437.html
Ellingson, R. G., and Y. Fouquart. 1991.The Intercomparison of Radiation Codes in Climate Models:
An Overview.Journal of Geophysical Research 96 (D5): 89258927. doi:10.1029/90JD01618.
Flato, G., J. Marotzke, B. Abiodun, P. Braconnot, S. C. Chou, W. Collins, P. Cox, et al. 2013.Evaluation
of Climate Models.In Climate Change 2013: The Physical Science Basis. Contribution of Working
Group I to the Fifth Assessment Report of the Intergovernmental Panel on Climate Change, edited by
T. F. Stocker, D. Qin, G.-K. Plattner, M. Tignor, S. K. Allen, J. Boschung, A. Nauels, Y. Xia, V. Bex, and
P. M. Midgley. Cambridge, UK: Cambridge University Press.
Temperature Products for Assessing Climate (RATPAC): A New Data Set of Large-Area Anomaly Time
Series.Journal of Geophysical Research 110: D22101. doi:10.1029/2005JD006169.
Haimberger, L. 2007.Homogenization of Radiosonde Temperature Time Series Using Innovation
Statistics.Journal of Climate 20: 13771403. doi:10.175/JCLI4050.1.
Haimberger, L., C. Tavolato, and S. Sperka. 2012.Homogenization of the Global Radiosonde
Temperature Dataset through Combined Comparison with Reanalysis Background Series and
Neighboring Stations.Journal of Climate 25: 81088131. doi:10.1175/jcli-d-11-00668.1.
Hartmann, D. L., A. M. G. Klein Tank, M. Rusticucci, L. V. Alexander, S. Bronnimann, U. Charabi, F. J.
Dentener, et al. 2013.Observations: Atmosphere and Surface.In Climate Change 2013: The
Phyiscal Science Basis. Contribution of Working Group I to the Fifth Assessment Report of the
Intergovernmental Panel on Climate Change, edited by T. F. Stocker, D. Qin, G.-K. Plattner, M.
Tignor, S. K. Allen, J. Boschung, A. Nauels, Y. Xia, V. Bex, and P. M. Midgley. Cambridge, United
Kingdom and New York, NY, USA: Cambridge Universitiy Press.
Kobayashi, S., Y. Ota, Y. Harada, A. Ebita, M. Moriya, H. Onoda, K. Onogi, et al. 2015.The JRA-55
Reanalysis: General Specications and Basic Characteristics.Journal Meteor Social Japan 93: 5
48. doi:10.2151/jmsj.2015-001.
Mears, C. A., and F. J. Wentz. 2016.Sensitivity of Satellite-Derived Tropospheric Temperature Trends
to the Diurnal Cycle Adjustment.Journal of Climate 29: 36293646. doi:10.1175/JCLI-D-15-0744.1.
Mo, T. 2009.A Study of the NOAA-15 AMSU-A Brightness Temperatures from 1998 through 2007.
Journal of Geophysical Research 114D11. doi:10.1029/2008JD011267.
Po-Chedly, S., T. J. Thorsen, and Q. Fu. 2015.Removing Diurnal Cycle Contamination in Satellite-
Derived Tropospheric Temperatures. Understanding Tropical Tropospheric Trend
Discrepancies.Journal of Climate 28: 22742290. doi:10.1175/JCLI-D-13-00767.1.
Rosenkranz, P. W. 1993.Absorption of Microwaves by Atmospheric Gases.In Atmospheric Remote
Sensing by Microwave Radiometry, edited by M. A. Janssen, 572. New York: John Wiley and Sons.
Chapter 2. ISBN # 0471628913. doi:10.1002/joc.3370140912
Santer, B. D., S. Solomon, G. Pallotta, C. Mears, S. Po-Chedley, Q. Fu, F. Wentz, et al. 2017.
Comparing Tropospheric Warming in Climate Models and Satellite Data.Journal of Climate
30 (4): 373392. doi:10.1175/JCLI-D-16-0333.1.
Sherwood, S. C., and N. Nishant. 2015.Atmospheric Changes through 2012 as Shown by
Iteratively Homogenized Radiosonde Temperature and Wind Data (Iukv2).Environment
Researcher Letters 10: 054007. doi:10.1088/1748-9326/10/5/054007.
Spencer, R. W., and J. R. Christy. 1990.Precise Monitoring of Global Temperature Trends from
Satellites.Science 247: 15581562. doi:10.1126/science.247.4950.1558.
Spencer, R. W., J. R. Christy, and W. D. Braswell. 2017.UAH Version 6 Global Satellite Temperature
Products: Methodology and Results. Asia-Pac.Journal Atmos Sciences 53 (1): 110. doi:10.1007/
s13143.
van Oldenbrogh, G. J. 2016. Climate Data Explorer. http://climexp.knmi.nl/selecteld_co2.cgi?
someone@somewhere
3606 J. R. CHRISTY ET AL.
Zhang, Y., and D. J. Seidel. 2011.Challenges in Estimating Trends in Arctic Surface-Based Inversions
from Radiosonde Data.Geophys Researcher Letters 38 (17). doi:10.1029/2011GL048728.
Zou, C.-Z., and W. Wang. 2011.Intersatellite Calibration of AMSU-A Observations for Weather
and Climate Applications.Journal of Geophysical Research 116: D23113. doi:10.1029/
2011JD016205.
INTERNATIONAL JOURNAL OF REMOTE SENSING 3607
... This discontinuity spuriously inflates CESM2 tropical TMT trends by 0.04 KÁdecade À1 , though the magnitude of this effect remains to be quantified in other GCMs. In addition to model forcing biases and internal variability, GCM response errors and residual satellite biases are also likely to contribute to model-versussatellite differences in tropospheric warming (9,(49)(50)(51). ...
Article
Full-text available
Climate-model simulations exhibit approximately two times more tropical tropospheric warming than satellite observations since 1979. The causes of this difference are not fully understood and are poorly quantified. Here, we apply machine learning to relate the patterns of surface-temperature change to the forced and unforced components of tropical tropospheric warming. This approach allows us to disentangle the forced and unforced change in the model-simulated temperature of the midtroposphere (TMT). In applying the climate-model-trained machine-learning framework to observations, we estimate that external forcing has produced a tropical TMT trend of 0.25 ± 0.08 K⋅decade-1 between 1979 and 2014, but internal variability has offset this warming by 0.07 ± 0.07 K⋅decade-1. Using the Community Earth System Model version 2 (CESM2) large ensemble, we also find that a discontinuity in the variability of prescribed biomass-burning aerosol emissions artificially enhances simulated tropical TMT change by 0.04 K⋅decade-1. The magnitude of this aerosol-forcing bias will vary across climate models, but since the latest generation of climate models all use the same emissions dataset, the bias may systematically enhance climate-model trends over the satellite era. Our results indicate that internal variability and forcing uncertainties largely explain differences in satellite-versus-model warming and are important considerations when evaluating climate models.
... Efforts to intercompare and understand differences across tropospheric temperature datasets is ongoing (Steiner et al. 2020). For example, Christy et al. (2018) compares radiosonde and satellitebased measurements of tropospheric temperature and concludes that satellite datasets likely have spurious warming over the 1990s. On the other hand, this analysis also documents unexplained cooling in satellite datasets over the early 2000s. ...
Article
Full-text available
... Efforts to intercompare and understand differences across tropospheric temperature datasets is ongoing (Steiner et al. 2020). For example, Christy et al. (2018) compares radiosonde and satellitebased measurements of tropospheric temperature and concludes that satellite datasets likely have spurious warming over the 1990s. On the other hand, this analysis also documents unexplained cooling in satellite datasets over the early 2000s. ...
Article
Full-text available
Editors note: For easy download the posted pdf of the State of the Climate in 2021 is a low-resolution file. A high-resolution copy of the report is available by clicking here . Please be patient as it may take a few minutes for the high-resolution file to download.
Preprint
Full-text available
The Coupled Model Intercomparison Project (phase 6) (CMIP6) global circulation models (GCMs) predict equilibrium climate sensitivity (ECS) values ranging between 1.8 and 5.7 $^\circ$C. To narrow this range, we group 38 GCMs into low, medium and high ECS subgroups and test their accuracy and precision in hindcasting the mean global surface warming observed from 1980-1990 to 2011-2021 in the ERA5 T2m, HadCRUT5, GISTEMP v4, and NOAAGlobTemp v5 global surface temperature records. We also compare the GCM hindcasts to the satellite-based UAH MSU v6 lower troposphere global temperature record. We use 143 GCM ensemble averaged simulations under four slightly different forcing conditions, 688 GCM member simulations, and Monte Carlo modeling of the internal variability of the GCMs under three different model accuracy requirements. We found that the medium and high ECS GCMs run too hot up to over 95% and 97% of cases, respectively. The low ECS GCM group agrees best with the warming values obtained from the surface temperature records, ranging between 0.52 and 0.58 $^\circ$C. However, when comparing the observed and GCM hindcasted warming on land and ocean regions, the surface-based temperature records appear to exhibit a significant warming bias. If the satellite based UAH MSU lt record is accurate, actual surface warming from 1980 to 2021 may have been around 0.40 $^\circ$C (or less), i.e. up to about 30% less than what is reported by the surface based temperature records. The latter situation implies that even the low ECS models would have produced excessive warming from 1980 to 2021. These results suggest that the actual ECS may be relatively low, i.e. lower than 3 $^\circ$C or even less than 2 $^\circ$C. Therefore, the projected global climate warming over the next few decades could be moderate and probably not particularly alarming.
Article
Full-text available
The Coupled Model Intercomparison Project (phase 6) (CMIP6) global circulation models (GCMs) predict equilibrium climate sensitivity (ECS) values ranging between 1.8 and 5.7 $${^\circ }$$ ∘ C. To narrow this range, we group 38 GCMs into low, medium and high ECS subgroups and test their accuracy and precision in hindcasting the mean global surface warming observed from 1980–1990 to 2011–2021 in the ERA5-T2m, HadCRUT5, GISTEMP v4, and NOAAGlobTemp v5 global surface temperature records. We also compare the GCM hindcasts to the satellite-based UAH-MSU v6 lower troposphere global temperature record. We use 143 GCM ensemble averaged simulations under four slightly different forcing conditions, 688 GCM member simulations, and Monte Carlo modeling of the internal variability of the GCMs under three different model accuracy requirements. We found that the medium and high-ECS GCMs run too hot up to over 95% and 97% of cases, respectively. The low ECS GCM group agrees best with the warming values obtained from the surface temperature records, ranging between 0.52 and 0.58 $${^\circ }$$ ∘ C. However, when comparing the observed and GCM hindcasted warming on land and ocean regions, the surface-based temperature records appear to exhibit a significant warming bias. Furthermore, if the satellite-based UAH-MSU-lt record is accurate, actual surface warming from 1980 to 2021 may have been around 0.40 $${^\circ }$$ ∘ C (or less), that is up to about 30% less than what is reported by the surface-based temperature records. The latter situation implies that even the low-ECS models would have produced excessive warming from 1980 to 2021. These results suggest that the actual ECS may be relatively low, i.e. lower than 3 $${^\circ }$$ ∘ C or even less than 2 $${^\circ }$$ ∘ C if the 1980–2021 global surface temperature records contain spurious warming, as some alternative studies have already suggested. Therefore, the projected global climate warming over the next few decades could be moderate and probably not particularly alarming.
Article
5-page manuscript plus 5 figures. ABSTRACT: Proving the Sun's strong influence on climate change, published graphs of 1) global temperature fluctuations and 2) solar-magnetic output of the last 9,000 years are strikingly alike. Cross-matching of major spikes (decades-long peaks and troughs) reveals a ~150-year temperature lag, ascribed here to ocean thermal inertia. Non-matching spikes are also numerous: cold ones attributable to mega-eruptions ('volcanic winters'); and warm ones probably due to surges in submarine volcanism and/or hydrothermal venting. The graphs clearly decouple in the 20th Century: post-1940 warmth (NASA-HadCRUT charts) is disproportionately high for the corresponding solar output (150 years earlier), implicating either A) a submarine heat surge or B) one or more of three potential global-warming drivers produced by humans burning fossil fuels and wood, i.e. carbon dioxide (CO2), waste heat, and particulate black carbon (BC). Exonerating submarine heating, post-1940 temperature inflections (change points) correlate with abrupt changes in world annual consumption of coal (10-year lag) and oil. Further condemning coal, ~10 years after its usage suddenly doubled in 1974 (First Oil Shock), in 1985 terrestrial air oddly began warming twice as fast as the sea surface (NASA-HadCRUT; contrast previous lockstep). This preferential land warming is not attributable to CO2 as, even in heavily industrial regions, atmospheric CO2 is <0.5% above the global average (now 415 ppm). Nor can rising CO2 explain long temperature hiatuses (1957-75; 1998-2012). Evidently CO2's net 'greenhouse effect' is near zero, possibly negative due to poorly known feedbacks. Thus 'carbon capture' is misconceived. It is also counterproductive, e.g. if CO2 emissions were allowed to grow until 2080 and then slowly decline, atmospheric CO2 would stabiliize at ~750 ppm in ~2140 (IPCC scenario RPC6.0), near the agricultural optimum of ~1000 ppm. Given CO2's innocence, the coal link (above) incriminates BC and/or waste heat. Substituting nuclear power for coal will reduce the BC threat, but not the waste-heat menace. Solar-magnetic output rose an extraordinary 130% from 1901 to 1992, to its highest level in >9,000 years. Corresponding warming ~150 years later (lag) will add to man-made warming, unless world energy usage (proxy for BC and waste heat) stops growing by then. Simply reducing annual energy growth to zero (cf. 5% in 1960, 2% in 2020) would halt anthropogenic global warming within ~10 years (coal lag-time). PLEASE CLICK 'REQUEST FULL TEXT'.
Preprint
Full-text available
The equilibrium climate sensitivity (ECS) of the CMIP6 global circulation models (GCMs) varies from 1.83 {\deg}C to 5.67 {\deg}C. Herein, 38 GCMs are grouped into three ECS classes (low, 1.80-3.00 {\deg}C; medium, 3.01-4.50 {\deg}C; high, 4.51-6.00 {\deg}C) and compared against the ERA5-T2m records from 1980-1990 to 2011-2021. We found that all models with ECS > 3.0 {\deg}C overestimate the observed global surface warming and that spatial t-statistics rejects the data-model agreement over 60% (using low-ECS GCMs) to 81% (using high-ECS GCMs) of the Earth's surface. Thus, the high and medium-ECS GCMs are unfit for prediction purposes. The low-ECS GCMs are not fully satisfactory yet, but they are found unalarming because by 2050 they predict a moderate warming ($\Delta T_{preindustrial\rightarrow2050}\lesssim2\:{^\circ}C$).
Article
Full-text available
Plain Language Summary The last‐generation Coupled Model Intercomparison Projects (CMIP6) global circulation models (GCMs) are used by scientists and policymakers to interpret past and future climatic changes and to determine appropriate (adaptation or mitigation) policies to optimally address scenario‐related climate‐change hazards. However, these models are affected by large uncertainties. For example, their equilibrium climate sensitivity (ECS) varies from 1.83°C to 5.67°C, which makes their 21st‐century predicted warming levels very uncertain. This issue is here addressed by testing the GCMs' global and local performance in predicting the 1980–2021 warming rates against the ERA5‐T2m records and by grouping them into three equilibrium climate sensitivity (ECS) classes (low‐ECS, 1.80–3.00°C; medium‐ECS, 3.01–4.50°C; high‐ECS, 4.51–6.00°C). We found that: (a) all models with ECS > 3.0°C overestimate the observed global surface warming; (b) Student t‐tests show model failure over 60% (low‐ECS) to 81% (high‐ECS) of the Earth's surface. Thus, the high and medium‐ECS GCMs do not appear to be consistent with the observations and should not be used for implementing policies based on their scenario forecasts. The low‐ECS GCMs perform better, although not optimally; however, they are also found unalarming because for the next decades they predict moderate warming: ΔTpreindustrial→2050 ≲ 2°C.
Article
Full-text available
Editor’s note: For easy download the posted pdf of the State of the Climate for 2017 is a low-resolution file. A high-resolution copy of the report is available by clicking here. Please be patient as it may take a few minutes for the high-resolution file to download.
Chapter
Full-text available
Climate models have continued to be developed and improved since the AR4, and many models have been extended into Earth System models by including the representation of biogeochemical cycles important to climate change. These models allow for policy-relevant calculations such as the carbon dioxide (CO2) emissions compatible with a specified climate stabilization target. In addition, the range of climate variables and processes that have been evaluated has greatly expanded, and differences between models and observations are increasingly quantified using ‘performance metrics’. In this chapter, model evaluation covers simulation of the mean climate, of historical climate change, of variability on multiple time scales and of regional modes of variability. This evaluation is based on recent internationally coordinated model experiments, including simulations of historic and paleo climate, specialized experiments designed to provide insight into key climate processes and feedbacks and regional climate downscaling. Figure 9.44 provides an overview of model capabilities as assessed in this chapter, including improvements, or lack thereof, relative to models assessed in the AR4. The chapter concludes with an assessment of recent work connecting model performance to the detection and attribution of climate change as well as to future projections.
Article
Full-text available
Temperature sounding microwave radiometers flown on polar-orbiting weather satellites provide a long-term, global-scale record of upper-atmosphere temperatures, beginning in late 1978 and continuing to the present. The focus of this paper is the midtropospheric measurements made by the Microwave Sounding Unit (MSU) channel 2 and the Advanced Microwave Sounding Unit (AMSU) channel 5. Previous versions of the Remote Sensing Systems (RSS) dataset have used a diurnal climatology derived from general circulation model output to remove the effects of drifting local measurement time. This paper presents evidence that this previous method is not sufficiently accurate and presents several alternative methods to optimize these adjustments using information from the satellite measurements themselves. These are used to construct a number of candidate climate data records using measurements from 15 MSU and AMSU satellites. The new methods result in improved agreement between measurements made by different satellites at the same time. A method is chosen based on an optimized second harmonic adjustment to produce a new version of the RSS dataset, version 4.0. The new dataset shows substantially increased global-scale warming relative to the previous version of the dataset, particularly after 1998. The new dataset shows more warming than most other midtropospheric data records constructed from the same set of satellites. It is also shown that the new dataset is consistent with long-term changes in total column water vapor over the tropical oceans, lending support to its long-term accuracy.
Article
Full-text available
We present an updated version of the radiosonde dataset homogenized by Iterative Universal Kriging (IUKv2), now extended through February 2013, following the method used in the original version (Sherwood et al 2008 Robust tropospheric warming revealed by iteratively homogenized radiosonde data J. Clim. 21 5336–52). This method, in effect, performs a multiple linear regression of the data onto a structural model that includes both natural variability, trends, and time-changing instrument biases, thereby avoiding estimation biases inherent in traditional homogenization methods. One modification now enables homogenized winds to be provided for the first time. This, and several other small modifications made to the original method sometimes affect results at individual stations, but do not strongly affect broad-scale temperature trends. Temperature trends in the updated data show three noteworthy features. First, tropical warming is equally strong over both the 1959–2012 and 1979–2012 periods, increasing smoothly and almost moist-adiabatically from the surface (where it is roughly 0.14 K/decade) to 300 hPa (where it is about 0.25 K/decade over both periods), a pattern very close to that in climate model predictions. This contradicts suggestions that atmospheric warming has slowed in recent decades or that it has not kept up with that at the surface. Second, as shown in previous studies, tropospheric warming does not reach quite as high in the tropics and subtropics as predicted in typical models. Third, cooling has slackened in the stratosphere such that linear trends since 1979 are about half as strong as reported earlier for shorter periods. Wind trends over the period 1979–2012 confirm a strengthening, lifting and poleward shift of both subtropical westerly jets; the Northern one shows more displacement and the southern more intensification, but these details appear sensitive to the time period analysed. There is also a trend toward more easterly winds in the middle and upper troposphere of the deep tropics.
Article
Independent research teams have constructed long-term tropical time series of the temperature of the middle troposphere (TMT) using satellite Microwave Sounding Unit (MSU) and Advanced MSU (AMSU) measurements. Despite careful efforts to homogenize the MSU/AMSU measurements, tropical TMT trends beginning in 1979 disagree by more than a factor of 3. Previous studies suggest that the discrepancy in tropical TMT trends is caused by differences in both the NOAA-9 warm target factor and diurnal drift corrections. This work introduces a new observationally based method for removing biases related to satellite diurnal drift. Over land, the derived diurnal correction is similar to a general circulation model (GCM) diurnal cycle. Over ocean, the diurnal corrections have a negligible effect on TMT trends, indicating that oceanic biases are small. It is demonstrated that this method is effective at removing biases between coorbiting satellites and biases between nodes of individual satellites. Using a homogenized TMT dataset, the ratio of tropical tropospheric temperature trends relative to surface temperature trends is in accord with the ratio from GCMs. It is shown that bias corrections for diurnal drift based on a GCM produce tropical trends very similar to those from the observationally based correction, with a trend difference smaller than 0.02 K decade−1. Differences between various TMT datasets are explored further. Large differences in tropical TMT trends between this work and that of the University of Alabama in Huntsville (UAH) are attributed to differences in the treatment of the NOAA-9 target factor and the diurnal cycle correction.
Article
Closing and balancing Earth's global water cycle remains a challenge for the climate community. Observations are limited in duration, global coverage, and frequency, and not all water cycle terms are adequately observed. Reanalyses aim to fill the gaps through the assimilation of as many atmospheric water vapor observations as possible. Former generations of reanalyses have demonstrated a number of systematic problems that have limited their use in climate studies, especially regarding low-frequency trends. This study characterizes the NASA Modern-Era Retrospective Analysis for Research and Applications version 2 (MERRA-2) water cycle relative to contemporary reanalyses and observations. MERRA-2 includes measures intended to minimize the spurious global variations related to inhomogeneity in the observational record. The global balance and cycling of water from ocean to land is presented, with special attention given to the water vapor analysis increment and the effects of the changing observing system. While some systematic regional biases can be identified, MERRA-2 produces temporally consistent time series of total column water and transport of water from ocean to land. However, the interannual variability of ocean evaporation is affected by the changing surface-wind-observing system, and precipitation variability is closely related to the evaporation. The surface energy budget is also strongly influenced by the interannual variability of the ocean evaporation. Furthermore, evaluating the relationship of temperature and water vapor indicates that the variations of water vapor with temperature are weaker in satellite data reanalyses, not just MERRA-2, than determined by observations, atmospheric models, or reanalyses without water vapor assimilation.
Article
Updated and improved satellite retrievals of the temperature of the mid-to-upper troposphere (TMT) are used to address key questions about the size and significance of TMT trends, agreement with model-derived TMT values, and whether models and satellite data show similar vertical profiles of warming. A recent study claimed that TMT trends over 1979 and 2015 are 3 times larger in climate models than in satellite data but did not correct for the contribution TMT trends receive from stratospheric cooling. Here, it is shown that the average ratio of modeled and observed TMT trends is sensitive to both satellite data uncertainties and model-data differences in stratospheric cooling. When the impact of lower-stratospheric cooling on TMT is accounted for, and when the most recent versions of satellite datasets are used, the previously claimed ratio of three between simulated and observed near-global TMT trends is reduced to approximately 1.7. Next, the validity of the statement that satellite data show no significant tropospheric warming over the last 18 years is assessed. This claim is not supported by the current analysis: in five out of six corrected satellite TMT records, significant global-scale tropospheric warming has occurred within the last 18 years. Finally, long-standing concerns are examined regarding discrepancies in modeled and observed vertical profiles of warming in the tropical atmosphere. It is shown that amplification of tropical warming between the lower and mid-to-upper troposphere is now in close agreement in the average of 37 climate models and in one updated satellite record.
Article
The Japan Meteorological Agency (JMA) conducted the second Japanese global atmospheric reanalysis, called the Japanese 55-year Reanalysis or JRA-55. It covers the period from 1958, when regular radiosonde observations began on a global basis. JRA-55 is the first comprehensive reanalysis that has covered the last half-century since the European Centre for Medium-Range Weather Forecasts 45-year Reanalysis (ERA-40), and is the first one to apply four-dimensional variational analysis to this period. The main objectives of JRA-55 were to address issues found in previous reanalyses and to produce a comprehensive atmospheric dataset suitable for studying multidecadal variability and climate change. This paper describes the observations, data assimilation system, and forecast model used to produce JRA-55 as well as the basic characteristics of the JRA-55 product. JRA-55 has been produced with the TL319 version of JMA’s operational data assimilation system as of December 2009, which was extensively improved since the Japanese 25-year Reanalysis (JRA-25). It also uses several newly available and improved past observations. The resulting reanalysis products are considerably better than the JRA-25 product. Two major problems of JRA-25 were a cold bias in the lower stratosphere, which has been diminished, and a dry bias in the Amazon basin, which has been mitigated. The temporal consistency of temperature analysis has also been considerably improved compared to previous reanalysis products. Our initial quality evaluation revealed problems such as a warm bias in the upper troposphere, large upward imbalance in the global mean net energy fluxes at the top of the atmosphere and at the surface, excessive precipitation over the tropics, and unrealistic trends in analyzed tropical cyclone strength. This paper also assesses the impacts of model biases and changes in the observing system, and mentions efforts to further investigate the representation of low-frequency variability and trends in JRA-55.
Article
A procedure is described to construct time series of regional surface temperatures and is then applied to interior central California stations to test the hypothesis that century-scale trend differences between irrigated and nonirrigated regions may be identified. The procedure requires documentation of every point in time at which a discontinuity in a station record may have occurred through (a) the examination of metadata forms (e.g., station moves) and (b) simple statistical tests. From this "homogeneous segments" of temperature records for each station are defined. Biases are determined for each segment relative to all others through a method employing mathematical graph theory. The debiased segments are then merged, forming a complete regional time series. Time series of daily maximum and minimum temperatures for stations in the irrigated San Joaquin Valley (Valley) and nearby nonirrigated Sierra Nevada (Sierra) were generated for 1910-2003. Results show that twentieth-century Valley minimum temperatures are warming at a highly significant rate in all seasons, being greatest in summer and fall (> +0.25°C decade-1). The Valley trend of annual mean temperatures is +0.07° ± 0.07°C decade-1. Sierra summer and fall minimum temperatures appear to be cooling, but at a less significant rate, while the trend of annual mean Sierra temperatures is an unremarkable -0.02° ± 0.10°C decade-1. A working hypothesis is that the relative positive trends in Valley minus Sierra minima (>0.4°C decade-1 for summer and fall) are related to the altered surface environment brought about by the growth of irrigated agriculture, essentially changing a high-albedo desert into a darker, moister, vegetated plain.