Retrieval of upper tropospheric humidity data from microwave measurements

(1)

Retrieval of upper tropospheric

humidity data from microwave

measurements

Vom Fachbereich f¨ur Physik und Elektrotechnik

der Universit¨at Bremen

Zur Erlangung des akademischen Grades eines

Doktor der Naturwissenschaften (Dr. rer. nat.)

genehmigte Dissertation

von

M.Sc. Phys. Mashrab Kuvatov

1. Gutachter:

Prof. Dr. rer. nat. J. Notholt

(2)

(3)

Abstract

The current thesis shows three sub-studies in order to analyze the upper tropo-spheric humidity (UTH) from microwave measurements. First, a simple method of comparing the symmetric measurements across the scan-line revealed that the asymmetry errors in the AMSU-B Channel 18 for NOAA-16 and 17 are within the instrument noise temperatures. However, for the same AMSU-B channel, but on NOAA-15 for the recent data the asymmetry errors exceed the instru-ment noise temperature. The asymmetry errors for Channel 18 are less than 1.90, -0.53, and 0.49 K for NOAA-15, 16, and 17, respectively.

Fairly large asymmetry errors were found in the AMSU-B Channel 17, 19, and 20 on NOAA-15. This seems to be related to the known radio frequency interference (RFI) problem for this instrument. If the appropriate set of the recent RFI correction coefficients are used, the errors due to asymmetry are significantly reduced.

Second, this thesis also presents a cloud filtering method for the UTH mea-surements at 183.31±1.00 GHz. This method combines two known cloud de-tection techniques. Namely, it utilizes the difference between the brightness temperatures at 183.31±7.00 and 183.31±1.00 GHz, and a threshold for the brightness temperature at 183.31±1.00 GHz. The robustness of this cloud filter is demonstrated by a mid-latitudes winter case-study.

Studies on possible cloud effects on the microwave UTH climatology also presented. Using cloud contaminated measurements will overestimate the de-rived UTH, and not using such measurements will underestimate UTH, since clouds are associated with high humidity. These biases due to clouds are esti-mated and compared. The simulations of a cloud event and cloud database with a reasonable statistics for the midlatitude conditions are used.

The consistent result is that both cloud wet bias (0.8 %RH) and cloud filtering dry bias (-2.4 %RH) are modest for microwave data, where the numbers are from the cloud database analysis. For the case study the biases are slightly higher, -3% and 2%, respectively, because the studied case was a particularly strong ice cloud event. This indicates that for microwave data the cloud-filtered UTH and unfiltered UTH can be taken as error bounds for errors due to clouds. This is not possible for the more traditional infrared data, since the radiative effect of clouds is much stronger there.

(6)

at hand, the effects of the cloud filter only on the midlatitude data are discussed. However, those effects are expected to be valid for subtropical and tropical data, too.

Finally, the studies on UTH retrieved from 183.31±1.00 GHz radiances of the three operational microwave sounders are presented. The long-term UTH climatology data show similar patterns and features known from the meteorol-ogy and other available UTH data. Most of the high humidity values were found around the intertropical convergence zone (ITCZ). The inter-satellite calibration is investigated by comparing different sensors in the UTH space.

A study of the zonal averaged UTH showed that there are two maxima around the ITCZ in the annual UTH cycle. They coincide with the shift of the Indian Monsoon system around the ITCZ. A seasonal variation in UTH data for differ-ent latitude bands is also studied. It shows that the Northern hemisphere is more humid than its Southern counterpart. Also, the seasonal cycle of UTH is more pronounced in the Northern than in the Southern hemisphere.

A trend analysis is conducted by using a simple linear fit method. It revealed that there are visible trends in UTH on a regional scale. However, the trend values are contradictive. Also a time period of the available data is relatively short. Therefore, it is yet too early to make conclusive analysis of the UTH trends.

(7)

Publications

Journal Articles

1. Buehler, S. A., M. Kuvatov, V. O. John, M. Milz, B. J. Soden and J. Notholt (submitted 2007),

An upper tropospheric humidity data set from operational satellite microwave data,

J. Geophys. Res.

2. Buehler, S. A., M. Kuvatov, T. R. Sreerekha, V. O. John, B. Rydberg, P. Eriksson and J. Notholt (2007),

A cloud filtering method for microwave upper tropospheric humidity measurements,

Atmos. Chem. Phys. Discuss.. Online version is available from

http://www.atmos-chem-phys-discuss.net/7/7509/2007/ acpd-7-7509-2007.html

3. Buehler, S. A., M. Kuvatov, V. O. John (2005), Scan asymmetries in AMSU-B data,

Geophys. Res. Lett., 32, L24810, doi:10.1029/2005GL024747.

4. Buehler, S. A., M. Kuvatov, V. O. John, U. Leiterer and H. Dier (2004), Comparison of microwave satellite humidity data and radiosonde profiles: A case study,

J. Geophys. Res., 109, D13103, doi:10.1029/2004JD004605.

Articles in Conference Proceedings

5. John, V. O., S. A. Buehler, M. Kuvatov, B. J. Soden and T. R. Sreerekha (2006),

Toward a long-term homogenized UTH data set derived from satellite microwave measurements,

In: Proceedings of SPIE Conference on Microwave Remote Sensing of the Atmosphere and Environment V, November 13-17, 2006, Goa, India, doi:10.1117/12.694305.

(8)

6. John, V. O., S. A. Buehler and M. Kuvatov (2003),

Comparison of AMSU-B Brightness Temperature with Simulated Brightness Temperature using Global Radiosonde Data,

In: Thirteenth International TOVS Study Conference (ITSC - XIII), St. Adele, Montreal, Canada.

(9)

1 Introduction

I more than agree with Soden et al. [2005] who writes that the importance of wa-ter vapor in regulating climate is undisputed. Wawa-ter vapor is one of the most ra-diatively active gases in the atmosphere. It traps the re-emitted radiation, which originally comes from the Sun, so that the Earth stays warm.

Recently, Buehler et al. [2006b] studied the sensitivity of outgoing long-wave radiation to changes in humidity, carbon dioxide concentrations, and tempera-ture. They concluded that for the tropical scenario a 20% change in humidity has a larger impact than a doubling of the carbon dioxide concentration. This is just one example of how important water vapor in the atmosphere is.

The horizontal and vertical distributions of water vapor in the atmosphere are very variable. While the deep tropics (10◦N–10◦S latitude band) are very humid, the subtropics (25◦N–30◦N and 25◦S–30◦S latitude bands) are very dry. There is a strong longitudinal variability in the water vapor distribution, too.

The free troposphere holds significant part of the water vapor in the atmo-sphere. Its concentration beyond the troposphere is very small. Most of the water vapor is concentrated in the low troposphere. In the upper troposphere the concentration of water vapor is the least. So, if so little of the water vapor is in the upper troposphere, why are the scientists interested in the upper tropospheric humidity (UTH)?

As Soden et al. [2005] write, since the absorptivity of water vapor is propor-tional to the logarithm of its concentration, it is the fracpropor-tional change in water vapor mass, not the absolute change, that governs its strength as a feedback mechanism. Since there is very little water vapor in the upper troposphere, even a little change of UTH is crucial for the atmospheric processes linked to the water vapor. For instance, Schneider et al. [1999] concluded that extratropical water vapor feedback affected warming twice as much as that of tropical feed-back did. This can be explained by the fact that in general extratopics are dryer than tropics.

Poor knowledge of UTH and its change over the last decades resulted in the long ongoing dispute and controversy in the scientific community over the role of UTH in the global warming studies. While there is now a general agree-ment in the scientific community on the basic role and sign of the water vapor feedback [Held and Soden, 2000], we still need a more precise quantitative un-derstanding of this important aspect of the climate system, which can only be

(10)

achieved through long-term and accurate humidity measurements.

The positive feedback of water vapor was proposed by many scientists. It was suggested that water vapor feedback alone compared to the fixed water vapor concentration approximately doubles the warming [Schneider et al., 1999; Hall and Manabe, 1999; Held and Soden, 2000]. Moreover, water vapor feedback can amplify cloud and ice albedo feedbacks in models. For example, Hall and Manabe [1999] report that if there is a strongly positive cloud feedback, the water vapor feedback can lead to 3.5 times as much warming as compared to the fixed water vapor concentration.

An alternative view to this issue proposes the negative feedback of water vapor. For instance, Lindzen et al. [2001] suggested an adaptive infrared iris, where decreasing areas of high clouds and high humidity with increasing sur-face temperature would result in the negative feedback. In other words, this means that increasing surface temperature would lead to a reduced supply of water vapor into upper troposphere. However, this view has not been supported by convincing experimental or modeling evidence.

Minschwaner and Dessler[2004] studied a response of the tropical upper tro-pospheric water vapor to the increasing surface temperatures. The main conclu-sion of this study is that the relative humidity for the observed data is decreasing. Supporting argument for this conclusion is that the increase in mixing ratio is not as large as the increase in saturation mixing ratio in the upper troposphere. This would suggest a weak UTH feedback.

Recently, Su et al. [2006] suggested that there is an enhanced positive water vapor feedback because of moistening of the upper troposphere by tropical deep convection. It is about 3 times that implied only by thermodynamics. However, John and Soden[2006] continued in this direction and showed that, in fact, there is no such enhanced positive water vapor feedback.

I think that all these discussions show the need of as accurate as possible long-term UTH data available to a wide scientific community. One of the sources of humidity data are radiosonde measurements. Although they provide a long-term record of humidity with good vertical resolution, there are a few drawbacks in using them in global UTH studies. First of all, they do not provide a global coverage and most of the data are available only over land in the Northern hemi-sphere. They also underestimate the humidity at high altitudes. Moreover, the changes in the technology used and poor calibration does not allow one to use the radiosonde data in long-term studies of UTH [Elliot and Gaffen, 1991]. As Soden et al. [2005] write, due to changes in data quality and data coverage, global reanalysis data are also not usable for long-term UTH studies.

One alternative is to use existing satellite infrared (IR) measurements to derive humidity data [Soden and Bretherton, 1996]. Since clouds are not transparent to IR radiation, the usual practice is to remove cloud contaminated measurement

(11)

1 Introduction 7

from the analysis. This practice, however, will introduce a systematic clear-sky bias to the final UTH product [Lanzante and Gahrs, 2000].

We propose to use available measurements from satellite microwave sensors for the long-term UTH studies. One important advantage of microwave over IR remote sensing is that the clouds are in general transparent to microwave radiation. In this thesis I will demonstrate the possibility of retrieving UTH from one of the available microwave sensors. If the proposed methodology will be proven to give the reasonable results, the study can be extended for more than a decade humidity from the existing microwave measurements. I believe such study will shed more light on the water vapor feedback and long-term water vapor trends.

The structure of this thesis is as follows. In Chapter 2 an overview of the radiative transfer models used in the studies presented in this thesis will be given. In the next Chapter 3 the basics of the instrument from which the measurements are used will be given. Information on its channels, their sensitive altitudes, etc. will be presented.

This thesis presents a number of studies which address several important is-sues that need to be solved prior to the UTH retrievals. The first important issue is a possible asymmetry in the measurements. If present, this can introduce a systematic bias into the final UTH products. Therefore, this possible bias should be estimated. Chapter 4 of this thesis describes how this issue was investigated. It shows, and discusses the results.

The second important issue is about the possible cloud impact on the final UTH products. High ice clouds are expected to introduce a positive bias into UTH. However, filtering them out is expected to introduce a negative bias into UTH. In Chapter 5 we will discuss this issue and show the results of the study on it.

Finally, the UTH retrieved from the microwave measurements will be shown in Chapter 6. The long-term UTH, the seasonal variation of UTH, and a simple trend analysis of UTH will be presented and discussed.

In Chapter 7 the studies presented in this thesis will be summarized and the conclusions of overall study of UTH from the microwave measurements will be given.

(12)

(13)

2 Radiative Transfer Models

This chapter will give a brief introduction to the radiative transfer (RT) models used in the studies presented in this thesis. The aim of this chapter is to give an overview of the RT models. For detailed information please have a look at the corresponding referenced documents.

2.1 ARTS Model

The Atmospheric Radiative Transfer Simulator (ARTS) is a computer program, from now on the ARTS model, built to simulate atmospheric radiative transfer. It is the result of several years of collaboration between Bremen and Chalmers universities. At the moment, there are many more active collaborators and users of this model. The ARTS model is developed under the terms of the GNU general public license [Stallman, 2002], which means that anyone is granted to freely use, modify and distribute the software as long as the source code is freely available.

ARTS is a general purpose model which was designed with modularity and extendability in mind. Because of this, it can simulate almost any kind of in-strument for all kinds of observation geometry. The observation geometries it can handle are up, down, and limb looking. The sensor can be placed inside (for example, sensor on the aircraft) or outside (for example, ground based sensor or sensor on board satellite) the atmosphere.

The ARTS model is capable of handling any given frequency grid. Thus, it can be used to simulate high and low spectral resolution sensors. The frequency range the ARTS model can work with is from the microwave to the thermal infrared.

Absorption models of Liebe [1989] and Rosenkranz [1993] are fully inte-grated into the ARTS model. Detailed information on the absorption part of the model can be found in Kuhn [2004].

Also, the ARTS model can analytically or semi-analytically calculate Jaco-bians for the different atmospheric parameters. JacoJaco-bians give the change in radiance as a consequence of a change of another variable determining the ra-diative transfer. Those Jacobians are for trace gas concentration, continuum absorption, temperature, and surface emissivity.

(14)

The ARTS is a line-by-line (LBL) model. It means that the calculation is done taking into account the contribution of each absorbing line in a given frequency range. Thus, simulations which consider absorption are computationally very intensive procedures.

The ARTS model exists in two versions. One is for clear-sky simulations [Buehler et al., 2005a]. This version of the model does not take scattering into account, i.e. the effect of the ice particles in the clouds or rain drops is neglected. In this case, the atmosphere is considered to be a one-dimensional where the atmospheric parameters change with only the vertical coordinate. Pressure is taken as vertical coordinate.

This version of the ARTS model was compared to other radiative transfer models [Melsheimer et al., 2005; Buehler et al., 2006a] and to microwave mea-surements [Kuvatov, 2002; Buehler et al., 2004]. It was concluded that the ARTS model is very consistent with other models and measurements.

The next version of the ARTS model, currently under active development, can simulate atmospheric radiative transfer taking into account scattering [Emde, 2005; Sreerekha, 2005; Emde et al., 2004]. In this version of the ARTS a dis-crete ordinate iterative method is used to solve the 3D vector radiative transfer equation. In this case the ARTS model uses spherical three-dimensional coordi-nate system where the dimensions are longitude, latitude, and radius. With this version of the ARTS model it is still possible to simulate 1D atmospheric radia-tive transfer. The first validations showed consistent results with model [Emde, 2005] and microwave measurements [Sreerekha, 2005]. Since this is the work in progress, more thorough validations are underway. More information on the ARTS model can be found at http://www.sat.ltu.se/arts/

2.2 RTTOV Model

The Radiative Transfer for the TIROS Operational Vertical Sounder (RTTOV) [Saunders et al., 1999] is also a computer program built to simulate atmospheric radiative transfer. It is a result of a collaborative effort of several numerical weather prediction (NWP) centers and research centers like MetOffice, Mete-oFrance, European Centre for Medium-Range Weather Forecasts (ECMWF). The RTTOV model is used in more than 40 NWP centers worldwide including major NWP centers, in data assimilation centers like ECMWF, and by a number of scientific groups in the atmospheric research community.

Predefined set-ups for simulations are available for a variety of different sen-sors like AMSU-B, AMSU-A, AVHRR, SSMI, HIRS, MODIS and many more operational instruments.

(15)

2.2 RTTOV Model 11

of the instrument channel is calculated by integrating the LBL model opacities over the instrument channel responses. Apart from this, the computation is done on fixed pressure levels (43). Using the limited, fixed pressure levels speeds up the simulations, too.

The RTTOV uses a built-in emissivity model FASTEM [English and Hewi-son, 1998] which calculates surface emissivities depending on surface type, ob-servation viewing angle and frequency. The surface types include sea, low and high vegetation, snow, and bare soil. It can also do RT simulations for a user defined value for emissivity, if needed.

Wherever the RTTOV is mentioned in this thesis, it implies that the RTTOV-7 version [Saunders, 2002] was used, if not explicitly stated otherwise. Also note that Buehler et al. [2006a] showed that the RTTOV is in fairly good agreement with the more physical model ARTS.

(16)

(17)

3 Instrumentation

This chapter gives the essential information on the instrument used in all studies presented in this thesis. The technical characteristics like channel frequencies, scan angles, spacial resolution and swath width will be covered. Moreover, im-portant physical characteristics of the instrument’s channels, absorption spec-trum and Jacobians will also be presented. The limb effect in all insspec-trument channels will be explained and its estimate will be given.

3.1 The Advanced Microwave Sounding Unit B

The Advanced Microwave Sounding Unit B (AMSU-B) is a cross-track scan-ning, passive, total power microwave radiometer. Together with AMSU-A it composes the AMSU instrument. In the current thesis data from AMSU-A are not used. Therefore, the AMSU-A instrument will not be discussed here.

AMSU-B has two channels centered at 89 and 150 GHz and three channels centered at 183.31 GHz [Saunders et al., 1995]. These channels are referred to as Channel 16 to 20 of the overall AMSU instrument. All channels work in a double sideband mode, which increases overall sensitivity at above frequencies compared to a single sideband mode [Saunders et al., 1995]. The details of the channels are summarized in Table 3.1. Note that the values of the noise equivalent temperature are from the first flight model, not from the instrument operating currently. They might slightly differ for different satellite platforms carrying the AMSU-B instrument.

The instrument has a swath width of approximately 2300 km, which is sam-pled at 90 scan positions. A sketch of the scan geometry is shown in Figure 3.1. The scan positions are defined such that, relative to the flight direction, posi-tion 1 is on the left edge of the scan, posiposi-tions 45 and 46 are in the middle, and position 90 is on the right edge of the scan. The nadir viewing angle for the two innermost scan positions 45 and 46 is 0.55◦, the nadir viewing angle for the two outermost scan positions 1 and 90 is 48.95◦. Due to the Earth’s sphericity, this corresponds to incidence angles of 0.62◦for the innermost scan positions and 58.5◦for the outermost scan positions. This corresponds to a foot-print, target area, size of 20 × 16 km2 at nadir viewing geometry and increased to 64 × 52 km2at outermost off-nadir viewing geometry. The footprint is larger

(18)

Channel Center freq. [GHz] Noise equ. temp. [K] Bandwidth [GHz] 16 89.0±0.9 0.37 2×1 17 150.0±0.9 0.84 2×1 18 183.31±1.0 1.06 2×0.5 19 183.31±3.0 0.7 2×1 20 183.31±7.0 0.6 2×2

Table 3.1: Channel characteristics of AMSU-B instrument. “Noise equ. temp.”

means noise equivalent temperature. The values are taken from

NOAA[2000]

Orbit type Satellite Equator crossing Mean Launch

time (ECT) alt. [km] date

morning NOAA-17 10:24 (D) 810 Jun. 2002

afternoon NOAA-16 14:11 (A) 850 Sep. 2000

early morning NOAA-15 05:58 (D) 807 May 1998

Table 3.2: Characteristics of NOAA satellites. All satellites are

polar-orbiting on a sun-synchronous orbit. In the equator crossing

time column, “A” means ascending (satellite moves northward)

and “D” means descending (satellite moves southward).

In-formation is taken from WMO Space Programme web-page at http://www.wmo.ch/web/sat/POLpresent.html

along the scan line than along the flight direction.

This instrument is operated and the data it delivers are processed by Na-tional Oceanic and Atmospheric Administration (NOAA), USA. It flies on board NOAA-15, 16, and 17 satellites. Launch times of these satellites and other de-tails are given in Table 3.2.

The location of AMSU-B channels in frequency space with respect to atmo-spheric zenith opacity is demonstrated in Figure 3.2. The aim of Channel 16 and 17 is studying the information coming from the surface and the lowest layer of the atmosphere. Therefore, they are located at so-called window frequencies. At those frequencies the opacity of the atmosphere is very low allowing radiation originating from the bottom of the atmosphere to travel through the atmosphere and reach the sensor.

The remaining three channels of AMSU-B were designed to study humidity in the atmosphere. Therefore, they are located at a strong water vapor line at 183.31 GHz. At this frequency the atmosphere is very opaque to the radiation due to absorption of water vapor. Only radiation originating from the upper

(19)

3.1 The Advanced Microwave Sounding Unit B 15

Figure 3.1: AMSU-B scan geometry. Note that there is no measurement at exact nadir viewing angle. Adopted from NOAA [2000].

levels of the troposphere with less humidity and less dense atmospheric layers can reach the sensor. Channel 18 is 1, Channel 19 is 3, and Channel 20 is 7 GHz away from the line center. This separation allows us to receive signals from different layers of the atmosphere.

To know which altitudes the AMSU-B channels are sensitive to, one has to look at the corresponding Jacobians. A Jacobian, which is also called a weight-ing function elsewhere, is formally defined by Rodgers [2000] as follows:,

K = ∂ F(x)

∂ x (3.1)

where F is a forward model, and x is the atmospheric parameter we are interested in. Mathematically, the Jacobian is a derivative, i.e., the change in one variable due to a change in other variable. For example in the case of water vapor, the Jacobian shows the change in brightness temperature given by a forward model due to a change in water vapor concentration at a particular altitude.

The Jacobians for the AMSU-B sounding channels are demonstrated in Fig-ure 3.3. This figFig-ure shows Jacobians for different atmospheric scenarios. Upper and lower rows show Jacobians for a moist tropical and a dry midlatitude winter atmosphere, respectively. Figures from left to right are for AMSU-B channel 18,

(20)

80 100 120 140 160 180 200 Frequency [ GHz ] 10-6 10-4 10-2 100 102 Opacity [ Np ] 16 17 18 19 20 20 H2O O2 N2 Total

Figure 3.2: The zenith opacity of the main constituents of the atmosphere as a function of frequency. Short dashed line - H2O, dotted line - O2,

dash-dotted line - N2, and long dashed line - total opacity. Positions

of the passbands of the AMSU-B channels are indicated by gray bars. Figure from John [2005].

19, and 20. In the figures altitude is plotted against Jacobian in fractional units. The Jacobians were calculated from Chevallier [2001] data using the radiative transfer (RT) model ARTS, which is briefly covered in Section 2.1 of this thesis. First, let us have a look at the upper row of figures. The figure for the AMSU-B Channel 18 shows that the Jacobian peaks at around 8 km. This means that most radiation received by the sensor at this frequency originates from the at-mospheric layers around 8 km. This altitude corresponds approximately to the upper troposphere [Wallace and Hobbs, 1977]. The Jacobian for the AMSU-B Channel 19 has a peak at around 6 km. Thus, this channel provides informa-tion about middle tropospheric layers of the atmosphere. The Jacobian for the AMSU-B Channel 20 peaks at an altitude of less than 5 km. Thus, it is sensitive to the radiation coming from the lower troposphere.

In case of the midlatitudes the atmosphere is drier – contains less water vapor – compared to the tropical atmosphere. Therefore, the atmosphere is less opaque in the frequencies at which AMSU-B sounding channels operate. Thus, the altitudes to which the AMSU-B channels are sensitive in the dry midlatitudes

(21)

3.1 The Advanced Microwave Sounding Unit B 17 -0.8 -0.6 -0.4 -0.2 0.0 0.2 Jacobian [ K / 1 ] 0 2 4 6 8 10 12 14 Altitude [ km ] 18 -0.8 -0.6 -0.4 -0.2 0.0 0.2 Jacobian [ K / 1 ] 0 2 4 6 8 10 12 14 ₁₉ -0.8 -0.6 -0.4 -0.2 0.0 0.2 Jacobian [ K / 1 ] 0 2 4 6 8 10 12 14 ₂₀ -0.8 -0.6 -0.4 -0.2 0.0 0.2 Jacobian [ K / 1 ] 0 2 4 6 8 10 12 14 Altitude [ km ] 18 -0.8 -0.6 -0.4 -0.2 0.0 0.2 Jacobian [ K / 1 ] 0 2 4 6 8 10 12 14 ₁₉ -0.8 -0.6 -0.4 -0.2 0.0 0.2 Jacobian [ K / 1 ] 0 2 4 6 8 10 12 14 ₂₀

Figure 3.3: Water vapor Jacobians for channels 18 to 20 of AMSU-B. The two rows correspond to different atmospheric scenarios. Upper row: Ja-cobians for a moist tropical atmosphere. Lower row: JaJa-cobians for dry midlatitude winter atmosphere. Two kinds of lines correspond to different instrument viewing geometry: solid line - nadir, dashed line - outermost off-nadir viewing angle. Figure from John [2005]. are lower than ones in the moist tropics. This can be clearly seen in the lower row of Figure 3.3, which demonstrates the Jacobians for AMSU-B channels 18 to 20 in midlatitude winter conditions. The Jacobians for all sounding channels are shifted to lower altitudes by about 2 km.

One common feature all AMSU-B sounding channels share is that they are all sensitive to a thick layer of the atmosphere. However, only certain atmospheric layers contribute significantly to the signal the channel sensors receive. Those layers are centered around the Jacobian maxima of the channels. As demon-strated in Figure 3.3 these maxima are channel dependent. Therefore, different AMSU-B channels are sensitive to the atmospheric layers at different altitudes. To summarize, AMSU-B Channel 18, 19, and 20 are sensitive to upper, middle, and lower troposphere, respectively.

The Jacobians for AMSU-B Channel 16 and 17 are presented in Figure 3.4. Like in Figure 3.3, upper and lower rows are for moist tropical and dry midlat-itude atmospheric scenarios. These Jacobians peak at very low altmidlat-itudes. Also,

(22)

-0.5 0.0 0.5 Jacobian [ K / 1 ] 0 2 4 6 8 10 12 14 Altitude [ km ] 16 -0.5 0.0 0.5 Jacobian [ K / 1 ] 0 2 4 6 8 10 12 14 ₁₇ -0.5 0.0 0.5 Jacobian [ K / 1 ] 0 2 4 6 8 10 12 14 Altitude [ km ] 16 -0.5 0.0 0.5 Jacobian [ K / 1 ] 0 2 4 6 8 10 12 14 ₁₇

Figure 3.4: The same as Figure 3.3, but for AMSU-B surface channels 16 and 17. Figure from John [2005].

they are not as broad as the ones for the sounding channels. Thus, they are sensitive to Earth’s surface and the lowest layers of the atmosphere. In moist atmospheric conditions the Jacobian peak is also shifted to the higher altitudes.

As we have seen in Figure 3.3, Jacobians for a moist atmosphere peak at higher altitudes. This means, the more water vapor, the higher the sensitivity altitude of AMSU-B sounding channels. Since temperature in the troposphere rapidly decreases with the altitude [Wallace and Hobbs, 1977], the higher the sensitivity altitude, the colder the measurements. Therefore, the Jacobians, as defined in Equation 3.1, for AMSU-B sounding channels are negative.

Contrary to sounding channels, Jacobians for surface channels are positive. This can be explained by the fact that increasing water vapor concentration will lead to an increase in emission which then will lead to greater brightness temper-atures. Note that in moist atmospheric conditions AMSU-B Channel 17 behaves like a sounding channel i.e., its Jacobian peak is negative.

In Figure 3.3 the dashed lines depict Jacobians for the outermost off-nadir viewing geometry. For all AMSU-B sounding channels in both atmospheric scenarios those Jacobians are shifted to the higher altitudes. This feature can be explained by the limb darkening phenomenon. In an off-nadir viewing geometry signal has to travel to the sensor along the slanted path. Since this path is much

(23)

3.1 The Advanced Microwave Sounding Unit B 19 0 100 200 300 400 500 600 700 800 900 1000 −7 −6 −5 −4 −3 −2 −1 0 1 2

Distance from nadir [km]

∆

TB

[ K ]

Figure 3.5: Limb effect in AMSU-B channels. Differences of brightness tem-peratures measured at different viewing angles are plotted against the distance from the nadir of the AMSU-B swath. Different plotting symbols represent different AMSUB channels: Channel 16 -triangles, 17 - circles, 18 - crosses, 19 - pluses, and 20 - stars. Figure from Buehler et al. [2004].

longer than the path of a signal in a nadir viewing geometry, only the signal originating from the higher altitudes can reach the sensor. In other words, in an off-nadir viewing geometry sensitivity altitudes of the AMSU-B sounding channels are higher than the ones in a nadir viewing geometry. The difference of this sensitivity is about 1 km in altitude space.

The Jacobians for AMSU-B Channel 16 and 17, Figure 3.4, in maximum off-nadir viewing angle hardly show any difference. Obviously, the limb effect is very small for these channels and cannot be seen in the Jacobian plots.

The limb effect can be clearly seen in brightness temperature space. Fig-ure 3.5 demonstrates this effect in all AMSU-B channels. As expected, there is no or almost no brightness temperature difference between near-nadir measure-ments (0–200 km away from the nadir). However, the further from the nadir, the larger the limb effect. The largest limb effect is in the sounding channels, with the maximum reaching 7 K. This order of magnitude coincides with the tem-perature lapse in the troposphere for 1 km (see discussion about the sounding Jacobians for different viewing angles above).

This figure confirms that the limb effect in AMSU-B channels 16 and 17 is very small. It reaches up to 1 K for the outermost viewing angle of

(24)

Chan-nel 16. Note that in these chanChan-nels, the further from the nadir, the higher the brightness temperatures. This effect is called limb brightening. A column of the atmosphere contributing to the signal is significantly larger in off-nadir than in nadir viewing geometry. Therefore, the resulting brightness temperature is also greater in off-nadir than in nadir viewing geometry.

(25)

4 Asymmetry Error in Satellite

Data

The currently available microwave humidity sensors are cross-track scanners. This means that the instrument viewing angle is different for different measure-ments along the scan-line. This leads to limb darkening or limb brightening in the data, which has to be taken into account in the data analysis. The atmo-spheric limb effect should depend only on viewing angle, and hence should be symmetric about the nadir point in the middle of the instrument scan. However, in the real world instrumental effects may lead to scan asymmetries. The in-strumental origin of these asymmetries remains not fully understood. Possible explanations include a polarization misalignment or an antenna pointing angle error [Weng et al., 2003].

For the AMSU-A sensor, which is dedicated to temperature measurements, such asymmetries have been reported by Weng et al. [1999, 2003]. However, to my knowledge there is no published investigation of this issue for the humidity sensor AMSU-B. The issue is important, because uncorrected scan biases will introduce biases in humidity products derived from the data, for example hu-midity climatologies. The aim of the study described here is to close this gap, to estimate the scan dependent bias for the different channels of the different currently operational AMSU-B instruments, and to document the development of the biases with time.

Section 4.1 of this chapter will very briefly introduce the AMSU-B data. In Section 4.2 the methodology will be described. Then, the results and their dis-cussion will be presented in Section 4.3. Finally, Section 4.4 will give a sum-mary and the conclusions of this study. The results presented in this chapter are published in Buehler et al. [2005b].

4.1 AMSU-B Data

The time period of data used for this study varies with the satellite. It is March 2000 to August 2005 for NOAA-15, September 2000 to August 2005 for NOAA-16, and September 2002 to August 2005 for NOAA-17. All three sensors are still operational at the time of writing.

(26)

The AMSU data were calibrated with the ATOVS and AVHRR Processing Package (AAPP) which is briefly described in Atkinson and Whyte [2003]. Data of NOAA-15 before March 2000 were discarded, because the AMSU-B instru-ment, especially Channels 17 and 19, suffered strongly from the radio frequency interference (RFI) at the beginning [Atkinson, 2001]. Even for later NOAA-15 data, RFI correction coefficients have to be applied to the AMSU-B data, to re-duce the RFI problem. The procedure is described in Atkinson [2001]. Over time, ten sets of correction coefficients were made available by NOAA [2000]. These coefficients, their issue dates, and a brief description of each set of coef-ficients can be seen in Appendix M of NOAA [2000].

These RFI correction coefficients are included in the AMSU level 1b data available from the Comprehensive Large Array-data Stewardship System (CLASS) and are used by the AAPP software when the data are calibrated, i.e., converted to level 1c data. However, users of the CLASS 1b data should be aware that although a new set of RFI coefficients was documented in NOAA [2000] on August 13, 2004, the coefficients in the 1b data were still the old ones as of writing this. This was discovered in the course of the study described here and is discussed further in Section 4.3.

It should be noted that the RFI problem is most affecting the AMSU-B instru-ment on NOAA-15. The AMSU-B instruinstru-ment on NOAA-17 also suffers RFI problems to a small extend, with a maximum of 2 K for Channel 19 and 1 K for Channel 18. Correction coefficients for that instrument were issued on July 12, 2002 [NOAA, 2000]. The error due to RFI problem for AMSU-B on NOAA-16 is within the instrument noise level, therefore no correction is applied.

4.2 Methodology

For an individual instrument scan-line, it is quite obvious that there will be a natural cross-track asymmetry, due to the inhomogeneity of the atmosphere and the surface. Different measurements along the scan-line sample different atmo-spheric and surface states. However, one can assume that this natural variability cancels out when a large amount of data over a sufficiently long period of time is averaged.

The asymmetry ∆Tb for a particular viewing angle is defined as the mean

brightness temperature for the scan position corresponding to this viewing angle on the left side of the scan minus the mean brightness temperature for the scan position corresponding to this viewing angle on the right side of the scan. So,

(27)

4.2 Methodology 23

0 10 20 30 40 50

Viewing angle [degrees] 226 228 230 232 234 236 238 240 Tb [K] CH16 glob. 0 10 20 30 40 50

Viewing angle [degrees] 250 252 254 256 258 260 262 264 Tb [K] CH20 glob.

Figure 4.1: The mean brightness temperature as a function of the instrument viewing angle for different AMSU-B channels. Viewing angle zero is nadir. Different plots correspond to the AMSU-B channels from 16 to 20. The solid and dash-dotted lines correspond to the left and right side of the satellite track in flight direction, respectively. Shown are the results for NOAA-16 data for the MAM 2004 time period. The data used cover all longitudes and latitudes (global data).

∆Tb(0.55◦) = Tb45− Tb46

..

. (4.1)

∆Tb(48.95◦) = Tb1− Tb90

where Tb means brightness temperature, indices indicate the scan position,

and the overbar denotes the mean. For better understanding, please look at the AMSU-B scan-line sketch shown in Figure 3.1 on page 15. Note that on that figure Tb1, Tb2 etc. are called Cell 1, Cell 2 etc.

Figure 4.1 shows the mean brightness temperatures as a function of the in-strument viewing angle, for different channels of AMSU-B, for NOAA-16 data, for the time period March to May 2004. Measurements on the two symmetric around a nadir sides of the scan are indicated by different line styles.

(28)

0 10 20 30 40 50 Viewing angle [degrees] 250 252 254 256 258 260 262 264 Tb [K] CH16 trop. 0 10 20 30 40 50

Viewing angle [degrees] 276 278 280 282 284 286 288 290 Tb [K] CH17 trop. 0 10 20 30 40 50

Viewing angle [degrees] 266 268 270 272 274 276 278 280 Tb [K] CH20 trop.

Figure 4.2: The same as in Figure 4.1, but the data used are limited to the tropics (latitudes ±30◦).

This figure shows the result for the global data. First of all, it demonstrates the limb darkening effect for the sounding channels 18 to 20 and the limb bright-ening effect for the surface channel 16. These effects are explained on page 18. The behavior of Channel 17 is mixed, for viewing angles close to nadir it be-haves like a surface channel and shows limb brightening, but for viewing angles close to the edge of the scan it starts to behave more like a sounding channel and shows limb darkening. Besides the general limb effect, the figure shows that there is a significant difference between the measurements from the symmetric sides of the satellite scan-line. The brightness temperatures from the right half of the scan are always warmer.

For the global data, the maximum brightness temperature differences ∆T_b between the two sides of the track occur at the edges of the scan-line. Maximum

∆Tb values are −2.56, −3.44, −1.85, −2.47, and −5.18 K for the AMSU-B

channels 16, 17, 18, 19, and 20, respectively.

The explanation for the largest part of this asymmetry is the geometry of the satellite track. Since the satellite is in a sun-synchronous orbit with an inclina-tion angle of 98.7◦, only the part of the right side of the track covers the North Pole, and only the part of the left side of the track covers the South Pole. This behavior is shown in Figure 4.3. The asymmetry in the mean brightness

(29)

tem-4.2 Methodology 25

Figure 4.3: The AMSU-B swath. The left half of the swath is colored in red, the right half in black. The satellite moves from the North to the South Pole. The data used to produce this plot are from NOAA-16 on May 5, 2004, at around 2 AM.

peratures is introduced by the asymmetry of atmospheric and surface conditions at the two poles. The atmosphere over the South Pole is significantly colder and drier. Furthermore, there are significant differences in snow and ice cover between the two poles, affecting the surface channels.

To eliminate this artifact, the analysis is restricted to the tropical region i.e., ±30 degrees of latitude. The results for this restricted area are shown in Fig-ure 4.2. From this plot one can conclude that the biases are greatly reduced, their maximum values drop down to 0.69, 0.81, −0.39, 1.40, and 1.73 K for the AMSU-B channels 16, 17, 18, 19, and 20, respectively. It is assumed that these remaining small biases are mostly of instrumental origin. Since similar atmo-spheric states are observed by both sides of the scan on average, the effect of atmospheric inhomogeneities should cancel out. The exception is the channels with strong surface influence, as will be discussed in Section 4.3. From now on, results will be shown only for the tropical region as defined above.

For the study, the available data were divided into three-monthly

sub-sets: March-April-May (MAM) 2000, June-July-August (JJA) 2000,

(30)

(DJF) 2000/2001, MAM 2001, JJA 2001, . . . , MAM 2005, JJA 2005. Each sub-set contains about 1,000,000 measurements for each scan position.

4.3 Results and Discussion

Figure 4.4 shows ∆Tb, the scan angle dependent asymmetry in mean

bright-ness temperature, as a function of nadir viewing angle for the AMSU-B surface channels. This is the data version available from CLASS, i.e., the RFI correc-tion coefficients available in the level 1b data were applied. Different sub-plots display results for different satellites and channels. The different lines represent different three-monthly time periods of the data, the lines for recent data are darker. As a help in interpreting these curves, rough estimates of the radiomet-ric noise for each channel are displayed as horizontal lines. Noise estimates for NOAA-15 and 16 are from Buehler et al. [2004], noise estimates for NOAA-17 are copied from NOAA-16.

To check the influence of different surface types, data were sorted into land and sea data. For the AMSU-B surface channels 16 and 17 the results depend somewhat on the surface type, therefore, both classes are shown separately. The dependence of the asymmetry curve on the surface type for the surface channels indicates that the assumption that inhomogeneities are random and, therefore, average out does not hold completely for these channels.

As expected, for the sounding channels both classes lead to similar results. Therefore, the data was not separated into land and sea, so only global data are shown in Figure 4.5. The further discussion will, therefore, focus on the three sounding channels.

In general, observations close to nadir (viewing angle 0◦) have the smallest asymmetries, as expected. The largest asymmetries tend to occur towards the edge of the scan, but not always directly at the edge. The figure confirms that the simple method used to calculate the asymmetry is stable, in the sense that the shape of the asymmetry curve for each satellite and channel is remarkably similar for the different time periods studied, except for a slow evolution in some cases. I will come back to the time evolution of the asymmetry later.

The different asymmetry shapes for the same channels of different instru-ments suggest that each instrument has its particular signature. The oldest AMSU-B instrument on NOAA-15 is most strongly affected by scan asym-metries. Its sounding channels 18 to 20 suffer significantly, with asymmetries reaching ∆Tbvalues of approximately 1.9, 9.7, and 3.0 K, respectively. Hence,

particularly the newer data from Channel 19 of this instrument should be used only with caution.

(31)

asym-4.3 Results and Discussion 27

0 10 20 30 40 50

Viewing angle [degrees] -2 -1 0 1 2 ∆Tb [K] N15, CH16, Land 0 10 20 30 40 50

Viewing angle [degrees] -2 -1 0 1 2 ∆Tb [K] N15, CH16, Sea 0 10 20 30 40 50

Viewing angle [degrees] -2 -1 0 1 2 ∆Tb [K] N17, CH17, Sea

Figure 4.4: The asymmetry ∆Tb (left side minus right side of the scan-line in

flight direction) as a function of viewing angle. Nadir is at zero degree viewing angle. The rows of plots from top to bottom corre-spond to the satellites NOAA-15, 16, and 17. The first two columns from left to right correspond to the AMSU-B Channel 16 where the analysis was made for the data over the land and sea only. The next columns are the same, but for the the AMSU-B Channel 17. The dif-ferent lines correspond to difdif-ferent three-monthly time periods, the lines for recent data are darker. Horizontal lines indicate the noise equivalent temperature.

(32)

0 10 20 30 40 50 Viewing angle [degrees] -2 -1 0 1 2 ∆Tb [K] n15, CH18 0 10 20 30 40 50

Viewing angle [degrees] -10 -8 -6 -4 -2 0 2 ∆Tb [K] N15, CH19 0 10 20 30 40 50

Viewing angle [degrees] -2 -1 0 1 2 ∆Tb [K] N15, CH20 0 10 20 30 40 50

Viewing angle [degrees] -2 -1 0 1 2 ∆Tb [K] n16, CH18 0 10 20 30 40 50

Viewing angle [degrees] -2 -1 0 1 2 ∆Tb [K] N17, CH20

Figure 4.5: The same as in Figure 4.4, but for the AMSU-B channels 18, 19, and 20. Note that there is no separation of the data into the land and sea as in Figure 4.4. Also note that the y-axis scale for Channel 19 on NOAA-15 is different from the others.

(33)

4.3 Results and Discussion 29

0 10 20 30 40 50

Viewing angle [degrees] -6 -4 -2 0 2 4 ∆Tb [K] N15, CH16 0 10 20 30 40 50

Viewing angle [degrees] -6 -4 -2 0 2 4 ∆T b [K] N15, CH19 0 10 20 30 40 50

Viewing angle [degrees] -6 -4 -2 0 2 4 ∆T b [K] N15, CH20

Figure 4.6: It is the same as in Figure 4.5, but shows the scan asymmetry for NOAA-15, 1999.

metries. The ∆Tbvalues are mostly below 1 K, except for the edge of the scan for

Channel 19, where they reach 1.4 K and for Channel 20, which shows significant asymmetries reaching negative values down to almost 1 K and positive values of approximately 2 K. As a rough help in interpreting these numbers, consider that according to Buehler and John [2005] a brightness temperature difference of 1 K for Channel 18 corresponds approximately to a relative difference in upper tropospheric relative humidity of 7%. Sensitivities can be assumed to be of the same order of magnitude for the other humidity sounding channels 19 and 20. Hence, the asymmetry in Channels 19 and 20 is large enough to introduce sig-nificant humidity errors, if neglected. In contrast, the asymmetry in Channel 18, with an absolute value below 0.5 K, can most likely be safely ignored compared to other sources of uncertainty.

(34)

be drawn, since not as much data are available as for the other instruments. The data available so far look very promising, with the asymmetry values even smaller than for NOAA-16. The only notable exception here is Channel 20, which can have asymmetries up to 2.3 K.

As it was mentioned before, the AMSU-B instrument on board NOAA-15 suf-fered strongly from the radio frequency interference (RFI) right after the launch in 1998 [Atkinson, 2001]. The scan asymmetry during this time is expected to be high. Figure 4.6 shows the scan asymmetry of the AMSU-B instrument on board NOAA-15 for spring, summer, autumn, and winter 1999 (March 1999 to February 2000). In the first half of 1999 (gray lines in Figure 4.6) there obvi-ously is an unrealistic bias in all the channels of the AMSU-B. However, starting from the autumn of 1999 (dark lines in Figure 4.6) the RFI problem seems to be corrected for at least Channel 18 – its scan asymmetry bias is within the instru-ment noise. Nevertheless, data prior to 2000 were not taken into account in this study.

While the development of the asymmetry with time has a random nature for some channels, for others the asymmetry is steadily increasing with time. For example, one can see from Figure 4.5 that the asymmetry for Channel 19 of NOAA-15 has significantly increased to -9.7 K for the most recent data. One can observe the same behavior for Channel 18 of the same instrument. To show the evolution more clearly, Figure 4.7 shows the maximum asymmetry for each three-monthly time period as a function of time for each channel and instrument. The most striking feature of this figure is the steadily increasing asymmetry for the AMSU-B instrument on NOAA-15. It can be observed for all channels except Channel 16. The increase can be approximately described by a linear trend, as indicated in the figure. The most likely cause of this problem is a change in the characteristics of the radio frequency interference that the AMSU-B instrument on this satellite has experienced from the start, as mentioned ear-lier. The figure also shows the impact of different RFI correction coefficients. The coefficients published by NOAA [2000] on August 13, 2004 do significantly reduce the asymmetry. Note that these coefficients were not implemented in the 1b data, so they have to be manually applied. As shown by Figure 4.7, the latest coefficients also significantly reduce the asymmetry in data before August 13, 2004. From this study, the use of the August 13, 2004 coefficients for Channel 19 data later than March 1, 2003 is recommended.

Compared to NOAA-15, the instruments on NOAA-16 and 17 show no sig-nificant trends. The apparent trends for Channels 19 and 20 on NOAA-17 are not significant because the time series of data is not yet long enough.

(35)

4.3 Results and Discussion 31 2000 2001 2002 2003 2004 2005 2006 Time [year] -2 -1 0 1 2 max( ∆ Tb ) [K] CH16 NOAA 15 NOAA 16 NOAA 17 2000 2001 2002 2003 2004 2005 2006 Time [year] -2 -1 0 1 2 max( ∆ Tb ) [K] CH17 NOAA 15 NOAA 16 NOAA 17

NOAA 15 post Sep. 04 RFI

2000 2001 2002 2003 2004 2005 2006 Time [year] -2 -1 0 1 2 max( ∆ Tb ) [K] CH18 NOAA 15 NOAA 16 NOAA 17

2000 2001 2002 2003 2004 2005 2006 Time [year] -10 -8 -6 -4 -2 0 2 max( ∆ Tb ) [K] CH19 NOAA 15 NOAA 16 NOAA 17

2000 2001 2002 2003 2004 2005 2006 Time [year] -2 -1 0 1 2 max( ∆ Tb ) [K] CH20 NOAA 15 NOAA 16 NOAA 17

Figure 4.7: The time evolution of the maximum asymmetry for three-monthly time periods. The plots from left to right correspond to the AMSU-B channels 16, 17, 18, 19, and 20. Different line styles indicate different satellites: NOAA-15 (thick solid), NOAA-16 (dotted), and NOAA-17 (dashed). The thin solid line is for NOAA-15 where RFI correction coefficients as of August 13, 2004 are manually applied to level 1b data. The diamonds indicate the last, as of writing, RFI coefficients update date. Straight lines show linear fits.

(36)

Table 4.1: The range of maximum absolute brightness temperature asymmetries. This table summarizes Figure 4.7. The maximum of the absolute value of ∆Tbwas calculated for three-monthly time periods for each

channel and satellite. Given here are the smallest and largest of these numbers.

Channel NOAA-15 NOAA-16 NOAA-17

16 0.98 – 1.45 0.38 – 0.89 0.19 – 0.50

17 0.15 – 2.93 0.22 – 1.07 0.07 – 0.91

18 0.71 – 1.90 0.23 – 0.53 0.28 – 0.49

19 0.88 – 9.67 0.62 – 1.41 0.30 – 0.67

20 0.83 – 2.95 1.21 – 2.13 0.80 – 2.34

4.4 Summary and Conclusions

Using a simple method of averaging the brightness temperatures for different instrument viewing angles, the scan asymmetry in measured brightness temper-atures of the AMSU-B instrument was quantified. The exercise was carried out for all available data of NOAA-15, 16, and 17. For the three sounding channels the results are independent of the surface type. This gives a reasonable con-fidence that they show instrumental effects. For the two surface channels the results depend somewhat on the surface type and should, therefore, be regarded with caution.

The development of the asymmetry over time was investigated by considering the maximum value of three-monthly averages. Table 4.1 summarizes the range of maximum asymmetries for the different satellites and channels.

The AMSU-B instrument that suffers the most asymmetry is the one on NOAA-15. Particularly, its most affected channels are 19 and 20, which have maximum asymmetry values of −9.67 and −2.95 K, respectively. Moreover, there is clear evidence that for Channels 17 to 20 of this instrument the asym-metry is steadily increasing with time. The most likely explanation for this in-crease are changes in the RFI characteristics, because the application of appro-priate RFI correction coefficients significantly reduces the asymmetry. Updated coefficients were reported in NOAA [2000] on August 13, 2004, but were not implemented in the level 1b data from CLASS. It is recommended to use this set of coefficients for NOAA-15, Channel 19 data after March 1, 2003.

In contrast to NOAA-15, asymmetries for NOAA-16 and 17 are smaller and show no significant trend. Channels with maximum asymmetry exceeding 1 K on NOAA-16 are 17, 19, and 20. On NOAA-17 it is only Channel 20. The in-strument on NOAA-17 has clearly the smallest asymmetries, but also the short-est time series so far. The asymmetries in some cases show a seasonal cycle,

(37)

4.4 Summary and Conclusions 33

most pronounced for NOAA-16 Channel 17 with an amplitude of approximately 0.8 K.

Of particular interest for the next studies is Channel 18 of all satellites, which is used to derive the upper tropospheric humidity climate data products. To check its scan asymmetry was an important motivation for the presented study. This channel was found to be very well behaved, with maximum asymmetry val-ues of 1.90, −0.53, and 0.49 K for NOAA-15, 16, and 17, respectively. Hence, with the exception of NOAA-15, asymmetries in Channel 18 can be neglected compared to other sources of error for most applications, since a brightness tem-perature difference of 0.5 K maps to a relative difference in relative humidity of only approximately 3.5%, according to Buehler and John [2005].

(38)

(39)

5 Cloud Impact on Microwave

UTH

Humidity in the atmosphere, and particularly in the upper troposphere, is one of the major factors in our climate system. Changes in its distribution affect the atmospheric energy balance. It is therefore essential to monitor and study upper tropospheric humidity (UTH), and to make such data available to the sci-entific community. In contrast to traditional direct measurements of atmospheric humidity by radiosondes, satellites provide humidity measurements with global coverage. Satellite measurements of UTH are typically made in two specific fre-quency regions: in the infrared at 6.3 µm and in the microwave at 183.31 GHz. The infrared instruments are the more established ones, whereas the microwave instruments became available only rather recently.

UTH can be retrieved from satellite radiances using an algorithm developed by Soden and Bretherton [1996]. They used a linear relation between the natu-ral logarithm of UTH and brightness temperature (ln(UTH) = a + b ∗ TB), and

derived the fit parameters a and b using linear regression. In this algorithm, UTH is defined as the Jacobian weighted mean of relative humidity in the upper troposphere which is roughly between 500 and 200 hPa. Soden and Bretherton [1996] applied the algorithm to infrared data from the HIRS instrument.

One of the available microwave instruments for measuring UTH is the Ad-vanced Microwave Sounding Unit B (AMSU-B), which is discussed in detail in section 3.1 on page 13 of this thesis. As we have seen, this instrument was specifically designed for humidity observations. Recently, Buehler and John [2005] demonstrated that UTH can be derived from 183.31±1 GHz channel brightness temperatures of this instrument with a precision of 2 %RH at low UTH values and 7 %RH at high UTH values. The same retrieval algorithm was used for the work described here. In the current study the above paper will be referred to as BJ.

In general, clouds are more transparent in the microwave than in the infrared. Therefore, data from microwave sensors are less contaminated by clouds than data from IR sensors. This is particularly true for Channel 18 of AMSU-B. As the Jacobian for this channel shows (see Figure 3.3), the signal it receives originates mostly from the upper part of the troposphere. Thus, it is not sensitive to low clouds. However, clouds can affect the measurement if there is a high

(40)

cloud with a high ice content in the line of sight (LOS) of the instrument. In such a case the radiation is scattered away from the LOS by ice particles in the cloud so that the brightness temperature measured by the instrument is colder than it would be without the cloud.

In clear-sky conditions brightness temperatures from Channel 18 (T_B18) are colder than brightness temperatures from Channel 20 (T_B20). This is due to the atmospheric temperature lapse rate, and the fact that Channel 18 is sensitive to a higher region of the troposphere than Channel 20. However, in the presence of ice clouds T_B18can be warmer than T_B20. Thus, the brightness temperature differ-ence, defined as ∆TB= T_B20− T_B18 can be used to detect the presence of clouds.

For example, Adler et al. [1990] showed, using aircraft microwave observations, that ∆TBcan reach up to −100 K in a strong convective system and Burns et al.

[1997] suggested to use ∆TB< 0 as a criterion to filter out convective cloud cases

before retrieving water vapor from these measurements.

Greenwald and Christopher[2002] investigated the effect of cold clouds (de-fined as 11 µm brightness temperatures less than 240 K) on T_B18. They concluded that non-precipitating clouds produce on average 5 %RH error in UTH retrieval, whereas precipitating clouds produce 18 %RH error. They used infrared data to estimate the clear-sky background T_B18 (which was found to be 242±2 K) in order to estimate this error. Unfortunately, it is not possible to use the above numbers directly to assess the impact of clouds on a UTH climatology, because the averages refer not to the total number of measurements, but only to all clouds in the given class, where the class definitions are somewhat arbitrary. For exam-ple if the non-precipitating cloud class is extended towards thinner clouds, then the average impact of clouds of this class on UTH will appear to be smaller.

Another application of AMSU-B data cloud filtering is given by Hong et al. [2005]. The authors used the three AMSU-B sounding channels centered around 183.3 GHz to detect tropical deep convective clouds. They conclude that the deep convective cloud fraction in the tropics is around 0.3 %, and that the con-tribution of overshooting convection to this is around 26 %.

In this study a cloud filter that uses only the microwave data is developed, no additional infrared data are used (section 5.1). This is achieved by combining the approaches from earlier studies. To demonstrate the robustness of the filter, a case study is used (section 5.2.1). Next, the same case study is used to estimate the bias in the retrieved UTH that is introduced by the cloud filtering, and com-pared that to the bias introduced by the radiative effect of the clouds themselves, if they are not filtered out (section 5.2.2). In this context, the impact of surface emissions on retrieved UTH and on the cloud filtering procedure must also be discussed (section 5.2.3). Finally, the results from the case study are put on a firmer statistical basis by analyzing the cloud bias and cloud filtering bias for a stochastic dataset of midlatitude cloud cases with realistic statistics (section

(41)

5.1 Cloud Filter Methodology 37

5.2.4). Section 5.3 contains a summary and the conclusions of this work. The results presented in this chapter are published in Buehler et al. [2007].

5.1 Cloud Filter Methodology

The proposed cloud filter combines a threshold value for T_B18 (240 K for nadir data) and the ∆TB (0 K). To demonstrate this, the two-dimensional histograms

of ∆TB versus TB18 are used, such as the ones shown in Figure 5.1. In the top

plot, one month of AMSU-B measurements were used to plot the histogram. The color coded contour levels show the frequency of measurements, normal-ized relative to the maximum. In other words, the figure shows the combined probability density function (PDF) for T_B18and ∆TB. The maximum of the PDF

is near T_B18 = 245 K and ∆TB = 20 K. Most cases are indeed above T_B18 = 240 K

and ∆TB= 0 K. There is a tail of cases with negative ∆TBas low as -60 K. These

cases are identified mostly with clouds.

In principle, negative ∆TBvalues can also be an indicator of surface influence

on T_B18. Under very dry atmospheric conditions, both channels measure radi-ation emitted from the Earth’s surface and the atmosphere. If we assume the surface emissivity to be the same for both channels, T_B18 will be warmer than T_B20 because the contribution of atmospheric emission will be more for Channel 18 as the frequencies are closer to the line center.

To validate the assumption that negative ∆TBvalues are caused by clouds and

not surface effects, a two-dimensional histogram was plotted with brightness temperatures simulated for a clear-sky scenario (not considering clouds in the RT model). The result of this exercise is demonstrated in the middle plot of Fig-ure 5.1. The brightness temperatFig-ures used in this figFig-ure are calculated with the radiative transfer model RTTOV (see page 10) using ECMWF ERA-40 reanal-ysis data. The surface emissivity model was FASTEM [English and Hewison, 1998] over the ocean, and one of five different fixed surface emissivity values, depending on terrain type, over land. Confirming our expectations this figure shows that for clear-sky conditions there are very little data where ∆TBis below

0 K. The simulated clear-sky data in the middle plot also confirm the criterion of Greenwald and Christopher [2002] that T_B18should be above 240 K for clear-sky cases, which was not evident from the measured AMSU-B data in the top plot. It can be concluded that it is valid to use the two criteria in combination as a cloud filter.

While the threshold of 240 K for T_B18is valid for nadir looking measurements, due to limb darkening this threshold shifts to colder brightness temperatures for off-nadir looking measurements. As shown in the bottom plot of Figure 5.1, the depression from nadir to off-nadir is approximately 7 K.

(42)

−60 −40 −20 0 20 40 60 Tb(20) − Tb(18) [K] 220 240 260 280 Tb(18) NOAA−16(Nadir),Lat 45−50 0.0001 0.0001 0.0001 0.001 0.001 0.001 0.001 0.01 0.01 0.1 0.5 0.0001 0.0010 0.0100 0.1000 0.5000 1.0000 −60 −40 −20 0 20 40 60 Tb(20) − Tb(18) [K] 220 240 260 280 Tb(18) ECMWF(RTTOV),Lat 45−50 0.0001 0.0001 0.001 0.001 0.01 0.1 0.0001 0.0010 0.0100 0.1000 0.5000 1.0000 −60 −40 −20 0 20 40 60 Tb(20) − Tb(18) [K] 220 240 260 280 Tb(18) NOAA−16(Off−nadir),Lat 45−50 0.0001 0.0001 0.0001 0.0001 0.001 0.001 0.001 0.01 0.01 0.1 0.5 0.0001 0.0010 0.0100 0.1000 0.5000 1.0000

Figure 5.1: Top plot: combined histogram of the measured difference between AMSU-B Channel 20 and 18 brightness temperatures (y-axis) and Channel 18 brightness temperature (x-axis). Middle plot: RTTOV simulation of clear-sky brightness temperatures from ECMWF data. The data are for January 2004, near nadir viewing geometry, and a 45–50 latitude band. Bottom plot: the same as on the top, but for off-nadir looking measurements.

(43)

5.2 Results and Discussion 39

Table 5.1: Viewing angle (θ in degrees from nadir) dependent thresholds for Channel 18 brightness temperatures (in K).

θ T_B18 θ T_B18 θ T_B18 θ T_B18 θ T_B18 0.55 240.1 10.45 239.8 20.35 239.2 30.25 238.2 40.15 236.4 1.65 240.1 11.55 239.8 21.45 239.2 31.35 238.0 41.25 236.1 2.75 240.1 12.65 239.7 22.55 239.1 32.45 237.8 42.35 235.8 3.85 240.1 13.75 239.7 23.65 239.0 33.55 237.6 43.45 235.5 4.95 240.1 14.85 239.6 24.75 238.8 34.65 237.4 44.55 235.2 6.05 240.1 15.95 239.6 25.85 238.7 35.75 237.2 45.65 234.9 7.15 240.1 17.05 239.5 26.95 238.6 36.85 237.0 46.75 234.4 8.25 239.9 18.15 239.4 28.05 238.5 37.95 236.7 47.85 233.9 9.35 239.9 19.25 239.3 29.15 238.3 39.05 236.6 48.95 233.3

To derive viewing angle dependent values for the T_B18 threshold, clear-sky AMSU-B measurements for each instrument angle are simulated. This simu-lation was done with a sampled ECMWF data set [Chevallier, 2001] using the Atmospheric Radiative Transfer Simulator (ARTS) (see page 9). For each view-ing angle, minima of T_B18for a number of ∆TBintervals around the ∆TBthreshold

were determined. The mean of these minima was taken as T_B18threshold for that viewing angle. A summary of the threshold values is given in Table 5.1.

Figures similar to Figure 5.1 were also generated for other latitude ranges (see supplementary Figure A.1 in Appendix A). Overall, they look rather similar. In particular, the assumed threshold values appear to be applicable also for tropi-cal and sub-tropitropi-cal data. Since the focus of this study is on mid-latitudes, no attempt to fine-tune the filter for these other latitudes is made.

5.2 Results and Discussion

In this section the cloud filter using a strong ice cloud event over northern mid-latitudes is demonstrated. The clear-sky bias in the retrieved UTH fields due to cloud screening is also estimated, and the impact of surface emissions is dis-cussed.

5.2.1 Case Study

For the case study model fields and microwave measurements from a strong ice cloud event that occurred over the UK on January 25, 2002 were used. The model fields are from the Met Office (UK) mesoscale model UKMES [Cullen,

(44)

Ice Water Path 340˚ 340˚ 350˚ 350˚ 0˚ 0˚ 10˚ 10˚ 50˚ 50˚ 60˚ 60˚ 0.00.1 0.3 0.5 1.5 2.5 3.5 kg/m2

Figure 5.2: Left figure shows the infrared image of an ice cloud event over the UK in January 25, 2002. The image is from Advanced Very High Resolution Radiometer (AVHRR). Right figure shows the cor-responding ice water path calculated from the model fields.

1993]. Profiles of pressure, temperature, relative humidity, cloud ice water con-tent and cloud liquid water concon-tent were used to simulate AMSU-B radiances.

The cloud event is shown in Figure 5.2. On the AVHRR image we see the strong cloud cover passing over the UK and moving to the continental Europe. Associated ice water path (IWP) is shown on the right. Most of the area covered by cloud has IWP more than 0.5 kg/m2, maximum of IWP in the cloud reaches 3.5 kg/m2. Note that although there is about half an hour time difference be-tween the observation and simulation, model quite successfully reproduced the measurements.

To put the results on the cloud impact in the right perspective, one should keep in mind the properties of the applied UTH retrieval method and its limi-tations. For this purpose it first was applied to simulated clear-sky brightness temperatures. The results are displayed in Figure 5.3, which shows the quanti-ties UTHJac and UTHTb in different ways. The quantity UTHJac is the Jacobian

weighted upper tropospheric humidity, calculated from the relative humidity profiles and the AMSU-B Jacobian (for details, see BJ). The quantity UTHTb

is UTH calculated from the simulated brightness temperatures by applying the coefficients derived by BJ. The humidity unit used here and everywhere in this

(45)

5.2 Results and Discussion 41 Jacobian UTH 340˚ 340˚ 350˚ 350˚ 0˚ 0˚ 10˚ 10˚ 50˚ 50˚ 60˚ 60˚ 0 10 20 30 40 50 60 70 80 90 100 %RH

Clear−sky UTH − Jacobian UTH

340˚ 340˚ 350˚ 350˚ 0˚ 0˚ 10˚ 10˚ 50˚ 50˚ 60˚ 60˚ −22 −15 −10 −5 0 5 10 %RH 0 20 40 60 80 100 Jacobian UTH [%RH] 0 20 40 60 80 100 Clear-sky UTH [%RH]

Figure 5.3: A comparison between clear-sky simulated brightness temperatures converted to UTHTb and UTHJac. Top left: Model UTHJac; top

right: UTHTb− UTHJac; bottom left: Scatter plot of UTHTb

ver-sus UTHJac. Relative humidity here, as everywhere in this study, is

defined over liquid water.

study is the relative humidity over liquid water (%RH).

The top left plot of the figure shows a map of the UTHJac field. The top right

plot shows a map of UTHTb− UTHJac. The bottom left plot is a scatter-plot of

UTHTb versus UTHJac. The figure shows that the retrieval method works well,

as most of the differences are within ±5 %RH. It is interesting to note where the discrepancies between UTHJac and retrieved UTH, which are referred to as

regression noise in BJ, happen. Strong differences of up to 22 %RH occur in areas with unusual atmospheric states, for example behind the cold front. In this area, where warm air is over-laying cold air, the temperature and humidity lapse rates are less steep than in the average state, resulting in a different relation between ln(UTH) and brightness temperature.

Retrieval of upper tropospheric humidity data from microwave measurements