SEARCH

SEARCH BY CITATION

Keywords:

  • RADAGAST;
  • flux closure;
  • aerosol

Abstract

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Background
  5. 3. Surface Results
  6. 4. Top of Atmosphere Results
  7. 5. Longwave Atmospheric Divergences
  8. 6. Conclusions
  9. Acknowledgments
  10. References
  11. Supporting Information

[1] Simultaneous observations of thermal radiative fluxes and radiances from the surface (Atmospheric Radiation Measurement Mobile Facility, Niamey) and top of atmosphere (Geostationary Earth Radiation Budget (GERB) instrument) during the Radiative Atmospheric Divergence using ARM Mobile Facility, GERB data, and AMMA Stations experiment are compared with results from a radiative transfer model (Edwards-Slingo). Emphasis is placed on diagnosing the accuracy of the cloud-free radiation measurements using multiple instruments at the surface. The surface forcing from aerosol is found to regularly exceed 20 Wm−2, and reached ∼100 Wm−2 during the March 2006 dust storm. Equivalent comparisons are made with top of atmosphere (TOA) measurements but here radiance closure is not achieved. A disagreement is found between the angular anisotropy derived from GERB products and that from radiative transfer (RT) calculations. A hybrid TOA radiative flux time series is created using RT-calculated TOA anisotropy and GERB-observed TOA radiance. At 1100 UT (local noon), this hybrid flux differs from the Edition 1 GERB product by a positive difference in the range ∼0–10 Wm−2. Three collections of fluxes exist to calculate column-integrated atmospheric heating (divergence) from surface and TOA fluxes. The first two are fluxes from observations only or from RT calculations only. The third is a combination of RT calculation and observed fluxes that includes the hybrid flux. The resulting divergences are binned by sonde launch times and averaged over the year. The range of divergence during a day depends on the flux collection used (−200 to −111 Wm−2, −212 to −116 Wm−2, or −205 to −112 Wm−2) for observations only, for RT calculations only, or for observation-calculation fluxes. All estimates agree as to the interday variation being larger than that of intraday variability.

1. Introduction

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Background
  5. 3. Surface Results
  6. 4. Top of Atmosphere Results
  7. 5. Longwave Atmospheric Divergences
  8. 6. Conclusions
  9. Acknowledgments
  10. References
  11. Supporting Information

[2] North Africa and the Arabian peninsula share one of the most unusual radiation environments on the planet. For example, this is the only tropical region with a negative annual mean net radiation balance at the top of the atmosphere [Hartmann, 1994]. There are few, permanent surface radiation measurement sites [Ohmura et al., 1998], in part because of the harsh environment. In addition, until recently broadband radiation budget measurements at the top of the atmosphere (TOA) were limited to twice per day from instruments on sun-synchronous platforms [Jacobowitz et al., 1984; Barkstrom, 1984; Hucek et al., 1996]. The launch of the Geostationary Earth Radiation Budget (GERB) instrument [Harries et al., 2005] in 2002 substantially improved the TOA sampling. The deployment of the Atmospheric Radiation Measurement (ARM) Program Mobile Facility (AMF) to Niamey, Niger for 2006 [Miller and Slingo, 2007] provided a similar improvement in surface sampling. The combination of GERB and the AMF within the Radiative Atmospheric Divergence using ARM Mobile Facility, GERB data, AMMA Stations (RADAGAST) experiment [Slingo et al., 2008] has resulted in a yearlong time series of shortwave and longwave radiation measurements from both surface and TOA, together with the associated meteorological variables.

[3] These observational developments provide an opportunity to test radiative transfer (RT) schemes within this environment. In turn, the simulations can be used to examine the consistency of the measurements and to arrive at an improved representation of the overall radiation budget. This work represents a first step in this process, by concerning itself with simulations of the longwave fluxes and divergences. The Sahelian location of Niamey is interesting from a modeling perspective for several reasons; in particular, the range of integrated column water vapor (CWV) and aerosol optical depth. Both are consequences of the location which causes Niamey to be affected by the moist southwesterly winds of the West African monsoon in boreal summer, and by the dry, and dust laden, northeasterly Harmattan winds from the Sahara in boreal winter.

[4] The geostationary orbit of the GERB instrument allows for ∼15 minute sampling over 24 hrs of the broadband radiances at TOA. This time resolution can be matched by observations from the AMF, thus making comparisons with radiative transfer simulations relatively straightforward. To derive the atmospheric heating it is therefore possible simply to combine the observations from the surface and TOA [Slingo et al., 2009]. However, this is strictly only valid if all the measurements are over identical spatial scales. The consequence of making this assumption from these observations is studied here. The principle difficulty is that the surface measurements are made from a single point whereas the satellite data are integrated over the “footprint” (the point spread function) of the instrument. This mismatch in spatial scales can result in radiative inconsistencies, such as in the description of the land surface or of cloudiness, which in turn results in increased random and systematic errors. The random error effect is analyzed in a companion paper [Settle et al., 2008] and is ignored here by assuming atmospheric homogeneity and surface stationarity. The systematic errors are studied and reduced by using satellite retrievals to characterize the wider area about the AMF.

[5] This paper utilizes measurements only from clear sky conditions, partly to satisfy the sky homogeneity assumption. The aim of this work is to establish the uncertainties and biases in the cloud-free RT calculations which affect subsequent divergence calculations. These errors are subsequently important in quantifying radiation closure errors when clouds are present.

[6] In the longwave, the aerosol direct effect at the surface varies greatly, depending on region and the nature of the transported particles: a literature review [Claquin et al., 1998; Miller et al., 2004; Ritter et al., 2005; Dey and Tripathi, 2008] found values quoted between ∼1 and ∼70 Wm−2. The upper limit was determined via RT calculations in a desert environment where dust particle radii were greater than 1 μm. Since there is a paucity of radiation studies within such environments but which do occur during RADAGAST [Slingo et al., 2006], the full range of long-wave (LW) direct effect values is expected here. This variation will be exploited in this paper to understand the inferred aerosol direct effect via correlations with visible retrievals. The use of surface infrared retrievals [Turner, 2008] as an input for the RT profiles is avoided to prevent potentially common errors in aerosol-free profiles causing consistent surface but inconsistent TOA radiative closures. Another issue is that aerosol can affect the anisotropy of the TOA radiance field and therefore affect the retrieval that is used to create TOA flux products. Therefore, an examination of aerosol effects in the longwave at both surface and TOA is required to permit a conclusive reduction of radiation residuals between observations and calculations.

[7] In the following section, the measurements of, and arrangements for modeling the longwave radiation during 2006 at Niamey are introduced, as well as the data that are used to represent the disparity in spatial scales between the surface and TOA data. Section 3 first analyzes these surface upwelling radiation properties. This is followed by simulations of downwelling aerosol-free fluxes and radiances at the surface, with an initial emphasis on diagnosing the accuracy of the RT humidity profile. The aerosol direct effect derived from these surface measurements and simulations is compared with aerosol retrievals in two fashions. The second of these leads to simulations of radiation using an estimated aerosol profile input. Having established the errors in the RT calculations at the surface, the radiances and fluxes at TOA are compared in section 4. Here, both the aerosol-free and aerosol-included RT calculations are utilized. Section 5 presents estimates of the cloud-free divergences using fluxes that are observed, simulated, and from a best estimate collection which is introduced in an attempt to reduce systematic errors.

2. Background

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Background
  5. 3. Surface Results
  6. 4. Top of Atmosphere Results
  7. 5. Longwave Atmospheric Divergences
  8. 6. Conclusions
  9. Acknowledgments
  10. References
  11. Supporting Information

2.1. Radiation Measurements

[8] The RADAGAST field experiment consisted of two primary measurement locations: the ARM Mobile Facility (AMF) [Miller and Slingo, 2007] and the Geostationary Earth Radiation Budget instrument aboard Meteosat-8. The AMF was located at Niamey airport (2.18°E, 13.48°N) and was present between mid-December 2005 and early January 2007: data relevant to this work range from 6 January to 31 December. The GERB instrument (GERB-2) was located above 3.3°W, 0.0°N. The AMF was active for each day of 2006, with short instrumental dropouts, whilst GERB suffered from only two outages longer than a day. (However, the GERB instrument was taken offline into sun avoidance mode during 4 hours about midnight for 110 days.)

[9] The measurements of the downwelling surface radiation are from two sources: broadband fluxes from a pyrgeometer (using one of the two instruments), and spectral zenith radiances from the Atmosphere Emitted Radiance Interferometer (AERI) instrument [Knuteson et al., 2004]. A distinct advantage of the AERI is its calibration sequence, scheduled between measurements of sky radiation, which results in known measurement characteristics for all times. The rawinsonde ascents from the AMF were made four times daily at times described as 0600, 1100, 1700, and 2300 UT. (Restricting sonde data to those launched within 1 hr of these times removed 4% of the launches.) The more physically accurate MWRRET [Turner et al., 2007] measurements of CWV replaced the canonical (MWRLOS) data stream. These data were all retrieved from the ARM archive and are summarized in Table 1. In addition, using AMF observations alone a surface-based cloud mask was created, analogous to other approaches [Henderson, 2006; Long and Ackerman, 2000].

Table 1. Surface Data Streams Used in This Worka
Data StreamVariables MeasuredTemporal Resolution
  • a

    These are either measurements from the AMF deployment to Niamey or derived thereof.

AerisummaryAveraged longwave zenith-pointing spectrally resolved radiance400 s
Skyrad60sDownwelling longwave flux, sky-looking infrared thermometer (down_long_hemisp_shaded2 variable is used for flux)60 s
Gndrad60sUpwelling longwave flux, surface-looking infrared thermometer60 s
MwrretImproved microwave radiometer column water vapor retrievals60 s
SondewnpnRawinsonde, Vaisala RS-92, ascents4 daily
Met2 m height temperature, pressure60 s

[10] The GERB [Harries et al., 2005] instruments measure the total and shortwave radiance from geostationary orbit, and their difference returns the longwave radiance. Also aboard the Meteosat platform is the Spinning Enhanced Visible and Infrared Imager (SEVIRI) spectral imager [Schmetz et al., 2002]. A regression, on the basis of a database of radiative transfer calculations, utilizes the contemporaneous SEVIRI spectral measurements to derive the GERB radiance-to-flux conversion. This is used in turn to produce the operational GERB flux products [Clerbaux et al., 2003] from the radiance measurements; the Edition 1 averaged, rectified, and geolocated (ARG) products [Dewitte et al., 2008] are used in this work. The SEVIRI radiances themselves are used by the Nowcasting Satellite Application Facility (NWCSAF) [Slingo et al., 2009; Derrien and Le Gleau, 2005] to produce a TOA cloud mask which is used here. These products are summarized in Table 2.

Table 2. Satellite Data Streams Used in This Work
NameDescriptionSpatial Resolution
GERB ARGEdition 1 radiances and fluxes∼46 × 46 km
NWCSAF cloud maskSEVIRI-based cloud mask∼3 × 3 km
MOD11A2/MYD11A2Surface temperature from MODIS Terra/Aqua1 × 1 km
CIMMS IR emissivitySurface LW emissivity via MODIS measurements0.05° × 0.05°

[11] The errors from radiative measurements are provided in two different fashions: for the surface measurements, a standard deviation is made from the 60 samples per minute, whilst for the GERB ARG values, the quoted error for the radiances is a relative value and there is an additional error for the fluxes. These are summarized [Slingo et al., 2006] as one standard deviations: ±5Wm−2 for the surface fluxes, 1% for the GERB radiances, and ∼5 Wm−2 for the GERB fluxes. The systematic errors of the instruments are analyzed in this work in the context of calculating the atmospheric divergence.

2.2. Tools and Other Measurements

[12] The radiative transfer simulations were made using the Edwards-Slingo code [Edwards and Slingo, 1996] (hereafter referred to as ES96). Despite being designed for computational efficiency, it remains flexible in several areas, e.g., two-stream and spherical harmonic solvers are available and the latter was used for calculation of fluxes and radiances. The optical parameters of gases and aerosols are specified via spectral files, and here the file is as used in the MetOffice HadGEM1 model [Martin et al., 2006]. This is a nine-band model which operates between 3 μm and 100 μm and whose aerosol specification is from the WMO [World Meteorological Organization, 1990] with a modified size distribution [d'Almeida et al., 1991]. Using the times of sonde launches restricts the calculations to 4 per day, but this is sufficient to resolve the diurnal cycle for this work.

[13] A (binary) cloud mask is used to remove the times in the measurements when cloud affects the radiation. The SEVIRI-NWCSAF cloud mask results in ∼270 values per GERB ARG pixel, whose average when greater than 0.1 is interpreted as cloud. This mask results in evident false negatives from the surface perspective therefore the surface-based cloud mask replaces it for surface comparisons. However, when combining TOA and surface fluxes for divergence estimates, a joint mask is produced by assuming independence of each mask and combining them via a Boolean OR operator. Using the joint mask the percentage of available observations is decreased by ∼20% during the dry season and ∼66% during the wet (monsoon) season. (These seasons are of approximately the same length, and the monsoon period is defined for 2006 [Slingo et al., 2008] as being between 5 May and 29 October inclusive.)

[14] The sonde data are used to produce the temperature and humidity profiles for the radiative transfer code. The humidity is then scaled [Soden et al., 2004; Turner et al., 2004] such that it produces a CWV which is equal to the value from MWRRET retrievals. The profiles have ∼150 pressure levels with a fixed ozone prescription [Ellingson et al., 1991] (250 DU), and the long-lived gas species (CO2, CH4, O2, N2O) have vertically constant mixing ratios consistent with values from 2005 while halocarbons are neglected. These assumptions cause (acceptable) biases at the surface of −0.5 Wm−2, and at TOA of 0.7 Wm−2.

[15] These profiles can be used to simulate pristine sky (aerosol- and cloud-free) radiation fields and then, by comparisons with surface downwelling measurements, an estimate can be made of the direct radiative effect of aerosols. A companion paper in this special section concerning infrared aerosol retrievals from AERI radiances [Turner, 2008] shows good agreement of 11 micron optical depths from Niamey with 1 micron Aerosol Robotic Network (AERONET) retrievals from Banizoumbou [Holben et al., 2001], ∼60 km east of Niamey. (The AERONET site at Niamey itself only operated from August 2006 onward.) The analysis showed that a reasonable assumption is for the Ångstrom exponent between 11 microns and 1 micron to be a constant (∼0.33). The Banizoumbou AERONET retrievals are used for the analysis of differences between observations and calculations, but not directly as inputs into the RT profiles.

[16] The airport location means the area underneath the AMF radiation instruments is unrepresentative of the larger GERB ARG footprint. In particular, the wider area includes the River Niger and the urban area of Niamey proper. Hence, the profile surface properties, emissivity and surface temperature (LST), use satellite retrievals to supplement AMF measurements. The retrievals are from the Moderate Resolution Imaging Spectroradiometer (MODIS) instruments: LST via the MOD11A2 and MYD11A2 (the LST with 8 day time resolution and 1 km spatial resolution from the Terra/Aqua MODIS instrument) products [Wan and Li, 1997] and emissivity via the Cooperative Institute of Mesoscale Meteorological Studies (CIMMS) IR emissivity database [Seemann et al., 2008].

[17] The CIMMS emissivity retrievals are monthlong averages for several narrow bands over the longwave spectrum, from which the broadband emissivity was obtained using two separate approaches: by a bandwidth-weighted average and by using predetermined weightings [Wang et al., 2005]. Over the GERB ARG footprint, both methods resulted in emissivities with a difference of 0.01 RMS. The yearlong (1 January 2006 to 30 November 2006) average of these area average emissivities is 0.93 ± 0.01 so the RT profile emissivity, ε, is fixed to this value.

[18] To specify the LST in the RT profiles, TARG, an estimate of the temperature local to the AMF, TAMF, is required. This LST is transformed to TARG by using MODIS retrievals as described below. The measurement of LST, TIRT(t), from the AMF datastream uses a IR thermometer (IRT) which derives temperature from measurements of surface-emitted radiances between 9.8 μm and 11.5 μm. Alternately, a second measurement of upwelling radiation is from the upwelling pyrgeometer which observed the broadband flux, FULR;AMF(t). This flux can be turned into a temperature by assuming black body and Lambertian emission

  • equation image

Here we assume εAMF = 1, and so there is some unknown scaling factor. This is accounted for by scaling the diurnal mean of Tpyrg such that it is then equal to the diurnal mean of TIRT

  • equation image

where the overbar represents an average over the day (24 hrs) that relates to t.

[19] The IRT assumes a constant emissivity relation between the IR window and LW broadband, and if this were valid at the AMF site then the difference between TIRT and the derived Tflux would be expected to be proportional, i.e., a time-independent constant scaling factor. Figure 1 shows the difference between TIRT and Tflux for the four sonde launch times. It is evident that there is time dependence, particularly at 1100 UT, but verifying whether TIRT or Tflux is the appropriate value for TAMF requires comparisons of TOA radiances (section 4.1).

image

Figure 1. The difference between TIRT and Tflux. The boxes run clockwise in time (0600, 1100, 1700, and 2300 UT).

Download figure to PowerPoint

[20] Given TAMF, a transformation is needed to produce the temperature representing the wider area of the GERB ARG pixel, TARG. The MODIS LST retrievals are used for this purpose. They are (essentially) 8-day averages to compromise between data availability and time resolution. The retrievals are used to estimate the relationship between the LST for the MODIS footprint nearest the AMF, TMOD;AMF, and the LST over the ARG footprint, TMOD;ARG. The MODIS LST area transform was calculated over all 8-day periods between May and December, and was found to be approximately linear

  • equation image

To use this transform, the assumption made is that the MODIS footprint LST about the AMF is equivalent to the AMF LST: TAMF and TARG replace their MODIS equivalents in the equation. The MODIS footprint resolution (∼500 m) is sufficiently small to negate effects of the river or its adjacent paddy fields (≥3.5 km SW of the airport location). High-resolution imagery and local cartography also suggest that there are no obvious changes in land cover which could invalidate the assumption.

3. Surface Results

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Background
  5. 3. Surface Results
  6. 4. Top of Atmosphere Results
  7. 5. Longwave Atmospheric Divergences
  8. 6. Conclusions
  9. Acknowledgments
  10. References
  11. Supporting Information

[21] The results of the ES96 calculations are compared to the measurements from the AMF in this section. These initial simulations assume pristine sky conditions (cloud- and aerosol-free), and demonstrates how omitting aerosol affects the surface LW radiation calculations. The estimated aerosol direct effect at the surface is then analyzed in conjunction with retrievals of aerosol optical thickness (AOT) to understand to what extent this effect accounts for the difference between calculations and observations.

[22] For the surface downwelling radiation (DLR) flux, the time series of observation and pristine simulation are shown in Figure 2, with the observed less calculated (obs-calc) flux difference in Figure 3. The corresponding time series for the surface upwelling radiation (ULR) component are shown in Figures 4 and 5. These two components combine to form the surface net flux in the atmospheric divergence calculation.

image

Figure 2. The flux measurements (circles) with pristine simulations (crosses). Here DLR is surface downwelling radiation. The boxes run clockwise in time (0600, 1100, 1700, and 2300 UT).

Download figure to PowerPoint

image

Figure 3. The difference between measurements and calculations (cf. Figure 2). Here DLR is surface downwelling radiation. The boxes run clockwise in time (0600, 1100, 1700, and 2300 UT).

Download figure to PowerPoint

image

Figure 4. The flux measurements (circles) with pristine simulations (crosses). Here ULR is surface upwelling radiation. The boxes run clockwise in time (0600, 1100, 1700, and 2300 UT).

Download figure to PowerPoint

image

Figure 5. The difference between measurements and calculations (cf. Figure 4). Here ULR is surface upwelling radiation. The boxes run clockwise in time (0600, 1100, 1700, and 2300 UT).

Download figure to PowerPoint

3.1. Upwelling Surface Radiation

[23] The comparison of ULR shows that the overall differences between the RT calculations and the AMF observations can be significant. Given that the calculations uses the scaled temperature, TARG, and that the emissivity, ε, is below one, the ULR obs-calc values are positive as expected. This is displayed in Figure 5, for which the differences at the four times are between 9−2+2 and 28−7+8 Wm−2. As relative percentages these are between ∼2% and ∼5%, which are small but not negligible numbers. Since no instruments other than those used to specify the surface properties measure the upwelling surface radiation, the verification of the ULR will be dealt with in the TOA radiation analysis, section 4.

3.2. Downwelling Surface Radiation

[24] Two initial comparisons were made to verify the simulations and measurements of DLR, in order to eliminate the possibility of biases larger than observation or simulation error. First, a comparison of the ES96 simulations with independent calculations (E. Mlawer, personal communication, 2007), using the LBLRTM code [Clough et al., 2005], shows a correlation r value of 0.95. The mean difference is −0.9 ± 2.2 Wm−2 which does not vary with time of day; a result which is consistent with previous ES96 comparisons to line-by-line codes [Edwards and Slingo, 1996]. Second, a comparison using a parameterized DLR formula [Prata, 1996] at the sonde launch times showed that the parameterisation overestimated the flux by ∼20 Wm−2 during the day and underestimated by a similar amount at night. This is thought to be due to the relatively large lapse rates which were not adequately represented when the parameterisation was created. Over the period of a day, however, these lapse rate-related differences cancel so the diurnally averaged DLR can be used to assess for systematic biases in the AMF observations. The resulting difference of averaged DLR between observation and parameterisation is −1.9−4.4+4.7 Wm−2. Therefore, it is concluded that there is no observation bias larger than ±5 Wm−2 and that the RT calculations are consistent with LBL calculations to within ∼2 Wm−2.

3.2.1. Pristine Sky Calculations

[25] The observed DLR primarily shows the effect of change in CWV and near-surface temperature (Figure 2). In time, the monsoon onset and retreat (circa days 120 and 300) marks the largest change in average DLR because of the change in atmospheric humidity. The changes in temperature are most evident in the latter half of the year; after October the CWV fluctuated but the dominating trend is the decrease in both temperature and DLR. The pristine calculations model these overall trends, but there remains a significant underestimate. For example, during the first dry season the observed DLR generally increases from January to May but the simulations do not show such a clear change.

[26] The scatter of the flux residuals (the obs-calc differences) with CWV (Figure 6) demonstrates a clear link between flux residuals and the atmospheric profile. If there were a bias in either temperature or water vapor then the lower limit of the obs-calc difference would not be expected to be close to zero regardless of CWV value. In particular, only a few points show a negative obs-calc value (1% are more than the measurement error below zero) whilst a large number (40%) have a difference of ≤10 Wm−2 (twice the measurement error). In contrast, the upper limit of the flux differences decreases with increases of CWV. These observations are consistent with the aerosol direct effect being primarily responsible for the difference; as the CWV lessens, the IR window becomes optically thinner and the efficacy of the direct effect larger.

image

Figure 6. Observation minus calculation (obs-calc) of DLR, Δ, versus CWV. The circled circles mark the times during March dust storm, and the squared circles mark the times during the early June dust event. (See text for more detail.)

Download figure to PowerPoint

[27] For example, during the dust storm on 9 March 2006 [Slingo et al., 2006], there was a sonde launch at 05h31 for which CWV was 0.9 cm, the AERONET AOT (1020 nm) from Banizoumbou was 2.3, and the obs-calc residual equaled 99 Wm−2. Conversely, on 12 June, a sonde launch at 05h30 occurred when CWV was 4.2 cm, the AOT was 1.8, and the obs-calc residual equaled 23 Wm−2. These two periods are highlighted in Figure 6. In general, AOT and CWV were not highly correlated, although the short-duration high-intensity events characteristic of dust storms (e.g., in early March and late December) did primarily occur during the dry season.

[28] The obs-calc analysis was repeated with zenith radiances from the AERI. Since the intention here is to establish the accuracy of the temperature and humidity profiles, it is useful to compare radiances in several regions of the spectrum where temperature, water vapor, and aerosol in turn have a dominating effect. The band passes utilized and their primary sources of emission are shown in Table 3. Band 4 is dominated by CO2 emission and is essentially a test of near-surface temperature. Similarly, band 8 is dominated by water vapor and temperature, so can be used to verify the boundary layer water vapor quantity. Finally, band 5 has the same contributions but at far lower optical depths, and it is here where the effect of aerosol emission would be most notable. Figure 7 (left) shows the radiances from these bands plotted against CWV. As expected, the obs-calc differences for bands 4 and 8 show little correlation with CWV and near zero difference. The scatter for band 5 is reminiscent of the flux scatter plot: at lower CWV the range of obs-calc radiance difference increases, whilst the lower bound remains approximately constant. Interestingly, the lower bound for band 5 seem to have a nonzero residual and here it is noted that there are known issues regarding AERI radiance measurements and obscurations of the viewing window [Turner, 2008]. Consequently, the use of AERI radiances is restricted to these comparison analyses.

image

Figure 7. Observation-calculation differences of surface zenith radiances (AERI-ES96) versus CWV. The three bands have their spectral range listed in Table 3. (left) Without aerosol in RT calculations. (right) With aerosol.

Download figure to PowerPoint

Table 3. ES96 Band Passes Used for Comparison With AERI Radiancesa
HadGEM1–3 Band NumberBandpassEmission Sources
cm–1Microns
  • a

    The parentheses represent subsidiary sources of emission, either because they are well determined or relatively unimportant.

4590–75013.3–17.0CO2, H2O, (O3)
5800–990, 1120–12008.3–8.9, 10.1–12.5H2O, aerosol, CO2, (N2O), (O3)
81330–15006.7–7.5H2O, (N2O), (CH4)

[29] Returning to the fluxes, the primary interest at this stage is to quantify whether the residuals are primarily due to the aerosol direct effect. This can be attempted via a correlation of the diurnal average of obs-calc flux difference and the equivalent average AOT. (The diurnal average is calculated for days with four cloud-free sonde launches.) The following assumptions are made: AOT at the AERONET site (Banizoumbou) is correlated with that at the AMF (Niamey), the longwave dust emission is a simple function of the shortwave AOT, and that changes in vertical structure of aerosol within the atmosphere are negligible. The r correlation coefficient for 561 near-contemporaneous AERONET retrievals from the two sites is 0.91. The second assumption is validated by the companion paper [Turner, 2008]. The final assumption rests on temperature variations in the atmosphere. By taking diurnally averaged values, the change in emission from diurnal variation is reduced. A simple model for vertical distribution of aerosol is an even mass mixing ratio from the surface to 800 hPa. The diurnal mean temperature averaged over this altitude is 298 ± 3 K for all days in 2006. Raised to the 4th power, this temperature becomes an aerosol emission equivalent value whose resulting standard deviation is 4%. This value is far smaller than the range of AOT extinction measured (0.03–4.07). Given these results, the aerosol direct effect can be expected to vary primarily because of changes in the vertically integrated aerosol loading, as measured by the visible AERONET retrievals.

[30] Given that the direct effect efficacy is controlled by the opacity of the IR window, the data are binned into CWV intervals of 0–1, 1–1.5, 1.5–2.5, and 2.5 cm and greater. The flux residual AOT correlation for these bins are shown in Figure 8. Within each bin, a fit is made of the form

  • equation image

where the averages represent diurnal means, FDLR;AMF the AMF pyrgeometer measurement, and FDLR;pristine the pristine DLR calculation. This equation was based on fits to theoretical RT calculations. Each parameter has a physical meaning: α0 represents the offset error from pristine DLR, α represents the blackbody emission from the sky (e.g., “infinite aerosol”) less pristine DLR, and β relates the AERONET AOT to the IR opacity.

image

Figure 8. Observation calculation (obs-calc) of diurnal mean DLR, equation image, versus the diurnally averaged AERONET AOT (1020 nm), equation image. The lines correspond to the fits described in the text. The data are divided into the following four CWV bins: (a) 0–1 cm, (b) 1–1.5 cm, (c) 1.5–2 cm, and (d) 2.0 cm and greater.

Download figure to PowerPoint

[31] The values of these parameters are shown in Table 4. The data do seem to indicate a nonzero offset error, but the RMS errors are relatively large and so it can be considered that the number of data points are inadequate for fitting. The parameter α represents the aerosol direct effect when the opacity of the atmosphere is very large and can be compared with an upper limit derived from the AMF 2-m height temperature, T2m

  • equation image

The CWV bin of 1.5–2.0 cm results in a fitted value of α which is unphysical, so this is ignored. The fit values of α are considerably smaller than the estimate: for CWV 0–1 cm, the value is 95 ± 17 Wm−2 versus the estimate of 150 ± 10 Wm−2, while for CWV >2.5 cm, the fitted value is 31 ± 6 Wm−2 versus the estimate of 53 ± 1 Wm−2.

Table 4. Parameters to the Fits and RMS Error of the Form Δ(F) = α0 + α (1 − exp[−βequation image])a
CWV Rangeα0 (Wm−2)α (Wm−2)βRMS (Wm−2)
Lower (cm)Upper (cm)
  • a

    Where Δ(F) is obs-calc for the diurnal mean DLR and equation image is diurnal mean AERONET AOT (1020 nm). As shown in Figure 8.

015 ± 295 ± 170.6 ± 0.233
11.54 ± 250 ± 51.3 ± 0.323
1.52.55 ± 2500 ± 50000.1 ± 0.717
2.5−3 ± 631 ± 62.2 ± 0.921

[32] The final fit parameter, β, is more difficult to analyze; the values seem inversely proportional to the parameter α. This is a consequence of equation (4), β is dependent on the value of α but cannot be constrained accurately with the existing data. Therefore the fitted values of β do not accurately represent the relation between AERONET AOT and IR opacity. Fixing β to 1 to examine the effect of this coupling between variables does not appreciably change the fitted values of α and α0 and their interpretation. (The exception being that a physically consistent value of α is found for the CWV bin 1.5–2.0 cm.)

3.2.2. Clear Sky Calculations

[33] The above analysis suggests that the pristine simulated and the measured DLR fluxes have a small bias at worst, but that the aerosol direct effect is primarily responsible for the obs-calc flux differences. An alternative approach for comparisons with the AERONET AOTs was also pursued given the issues with fitting raised above. Under the assumption that both the calculated and observed DLR are accurate, the obs-calc residual flux can be used to estimate an aerosol loading for input into the RT code via (an equivalent of) equation (4). The resulting clear sky (aerosol-included) profiles lead to simulated DLR which are the same as those measured. These aerosol profiles can be converted into AOT values, and these AOTs directly compared with the AERONET values.

[34] The clear sky profiles are based on the pristine sky equivalent with the addition of one type of aerosol, on the basis of desert dust. In ES96, the aerosol profile is specified as mass mixing ratios per profile level and is estimated for each profile individually as follows. To begin with, an empirical profile is created

  • equation image

where p is profile level pressure, pm = 933 hPa and σ = 210 hPa, and c is an altitude-independent scaling factor which is initially arbitrary. The parameter values were chosen to produce a plausible vertical profile, in comparison with measurements made in Niamey during the DABEX aircraft campaign [Johnson et al., 2008]. It places 99% of the aerosol below 7.3 km and 50% below 1.7 km. Next, the value of c is determined such that the resulting obs-calc residuals are zero. The required scaling factor, cestimate, is calculated by a two stage process. First, for each profile the parameters A(t) and B(t) in the following function are fitted

  • equation image

using several values of cES whose upper limit results in a column mass of 25.6 × 10−3 kg m−2 (or, AOT at 11 μm of ∼10). Hence, FDLR;c=0 is the pristine value viz. section 3.2. From having estimates of these parameters, equation (7) can be inverted to obtain an estimate of c for a measurement of FDLR;AMF

  • equation image

The dust profile corresponding to the determined direct effect is then

  • equation image

Finally, from MMRdust;estimate a AOT at 11 μm is calculated. The fitting procedure was only carried out when a positive aerosol loading could be identified, i.e., if the uncertainty allowed for a negative loading then the creation of that clear sky profile was abandoned. This reduced the percentage of profiles by ∼10%. There are two further advantages to having calculated the clear sky profiles. First, the effects of aerosol can be evaluated at TOA against both measured radiance and derived flux products (see section 4). Second, the diurnal variations in dust loading [N'tchayi Mbourou et al., 1997] are more accurately estimated. Such time variations can only be validated by comparison with the AERI radiances (not attempted here).

[35] From the resulting profile-derived AOTs and the AERONET AOTs, scatter plots can be produced. These comparisons are made only when an AERONET retrieval exists within an hour of the RT profile time. (The frequency of the 0600 and 1700 UT times are therefore approximately half that of 1100 UT time and no comparison is made at 2300 UT). Figure 9 shows the scatter plots, separated into the same CWV bins as chosen before, and with linear fits to the data. The fit results are summarized in Table 5. The zero offsets in these fits are difficult to interpret given the lack of correlation with CWV. For a CWV bin to return a negative zero offset implies that the AERONET AOT is zero for some positive value of ES96-derived AOT, i.e., the calculated pristine flux is too low or the measured flux is too high, leading to a positive aerosol loading bias which artificially makes up for the shortfall. These conclusions are consistent with the values of α0 in Table 4, but again there is no relation to CWV.

image

Figure 9. ES96-derived AOT versus AERONET AOT (1020 nm). The lines correspond to the fits described in the text. The data are divided into the following four CWV bins: (a) 0–1 cm, (b) 1–1.5 cm, (c) 1.5–2 cm, and (d) 2.0 cm and greater.

Download figure to PowerPoint

Table 5. Parameters to the Fits and RMS Error of the Form AOTAERONET = AOT0 + S(AOTES96)a
CWV RangeAOT0SRMS (Wm−2)
Lower (cm)Upper (cm)
01−0.09 ± 0.041.3 ± 0.11.0
11.5−0.33 ± 0.052.1 ± 0.11.7
1.52.5−0.18 ± 0.051.7 ± 0.21.2
2.50.14 ± 0.051.2 ± 0.13.0

[36] Were the dust to be composed solely of relatively large particles (≥3 μm), then the Angstrom exponent could be expected to be near zero, and S consequently unity. As S is always larger than one, this is consistent with a positive Angstrom exponent which is consistent with the expected relation between (near-IR) AERONET AOT and aerosol profile (IR) AOT. The average Angstrom exponent, assuming a AOT of 1, can be calculated to be 0.16 ± 0.06 which is comparable with the previously found relation of 0.33 [Turner, 2008]. Seasonal changes in either the aerosol size distribution (which would particularly affect the AERONET retrievals) or the mineralogical makeup (affecting more the RT calculations) which are not considered here, but that can correlate with CWV via larger, synoptic, changes in monsoon meteorology, are likely to cause the variability in S.

[37] The independence of the AERI radiances allows the clear sky profiles to be analyzed as the pristine sky profiles were. The analogous scatter of obs-calc radiances difference against CWV is shown in Figure 7 (right). As expected, the correlation for bands 4 and 8 shows no difference when aerosol is included in the RT calculations. The scatter for band 5, however, changes significantly: the median obs-calc radiance difference is reduced from 2.1 Wm−2sr−1 to −0.2 Wm−2sr−1. The correlation, however, does show some relation with CWV; for residuals where CWV is 0–1 cm, the relative median residual is −14−12+16 % whereas for CWV 2–5 cm, the median is 3−4+4 %. Possible sources for these remnant radiance errors in band 5 may be from errors in the fitting process to derive aerosol loading, such as neglecting the spectral characteristics of the aerosol and the use of a fixed vertical profile, or because of errors in the parameterisation of the radiation code which are exposed at the low values of CWV found here.

[38] To summarize, the AMF measurements of DLR show no significant bias in comparison with a parameterisation, when the DLR is diurnally averaged. Correspondingly, the ES96 calculations of pristine DLR show no significant bias in comparison with calculations using the LBLRTM code. The obs-calc DLR flux differences are primarily accounted for by the aerosol direct effect, and the same conclusion applies to the radiance comparison with AERI measurements. When using AERONET AOTs as proxies for IR aerosol loading, there are, however, uncertainties in the fitting of flux residuals which suggests nonzero flux offsets. Similarly, using the presumed aerosol direct effect to derive a AOT also implies nonzero flux offsets, when compared with AERONET retrievals. These suggestions of biases in the DLR cannot be ruled out here, but are of the order of measurement error and so are not important for the purpose of this work. Using the derived aerosol loadings to compare IR window radiance with the AERI measurements results in radiance residuals with no significant bias overall, although there exists evidence for a relationship with CWV. It is therefore concluded that the pyrgeometer flux measurements, the AERI zenith radiance measurements, and the RT calculations from ES96, both pristine and clear sky, are all consistent and any error will not unduly bias the divergence calculations.

4. Top of Atmosphere Results

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Background
  5. 3. Surface Results
  6. 4. Top of Atmosphere Results
  7. 5. Longwave Atmospheric Divergences
  8. 6. Conclusions
  9. Acknowledgments
  10. References
  11. Supporting Information

[39] Having compared the ES96 calculations at the surface with observations, the equivalent analysis with TOA observations is presented here. The measured and simulated pristine outgoing longwave radiation is shown in Figure 10, and the obs-calc difference of these in Figure 11. The differences show far less structure in time than the equivalent surface comparisons, leading to the conclusion there are more factors responsible for the residuals at TOA than at the surface. It is necessary in the analysis to use averages over defined time periods in order to minimize random error.

image

Figure 10. The flux measurements (circles) with pristine simulations (crosses). Here OLR is outgoing radiation. The boxes run clockwise in time (0600, 1100, 1700, and 2300 UT). The relative paucity for the 23 00UT time is due to GERB sun avoidance operations.

Download figure to PowerPoint

image

Figure 11. The difference between measurements and calculations (cf. Figure 10). Here OLR is outgoing radiation. The boxes run clockwise in time (0600, 1100, 1700, and 2300 UT).

Download figure to PowerPoint

[40] The top of atmosphere upwelling flux (Outgoing Longwave Radiation (OLR)) is dependent on the emissivity and temperature of the atmosphere as well as the surface upwelling flux. From the surface analysis in the previous section, the atmospheric components (less aerosol) are found to be consistent with observed values to within the measurement errors. Therefore, initially the surface temperature input into the RT profiles is compared to determine whether the derived temperature or that measured by the IRT is more accurate. From establishing this, the second question regarding surface temperatures, namely that of scaling, can then be investigated. Finally, the pristine profiles are then considered to be defined and the questions of the effects of aerosol can be approached.

[41] For the Edition 1 fluxes, a known omission is that aerosol is not included in the theoretical RT calculations which are used to develop the database that leads to the radiance-to-flux conversion. As the surface radiation comparisons show, there are instances of large aerosol direct effects. Hence evaluating the angular isotropy of the radiance-to-flux conversion is useful before using either observation or RT calculation to calculate the atmospheric divergences.

4.1. Pristine Sky Calculations

[42] As the surface emissivities are fixed (0.93) in the profiles, the result of using either the derived LST, Tflux, or the measured LST, TIRT, as TAMF is first considered. In section 3, the comparison of Tflux with TIRT revealed that whilst a small difference is evident for most of 2006, between August and November there is a more significant variation which reaches a maximum difference of ∼4 K (or ∼5% relative difference in surface upwelling flux). By analysing the GERB ARG radiances with the RT calculations in pristine conditions, the difference between the radiances during August–November and during the other months will be an independent measure of which LST is more accurate.

[43] In Figure 12 (top), the difference of 1100 UT sonde launch surface temperatures between Tflux and TIRT is plotted for cloud-free times. This time is chosen partly because it corresponds to the largest surface temperatures, and so to increased surface contribution to TOA radiance. However, it also corresponds to when Banizoumbou AERONET AOT retrievals are always available (assuming no cloud cover) unlike the 0600 or 1700 UT times. The circled circles in the Figure 12 show when AOT at 1020 nm is smaller than 0.25 (representing half of the points shown) and so corresponds to the more pristine conditions.

image

Figure 12. (top) Difference of surface temperatures for clear skies at 1100 UT and when AERONET AOT (1020 nm) <0.25. (bottom) Difference of obs-calc TOA radiances, L, using three different surface temperatures.

Download figure to PowerPoint

[44] Figure 12 (bottom) shows the obs-calc TOA radiance residuals using the two surface temperatures in the simulations. During the months prior to the monsoon and in December, the radiance residuals are smaller when using TIRT compared to using Tflux. However, between August and December, the residuals are smaller when using Tflux. Overall, for the selected lower-AOT times, the residual from using TIRT is −3.4−2.3+1.2 Wm−2sr−1 and for using Tflux it is −3.3−1.4+0.9 Wm−2sr−1. As expected, the variance is reduced when using Tflux over TIRT.

[45] The relative magnitudes of the radiance residuals shows a similarity in time to the magnitudes of the surface temperature differences, with both maximized between days 250 and 300. To investigate whether surface temperature differences dominate the radiance residual, the analysis in this paragraph does not restrict the clear sky data by AOT value. Instead, the division is by the difference between TIRT and Tflux. When it exceeds 1 K, the TOA radiance residuals using first TIRT and then Tflux are −5.6−2.2+1.8 Wm−2sr−1 and −4.8−2.0+1.3 Wm−2sr−1 respectively. When the temperature differences is instead less than 1 K, the radiance residuals become −4.9−1.8+1.9 Wm−2sr−1 and −4.8−2.0+1.7 Wm−2sr−1. Consistent with the observation above, the median changes more when using TIRT than Tflux. Given these results, the assumption now made is that TAMF is Tflux.

[46] The final aspect of the surface temperature investigation is the effect of scaling TAMF to form TARG. Returning to the use of times with AOT below 0.25, the radiance residuals using TARG are −1.2−1.4+0.9 Wm−2sr−1, which as a percentage is −0.8−2.4+0.5 %. This is now much closer to the desired zero obs-calc radiance difference, and particularly so during December and January/February. Notably, during August–November, the change in using Tflux rather TIRT is as large as the subsequent effect of scaling the Tflux. For comparison, repeating the analysis but using fluxes rather than radiances results in a residual of −3.9−6.9+1.6 Wm−2 or −1.2−2.3+0.5 %. This change in the relative differences of the fluxes, compared to that from the radiances, will be analyzed more carefully in section 4.2.

[47] From the perspective of these TOA radiances, the pyrgeometer-derived surface temperature, Tflux produces residuals with smaller variance and greater consistency than using the IRT measurement, TIRT. As the pyrgeometer, the simulations and the GERB ARG radiances are all broadband, whilst the IRT operates in the infrared window, the hypothesis is that using the IRT retrieval to specify a broadband emission temperature is erroneous. However, the diurnal IRT values used to scale Tflux do not seem to retain this error so it is assumed that the averaging correctly reduces any window/broadband disagreements. Subsequently, using the spatial area scaling to form TARG from Tflux further reduces the TOA radiances residuals. This latter result then implies that the ULR measured by the AMF is biased positive (cf. Figure 5) relative to the ULR over the area of the GERB ARG pixel, as deduced from the MODIS LST results.

4.2. Clear Sky Calculations

[48] The limited number of pristine calculations in the previous section was dictated by both the need for near-contemporaneous AOT measurements and that aerosol-free times in Niamey were not common. Accordingly, more members of the data set can be obtained by using the clear sky profiles to calculate the TOA radiances and fluxes. For the previous selection of low-AOT times, with the clear sky profiles and using TARG as all following calculation do, the radiance residual becomes 0.6−2.0+0.6 Wm−2sr−1. Compared to the previous result, which assumed no aerosol, this result is now consistent with radiance closure at TOA.

[49] A more comprehensive comparison is made by extending the comparison to all cloud-free times, but then subdividing the obs-calc differences into both first and second dry periods and also the wet season, and then finally by time of day. The obs-calc difference is calculated for the GERB ARG radiances and fluxes, with the RT calculations including both the pristine and clear sky profiles. The comparison between the latter types of profile will show the effect of the aerosol upon TOA residuals, whilst using time of day as another subdivision will more fully analyze the effects of the diurnal cycle.

[50] The obs-calc differences are shown in Table 6, and the radiances are first examined. It is noticeable that in the dry seasons, the residual is smaller than in the wet season. For the pristine results there appears to be no distinct diurnal trend in the residual difference medians. However, for the clear sky results, there does appear to such a trend, with a noticeable overestimate of simulated radiances at 1100 UT relative to the other times in the dry season periods. The trend is not so clear for the wet season radiances. The diurnal variations of the radiance differences in the two dry periods are similar.

Table 6. Observed RT Calculations Difference, Δ(radn) = radnobs − radncalc, Results for the GERB ARG TOA Radiances, L, and Fluxes, Fa
TimePristineClear Sky
RadianceOLRRadianceOLR
ΔL (Wm−2sr−1)ΔL (%)ΔF (Wm−2)ΔF (%)ΔL (Wm−2sr−1)ΔL (%)ΔF (Wm−2)ΔF (%)
  • a

    Results are subdivided by use of pristine (aerosol-free) and clear sky (with aerosol) RT profiles, by time period (before (D1), during (W), and after monsoon (D2)), and by time bin. Differences are given as medians and interquartile range, for absolute values and percentages relative to the GERB data. The dashes denote when the number of data points for that time was fewer than 10% of the total number for that day.

D1
0600−2.0−1.3+0.8−2.2−1.4+0.8−9.9−2.9+2.0−3.6−1.1+0.7−2.0−0.5+0.8−2.2−0.5+0.8−7.7−2.6+2.9−2.9−0.8+1.2
1100−2.6−2.7+1.5−2.5−2.3+1.5−8.6−7.9+4.6−2.6−3.0+1.3−0.1−2.8+1.1−0.1−2.7+1.03.7−7.1+4.01.2−2.3+1.1
1700−3.3−1.8+2.1−3.2−2.1+2.1−13.1−4.3+6.2−4.5−1.6+2.2−1.4−2.0+1.7−1.4−2.1+1.7−1.4−6.4+4.5−0.5−2.3+1.5
2300
 
W
0600−3.6−2.4+0.6−3.9−2.6+0.7−13.8−7.0+1.9−4.8−2.8+0.5−3.9−2.0+1.0−4.0−2.5+1.0−12.8−7.2+1.8−4.6−3.0+0.7
1100−4.4−1.4+1.6−4.3−1.8+1.8−12.2−3.8+5.1−4.0−1.5+1.8−2.8−1.7+1.9−2.8−1.5+1.9−5.4−5.7+5.4−1.8−1.9+1.8
1700−3.8−1.2+0.9−3.9−1.4+1.0−12.7−2.5+2.7−4.4−1.1+0.8−3.0−1.1+1.1−3.2−1.2+1.3−9.6−3.5+3.7−3.4−1.3+1.3
2300−3.2−0.7+0.5−3.5−0.7+0.6−12.2−1.5+1.7−4.3−0.7+0.7−2.9−1.0+0.7−3.1−1.3+0.7−10.1−2.0+2.5−3.7−0.6+1.0
 
D2
0600−2.1−0.3+0.5−2.3−0.2+0.6−9.7−1.0+1.2−3.6−0.3+0.4−2.0−0.2+0.5−2.2−0.2+0.5−8.4−1.0+1.1−3.1−0.5+0.5
1100−1.5−1.4+1.0−1.4−1.4+0.9−6.1−4.2+3.1−1.8−1.5+0.90.4−1.7+0.50.4−1.6+0.53.9−4.9+1.81.2−1.5+0.5
1700−1.5−1.3+0.3−1.5−1.3+0.3−8.2−3.6+0.9−2.8−1.2+0.3−1.2−1.1+0.4−1.2−1.1+0.4−5.1−2.4+2.1−1.8−0.8+0.7
2300−1.6−0.9+0.4−1.7−0.9+0.4−8.4−2.8+1.3−3.0−1.0+0.4−1.5−0.7+0.6−1.7−0.8+0.6−7.1−2.0+1.6−2.5−0.8+0.6

[51] Comparisons of GERB and CERES LW radiances ([Dewitte et al., 2008] and [Clerbaux et al., 2009]) show that under equivalent clear sky conditions as used in this work, and within the whole GERB field-of-view, the ratio between the GERB and CERES (FM2 and FM3 only) satellite radiance measurement is ∼98–100%. The radiance residuals in the dry season when using the clear sky profiles are consistent with this level of uncertainty. However, the apparent trend suggests that either the aerosol profile specified in the RT profiles is incorrect, in particular, too much in the 1100 UT bin, or, that there remains an issue with the LST in the profiles. Because of the smaller relative diurnal trend in the wet season clear sky obs-calc differences compared to those in the dry periods, the LST issue is perhaps more likely. This is because of both smaller absolute value of LST and its reduced daily variation in the wet season.

[52] An analogous comparison using the fluxes leads to similar conclusions as with the radiance differences. However, the magnitude of the clear sky obs-calc diurnal trend is larger for the fluxes. If the radiance-to-flux conversion used in GERB Edition 1 processing and that in the ES96 code were equivalent then this would not be expected: the relative differences would not vary between comparing radiance and flux obs-calc values. The connection between the radiances, L, and fluxes, F, can be studied by the angular conversion factor, R

  • equation image

where (θ, ϕ) are the viewing angles between the satellite and the atmosphere, Ω represents the state vector of the atmosphere and surface, LOLR is the TOA radiance and FOLR is the flux. Here, because GERB and the AMF are fixed relative to each other, the angular coordinates remain constant (a view angle of 17.2°) and the changes are in Ω. Figure 13 shows R from ES96 calculations and the equivalents from GERB ARG plotted against the profile LST. (LST is a value in Ω and also a proxy for time since the coldest temperatures correspond to predawn and the hottest to near-noon, which are closest to 0600 and 1100 UT, respectively.) It is clear that the GERB and ES96 values of R follow a substantially linear relationship with LST. Relative to its linear component, the GERB ARG-derived values show the smallest variance. The ES96-derived values show that the addition of variable quantities of aerosol both increases the variance about its linear component and increases the anisotropy more rapidly with changes in LST. Using a linear fit for R, the following gradients and offsets result:

  • equation image

The gradients of the 1996 fits do not encompass that of the GERB fit, and so there is a disagreement in the radiance-to-flux conversion between the Edition 1 processing for GERB ARG and simulations. The differences indicate that it is not aerosol alone which cause disagreement in R. The consequence is that for equal clear sky radiances, the ES96 fluxes change less than GERB fluxes between the 0600 and 1100 UT, and the 1700 and 2300 UT time bins. The flux obs-calc differences would then show a larger variance over a day than the equivalent radiance differences. This can be seen most clearly in Table 6 for the percentage obs-calc differences in the two dry periods.

image

Figure 13. The angular conversion factor, R, from radiance to flux using GERB observations, and two types of ES96 calculation. (left) Without aerosol. (right) With aerosol. (Note that the view angle here is 17.2°.)

Download figure to PowerPoint

[53] This raises an important question: are the Edition 1 GERB ARG fluxes in the presence of a nonnegligible aerosol loading incorrectly retrieved, or is there an error in the calculations with the RT code when using the clear sky profiles? The former cannot be answered with this work, and the latter is beyond the scope of this work. Instead, it is possible here to analyze the magnitude of disagreement in R and its importance. This is via a third time series of OLR to accompany the GERB ARG Edition 1 and ES96 clear sky values: FOLR;ARG;R=ES96 is called the GERB/ES96 hybrid and is generated from the measured GERB ARG radiances, LOLR;ARG(t), and the contemporaneous ES96 angular conversion factor, Rclear;ES96(t)

  • equation image

where t are the cloud-free times. Figure 14 (top) displays the GERB ARG Edition 1, ES96 clear sky, and hybrid OLRs for the cloud-free times during 2006. The difference between the GERB and hybrid OLR is plotted in Figure 14 (bottom) for two times (0600 and 1100 UT). The dawn times show a smaller overall difference than the noon times, and with less variance. During the day, the hybrid OLR is smaller than the Edition 1 fluxes, and especially so in April when the CWV reached its minimum values and surface temperature were near their maximum. (The other equally dry periods in mid-November or late December were distinctly colder.) In contrast, the times of the March dust storm (days 66–70) do show a notable difference between the two OLRs, but not as pronounced as that for April. On 11 April 2006, the maximum difference of ∼10 Wm−2 was seen. This day had a combination of near annual maximum surface temperature, near annual minimum column water vapor, and above average dust loading (AOT(1020 nm) = 0.85 c.f. the 2006 average of 0.6). Overall, however, the difference between the Edition 1 and hybrid OLRs is smaller than the GERB measurement error.

image

Figure 14. (top) The GERB ARG Edition 1, ES96 RT simulation, and the hybrid GERB/ES96 OLRs. (bottom) The difference in OLR between the GERB ARG and GERB/ES96 hybrid OLR for near-dawn and noon times.

Download figure to PowerPoint

[54] Therefore the nature of the disagreement between the ES96-derived values of R and those from GERB ARG Edition 1 processing is due to several factors in the environment of Niamey. Aerosol is one component, but the difference between the GERB ARG and hybrid OLRs suggests that CWV and boundary layer temperature structures dominate. The exception is during intense aerosol loadings such as during the March 2006 dust storm over Niamey.

5. Longwave Atmospheric Divergences

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Background
  5. 3. Surface Results
  6. 4. Top of Atmosphere Results
  7. 5. Longwave Atmospheric Divergences
  8. 6. Conclusions
  9. Acknowledgments
  10. References
  11. Supporting Information

[55] The previous two sections have analyzed the surface and TOA radiation to establish the accuracy of the RT calculations. Although no significant biases have been found in the downwelling flux or TOA radiances, disagreements have been found. Therefore, an attempt is made to derive several estimates of the LW divergences across the atmosphere using both the measured and simulated fluxes in this work. The fluxes required are the surface upwelling flux (ULR), the surface downwelling flux (DLR), and the TOA upwelling flux (OLR). These can be obtained via the measurements alone, as in a companion paper [Slingo et al., 2009], from the ES96 calculations alone, or from what is termed a “best estimate” collection. The latter provides the ULR from the RT profiles that use a GERB ARG area LST, the DLR from the AMF measurements, and the OLR from the GERB/ES96 hybrid OLR. A summary of these is shown in Table 7. These are chosen to reduce the bias in the ULR and possible radiance-to-flux conversion errors in the OLR.

Table 7. Summary of the LW Flux Components Available for Divergence Calculationsa
FluxSourcesComment
  • a

    The bold entries denote those which are considered the best estimate (least biased and/or most accurate).

Surface down, FDLRAMF, ES96 (clear sky)none, defined equivalent to AMF
Surface up, FULRAMF, ES96biased with respect to GERB ARG area, unbiased with respect to GERB ARG area
TOA up, FOLRGERB ARG Edition 1, ES96 (clear sky), GERB/ES96hybridfrom aerosol-free R, using simple aerosol representation, GERB radiances and ES96R

[56] The divergence estimates using each flux collection (measurement only, simulation only, and best estimate) are given in Table 8 for the four sonde launch times. The diurnal trend is similar between each calculation, and the interquartile ranges do not typically exceed the difference between bin medians. This latter result corresponds to a key result in the companion paper [Slingo et al., 2009]: diurnally averaged divergence for days that were cloud free is −167−9+7 Wm−2 during 2006. In other words, the averaged divergence shows a relatively small variation between days, through 2006. In contrast, the amplitude of changes in divergence within a day is larger.

Table 8. Atmospheric LW Divergences for Each Time Bin Through 2006, From the Three Flux Component Collectionsa
Flux CollectionDivergence (Wm−2)
0600 UT1100 UT1700 UT2300 UT
  • a

    As medians and interquartile range.

Observations−200−13+10−111−13+12−189−10+11−198−15+9
Simulations−212−13+10−116−13+11−199−10+9−209−17+8
Best estimate−205−12+10−112−11+12−191−9+10−204−12+10

[57] In comparing the three divergence results with each other, the simulation divergences show the most cooling for each bin whereas the observation divergences show the least. Additionally, over the diurnal cycle, the amplitude of the former is larger than the latter. Comparing the best estimate divergences with those from observations, there is little difference at 1100 or 1700 UT, but the observations suggest less cooling at 0600 and 2300 UT. Overall, the best estimate divergences per time bin fall between those from the observations and the simulations. As percentages, however, the differences between the medians of best estimate divergences from the observation equivalents are ∼−1 to −3%. The primary causes here of the disagreements are the bias in ULR, and the partial cancellation of this by the use of hybrid OLR in the best estimate collection which causes the day-night disparity. The simulations overestimate the OLR in general, relative to GERB, and so this leads to greater cooling.

[58] The expected diurnal mean difference of divergence would then be typically ∼1%, or ∼2 Wm−2. In the context of GERB ARG and AMF flux measurement errors and biases, this difference is small. It does verify that the approach in the companion paper [Slingo et al., 2009], of using only the observed fluxes to derive diurnally averaged components and then study their variability as a function of other variables, is acceptable. However, to study more finely time resolved divergences, such as the 4 time bins here, the nature of flux components used can result in a bias which is larger than the measurement errors.

6. Conclusions

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Background
  5. 3. Surface Results
  6. 4. Top of Atmosphere Results
  7. 5. Longwave Atmospheric Divergences
  8. 6. Conclusions
  9. Acknowledgments
  10. References
  11. Supporting Information

[59] In this work, a comparison has been made between the observed radiative fluxes and radiances in the thermal region of the spectrum, at both the surface and top of atmosphere over spatial scales of ∼50 km. Utilising the high time resolution of measurements from both locations, aerosol- and cloud-free contemporaneous calculations were made at the four sonde launch times per day, for 361 days of 2006. At the surface, flux comparisons were made of the upwelling and downwelling components. The upwelling flux was found to differ substantially from the observed measurements, but this is primarily the result of modelling fluxes from the wider area in order to be consistent with the satellite footprint.

[60] The simulated, aerosol-free downwelling flux was found to behave principally as the observed values did, with the effects of changes in column water vapor and temperature being correctly calculated. The residual differences were found to be strongly correlated with column water vapor and consistent with the aerosol direct effect. This was verified by comparison of radiances from zenith-pointing observations in three wavebands, which separately verified the temperature and humidity profiles in the lower troposphere. A correlation of the flux residuals with visible AERONET optical depth retrievals verified that estimates of the direct effect, by using the measured and aerosol-free simulation fluxes, were not dominated by any systematic bias.

[61] A similar analysis was carried out between the top of atmosphere observations of radiances and fluxes and the simulated values. Since the observed fluxes are a product, a comparison of radiances was initially undertaken to establish the effect of the surface temperature transformation and surface emissivity specification. These showed an improvement in using the transformed and derived surface temperatures in comparison to use of the observed surface temperature from the AMF.

[62] Utilising the measured direct effect from the surface broadband fluxes to specify the aerosol loadings in a simplified vertical structure, a set of aerosol-included simulated radiances and fluxes at TOA was calculated. It was found that radiance closure could not be explicitly achieved but that the residual errors were small and, particularly during the dry season, approached those found in GERB-CERES comparisons. The comparison between simulated and observed radiances, and the same for fluxes, made evident a disparity that was attributed to the radiance-to-flux conversion process. The cause is hypothesized to be primarily due to CWV and boundary layer temperature structure, with aerosol playing a more minor role except during dust storms. In order to create a potentially more accurate time series of TOA flux, a hybrid OLR product was created using the RT calculations of radiance-to-flux conversion with the measured GERB ARG radiances.

[63] Under the cloud-free conditions assumed in this work, it became possible to calculate the atmospheric heating (divergence) using one of three flux collections: using observations from the AMF and GERB Edition 1 ARG only, using the ES96 simulated fluxes only, or using a combination of model and measurements. The latter is described as a “best estimate collection.” The three estimates, treated as averages for the 4 time bins throughout 2006, were found to show both similar variations diurnally, and similar interquartile ranges within each time bin. However, a comparison of the values per bin from the three estimates showed differences as large as 5%. Therefore for time resolved calculations, care must be taken in using consistent calculations, which here only truly applies to the simulated flux collection. The simulated fluxes were not intended to provide a definitive radiation closure, but did achieve consistency of surface and TOA radiation values through the year without significant bias.

[64] Future work using the RADAGAST data set can utilize the infrared retrievals of mineral dust [Turner, 2007], and associated retrievals of chemical speciation, to better characterise the aerosol environment. This will then allow a more detailed investigation into the apparent difference between radiances at both the surface and top of atmosphere. Validating the discrepancies found in TOA flux retrievals, particularly in the presence of aerosol which is shown to be near-omnipresent in Niamey during 2006, is another area which merits study especially given the prominent view of the deserts which the GERB instruments have. Establishing if the results presented here are consistent with those from other radiation codes, for differing satellite instrument view angles, and for different types of dust source, will be important in this task. Finally, the omission of clouds results in this work completing only one part of the task in RADAGAST and several opportunities subsequently present themselves, such as using the clear sky simulations to model the cloud forcing using the data or using cloud retrievals to directly model cloud effects within the radiative transfer code.

Acknowledgments

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Background
  5. 3. Surface Results
  6. 4. Top of Atmosphere Results
  7. 5. Longwave Atmospheric Divergences
  8. 6. Conclusions
  9. Acknowledgments
  10. References
  11. Supporting Information

[65] This paper is dedicated to the memory of Tony Slingo, whose proposal led to the AMF deployment to Niamey and thus made this work possible. His contribution to the work herein is immeasurable and he is sorely missed. N.A.B. thanks both Helen Brindley for fruitful discussion on GERB radiance-to-flux issues and the three anonymous reviewers for their constructive comments that improved this work. N.A.B. is funded by NERC grant NE/D002370/1. The GERB Edition 1 data were obtained from the GGSPS, Rutherford Appleton Laboratory, http://ggsps.rl.ac.uk/. The IR emissivity data were obtained from CIMMS, University of Wisconsin-Madison, http://cimss.ssec.wisc.edu/iremis/. The MODIS data are distributed by the Land Processes Distributed Active Archive Center (LPDAAC), located at the U.S. Geological Survey (USGS) Earth Resources Observation and Science Center (EROS), http://LPDAAC.usgs.gov/. The AMF data were obtained from the ARM archive, http://www.archive.arm.gov/, provided by the U.S. Department of Energy as part of the Atmospheric Radiation Measurement Program Climate Research Facility. The NWCSAF cloud mask was obtained via the EUMETSAT online archive, http://archive.eumetsat.org/. The AERONET data were obtained from http://aeronet.gsfc.nasa.gov/, with thanks to Didier Tanré (LOA, Université des Sciences et Technologies de Lille) and Rick Wagener (Brookhaven National Laboratory, New York) for their effort in establishing and maintaining the sites at Banizoumbou and Niamey.

References

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Background
  5. 3. Surface Results
  6. 4. Top of Atmosphere Results
  7. 5. Longwave Atmospheric Divergences
  8. 6. Conclusions
  9. Acknowledgments
  10. References
  11. Supporting Information

Supporting Information

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Background
  5. 3. Surface Results
  6. 4. Top of Atmosphere Results
  7. 5. Longwave Atmospheric Divergences
  8. 6. Conclusions
  9. Acknowledgments
  10. References
  11. Supporting Information
FilenameFormatSizeDescription
jgrd14991-sup-0001-t01.txtplain text document1KTab-delimited Table 1.
jgrd14991-sup-0002-t02.txtplain text document0KTab-delimited Table 2.
jgrd14991-sup-0003-t03.txtplain text document0KTab-delimited Table 3.
jgrd14991-sup-0004-t04.txtplain text document0KTab-delimited Table 4.
jgrd14991-sup-0005-t05.txtplain text document0KTab-delimited Table 5.
jgrd14991-sup-0006-t06.txtplain text document2KTab-delimited Table 6.
jgrd14991-sup-0007-t07.txtplain text document1KTab-delimited Table 7.
jgrd14991-sup-0008-t08.txtplain text document0KTab-delimited Table 8.

Please note: Wiley Blackwell is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing content) should be directed to the corresponding author for the article.