SEARCH

SEARCH BY CITATION

Keywords:

  • RADAGAST;
  • AMMA;
  • atmospheric radiation

Abstract

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Data Sources, Processing, and Errors
  5. 3. Time Series of Nonradiative Variables
  6. 4. Time Series of Radiative Fluxes
  7. 5. Radiative Flux Scatterplots
  8. 6. Radiative Divergences
  9. 7. Interpretation of the Longwave Fluxes and Divergences Using a Simple Model
  10. 8. Discussion and Conclusions
  11. Acknowledgments
  12. References
  13. Supporting Information

[1] Broadband shortwave and longwave radiative fluxes observed both at the surface and from space during the Radiative Atmospheric Divergence using ARM Mobile Facility, GERB data and AMMA Stations (RADAGAST) experiment in Niamey, Niger, in 2006 are presented. The surface fluxes were measured by the Atmospheric Radiation Measurement (ARM) Program Mobile Facility (AMF) at Niamey airport, while the fluxes at the top of the atmosphere (TOA) are from the Geostationary Earth Radiation Budget (GERB) instrument on the Meteosat-8 satellite. The data are analyzed as daily averages, in order to minimize sampling differences between the surface and top of atmosphere instruments, while retaining the synoptic and seasonal changes that are the main focus of this study. A cloud mask is used to identify days with cloud versus those with predominantly clear skies. The influence of temperature, water vapor, aerosols, and clouds is investigated. Aerosols are ubiquitous throughout the year and have a significant impact on both the shortwave and longwave fluxes. The large and systematic seasonal changes in temperature and column integrated water vapor (CWV) through the dry and wet seasons are found to exert strong influences on the longwave fluxes. These influences are often in opposition to each other, because the highest temperatures occur at the end of the dry season when the CWV is lowest, while in the wet season the lowest temperatures are associated with the highest values of CWV. Apart from aerosols, the shortwave fluxes are also affected by clouds and by the seasonal changes in CWV. The fluxes are combined to provide estimates of the divergence of radiation across the atmosphere throughout 2006. The longwave divergence shows a relatively small variation through the year, because of a partial compensation between the seasonal variations in the outgoing longwave radiation (OLR) and surface net longwave radiation. A simple model of the greenhouse effect is used to interpret this result in terms of the dependence of the normalized greenhouse effect at the TOA and of the effective emissivity of the atmosphere at the surface on the CWV. It is shown that, as the CWV increases, the atmosphere loses longwave energy to the surface with about the same increasing efficiency with which it traps the OLR. When combined with the changes in temperature, this maintains the atmospheric longwave divergence within the narrow range that is observed. The shortwave divergence is mainly determined by the CWV and aerosol loadings and the effect of clouds is much smaller than on the component fluxes.

1. Introduction

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Data Sources, Processing, and Errors
  5. 3. Time Series of Nonradiative Variables
  6. 4. Time Series of Radiative Fluxes
  7. 5. Radiative Flux Scatterplots
  8. 6. Radiative Divergences
  9. 7. Interpretation of the Longwave Fluxes and Divergences Using a Simple Model
  10. 8. Discussion and Conclusions
  11. Acknowledgments
  12. References
  13. Supporting Information

[2] Radiative processes are of crucial importance in determining the basic characteristics of the climate system, its stability and its response to perturbations. Understanding the radiation budget and what controls it is central to climate science, because changes in the budget are driving climate change and are also involved in the associated feedbacks [Randall et al., 2007]. In recent decades, satellite observations have provided an accurate description of the radiation budget at the top of the atmosphere (TOA) and its seasonal and interannual variability [Wong et al., 2006; Loeb et al., 2007]. In contrast, the radiation budget at the surface is less well observed, because almost all of the long-term observing sites are over land and their spatial distribution is very heterogeneous [Ohmura et al., 1998]. Retrievals of the surface radiation budget over the whole globe are now available from satellite data, using additional data sets and modeling, which have been validated against the surface sites [Zhang et al., 2004; Stackhouse et al., 2004]. Nevertheless, estimates of the divergence of radiation across the atmosphere, obtained by differencing the radiation budget at the surface from that at the TOA, show significant variations depending on the data sets employed. For example, estimates of the global and annual mean shortwave (solar) radiation absorbed by the atmosphere vary from 67 to 93 W m−2 [Kiehl and Trenberth, 1997; Ramana et al., 2007]. Significant differences also exist between estimates of the longwave divergence across the atmosphere, even for clear skies [Allan, 2006, and references therein].

[3] One approach to making progress in characterizing the atmospheric divergence and in understanding what controls it is to combine radiative fluxes and other measurements from a single, well-instrumented surface site with satellite data. Unfortunately, until recently, all broadband satellite radiometers have flown in low Earth orbits, so the availability of data over a given surface site has been limited in its temporal resolution and diurnal coverage. This situation improved dramatically with the launch of the first Geostationary Earth Radiation Budget (GERB) instrument on the Meteosat-8 European weather satellite in August 2002 [Harries et al., 2005]. GERB provides broadband shortwave and longwave fluxes every 15 min at about 50 km resolution within the Meteosat field of view. Building on this advance, a proposal was submitted to the U.S. Department of Energy's Atmospheric Radiation Measurement (ARM) Program to deploy their new ARM Mobile Facility (AMF) to Niamey, the capital of Niger in West Africa, to make coordinated measurements with GERB in collaboration with the African Monsoon Multidisciplinary Analysis (AMMA) experiment [Redelsperger et al., 2006], the field phases of which were due to take place in 2006. This proposal was called RADAGAST (Radiative Atmospheric Divergence using ARM Mobile Facility, GERB data and AMMA Stations). Further details on the background to the AMF and its deployment to Niger for the RADAGAST experiment are provided by Miller and Slingo [2007].

[4] The African continent provides a unique environment within which to study the physical processes that control the radiation budget. North Africa is the most important source of wind-blown dust in the world [Prospero et al., 2002] and the burning of vegetation in the dry seasons to clear the ground for crops can lead to high loadings of biomass aerosols. The extended drought that occurred in the Sahel in the 1970s and 1980s [Bell and Lamb, 2006] stimulated extensive research to discover the cause, which included the suggestion that radiative and biogeophysical feedbacks could be involved [Charney, 1975]. The climate of the Sahel is strongly influenced by the West African monsoon, which regulates the seasonal cycle of dry and wet seasons in response to the changing solar forcing. This important system was the focus of the AMMA observations in 2006. The International Science Plan for AMMA, which may be downloaded from the AMMA website at http://amma-international.org/, provides an introduction to the dynamics and meteorology of the West African monsoon and includes an extensive list of recent papers. In association with AMMA, several focused experiments studied various aspects of the monsoon. In particular, the Dust and Biomass-burning Experiment (DABEX) built on previous airborne campaigns to make in situ measurements of the radiative properties of wind-blown dust and of aerosols from biomass burning in the dry season [Haywood et al., 2008].

[5] An overview of the RADAGAST observations of the meteorology and thermodynamic variables is provided in a companion paper [Slingo et al., 2008] (hereafter part 1). Here, the focus is on the radiative fluxes both at the top of the atmosphere (TOA) and at the surface, and on the first direct estimates of the divergence of radiation across the atmosphere. The surface fluxes were measured by the AMF at the main Niamey airport site (13°29′N, 2°10′E). The TOA fluxes were measured by the GERB broadband radiometer on Meteosat-8, stationed above 0° longitude.

[6] In contrast to previous field studies that have invariably been of short duration, this paper presents the radiative fluxes and derived divergences for the whole of 2006, and investigates the influences of controlling factors such as temperature, water vapor, clouds and aerosols. These results are the first estimates of broadband radiative divergences using both surface and satellite observations that are available both at high temporal resolution and for such an extended period of time. The philosophy of the approach taken in the analysis is to avoid the use of detailed radiative transfer modeling as much as possible, since this is the remit of some of the following papers in this special section. Rather, the emphasis is on presenting the time series of the various radiative quantities and to use scatterplots, simple models and to a limited degree the results from the other studies to identify the main factors that control the fluxes and divergences.

[7] Many difficult sampling issues arise when bringing together measurements from the surface and from space to calculate atmospheric flux divergences above a single location [Settle et al., 2008]. We minimize these issues in this first study of the divergences from RADAGAST by using daily averaged data and by analyzing only the divergences across the atmosphere as a whole, as opposed to the vertical structure of the divergences within the atmosphere. As a result, the effects of residual sampling uncertainties and of errors in the calibration of the instruments are much smaller than the magnitude of the divergences themselves, so they have a much less damaging impact on the analysis than would be the case for data with higher time and space resolution.

[8] In the following section, the data used and the processing are described. Section 3 presents time series of the screen-level temperature and column water vapor (CWV) data analyzed in part 1 and of cloud and aerosol retrievals, which are used later to investigate their influence on the fluxes and divergences. Time series of the radiative fluxes from the AMF at the surface and from GERB at the TOA are analyzed in section 4. Scatterplots are used in section 5 to investigate the relative roles of temperature, water vapor, clouds and aerosols on the fluxes. The derived divergences are presented in section 6, both as time series and as scatterplots, which are used to explore the factors that control the divergences. The longwave fluxes and divergences are analyzed further in section 7 using a simple model of the greenhouse effect. Finally, section 8 summarizes and discusses the main results of this work.

2. Data Sources, Processing, and Errors

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Data Sources, Processing, and Errors
  5. 3. Time Series of Nonradiative Variables
  6. 4. Time Series of Radiative Fluxes
  7. 5. Radiative Flux Scatterplots
  8. 6. Radiative Divergences
  9. 7. Interpretation of the Longwave Fluxes and Divergences Using a Simple Model
  10. 8. Discussion and Conclusions
  11. Acknowledgments
  12. References
  13. Supporting Information

2.1. Radiative Fluxes

[9] Upwelling and downwelling longwave (thermal) and shortwave (solar) radiative fluxes measured at the surface were obtained from the radiometers at the main Niamey airport site. Details of the AMF instruments and of their characteristics are given by Miller and Slingo [2007]. The total downwelling shortwave radiation was calculated using the component sum method [Michalsky et al., 1999], whereby the direct flux measured normal to the direction of the sun is multiplied by the cosine of the solar zenith angle and added to the measured diffuse flux. This provides a more accurate estimate of the total downwelling flux than a measurement of the sum of the direct and diffuse fluxes by a single radiometer.

[10] Outgoing longwave radiation (OLR) and reflected shortwave radiation at the TOA were obtained from the GERB instrument on Meteosat-8 [Harries et al., 2005]. During the experiment, Version 3 ARG (Averaged, Rectified and Geolocated) data from GERB were retrieved from the Royal Meteorological Institute of Belgium and were displayed on the project website in near real time, with a latency of about 40 min. After about 40 days, these data were replaced by quality-controlled Edition 1 ARG data, produced by the GERB Ground Segment Processing System (GGSPS) [Harries et al., 2005; Dewitte et al., 2008], which are used here. Documentation on the GGSPS may be downloaded from http://ggsps.rl.ac.uk/information.html#Publications. The GERB instrument was turned off for periods of several hours or longer on 16 individual days in 2006 and for a longer period from 24 September to 9 October, because of autonomous or commanded safety shutdowns. In addition, the instrument was routinely shut down for up to 6 hours around midnight during the eclipse seasons at the equinoxes, to prevent direct sunlight from damaging the detectors [Harries et al., 2005]. In 2006, these periods covered 9 February to 23 April and 18 August to 30 October.

[11] All of the radiative fluxes shown here are daily means, which simplifies combining the ARM point measurements with the GERB area-average measurements. The averaging process removes the diurnal variability and the high-frequency noise that arises from small-scale cloud features advecting over the AMF and through the GERB pixels. It also simplifies investigations of the influence of the meteorological changes on the fluxes and divergences over a timescale of days and longer, which is the primary focus of this work.

2.2. Sources of Error in the Radiative Fluxes and Divergences

[12] The errors in the daily mean fluxes and divergences arise from three main sources, which are considered below. Combining these sources enables an estimate of the overall error to be made.

2.2.1. AMF and GERB Fluxes

[13] This source of error includes the instruments themselves and the processing of the measurements to produce fluxes. Uncertainties in the AMF fluxes are given by Slingo et al. [2006] as 9.0 W m−2 for the total downwelling shortwave, although this could be as large as 25 W m−2 at the peak of the March 2006 dust storm, and 5.1 W m−2 for the downwelling longwave. Estimates of the absolute accuracy of the GERB radiances are given by Slingo et al. [2006] as 2.25% in the shortwave and 0.96% in the longwave. They note that angular models introduce additional errors of 10 W m−2 for typical shortwave fluxes and 5 W m−2 for longwave fluxes, with the expectation that the latter will be an underestimate for aerosols, which are not (as yet) identified in the processing system. Results shown by N. A. Bharmal et al. (Simulation of surface and top of atmosphere thermal fluxes and radiances from the RADAGAST experiment, submitted to Journal of Geophysical Research, 2009) are consistent with this conclusion. Dewitte et al. [2008] analyze comparisons between collocated GERB and CERES data and conclude that the GERB accuracy is 5% for reflected shortwave radiances and 2% for the emitted longwave radiances. For typical daily mean shortwave and longwave fluxes shown later of 100–200 W m−2 and 300 W m−2, respectively, these numbers correspond to 5–10 W m−2 and 6 W m−2, consistent with the estimates from Slingo et al. [2006] quoted above.

2.2.2. Interpolation Across Missing Data

[14] The most serious source of missing data for the daily means is the shutdown of GERB during the eclipse seasons, with the loss of up to 6 hours of data every night. Shortwave data are also masked as unavailable during the sun-glint period of about 1 hour around 1130 local time. A simple linear interpolation of the data across these periods was applied. The impact was quantified by applying the interpolation to days when data were present and comparing the actual and interpolated daily means. For the interpolation at night, the impact on the OLR was −0.1 ± 0.8 W m−2 for clear days and 0.1 ± 2.5 W m−2 for cloudy days. The small systematic component reflects the fact that the OLR typically changes quite slowly through the night, at least for clear skies, and this behavior is approximately linear [e.g., Comer et al., 2007]. For the shortwave interpolation, the impact on the reflected shortwave was −5 ± 5 W m−2 for clear days and −4 ± 17 W m−2 for cloudy days. The small negative bias is due to the linear interpolation. Extended periods during which GERB was turned off for other reasons could not be dealt with in this way (notably 24 September to 9 October), so these days were excluded from the analysis altogether.

2.2.3. Inhomogeneities and Spatial Sampling

[15] Settle et al. [2008] analyze the sampling errors that arise from using the AMF measurements at a single location to represent the surface radiation budget beneath GERB. For the shortwave fluxes, there is a systematic error from geographical variability of the surface albedo of about 10 W m−2, and a random error, mainly from clouds, of about 3 W m−2 in the dry seasons, rising to 12 W m−2 at the peak of the wet season in August. For the longwave fluxes, there is a systematic error from geographical variability of the surface temperature of about 4 W m−2 and a random error, mainly from clouds, of about 3 W m−2.

2.2.4. Estimate of the Errors in the Divergences

[16] The divergences are calculated as the difference between the TOA and surface radiation budgets, so the errors discussed above are cumulative. In combining the errors, we make the usual assumption that the random errors from different instruments or observing systems are uncorrelated. In reality, there may be some correlation and it is rarely possible to be entirely sure how much of a stated error is systematic and how much random. These estimates hopefully err on the side of caution and should therefore be regarded as guidelines rather than definitive.

[17] In the shortwave, we combine the GERB error of ±10 W m−2 with the AMF error of ±9 W m−2 for each of the two radiometers used to calculate the net flux at the surface, together with the interpolation and sampling errors mentioned above. The overall random error in the shortwave divergences is then ±16 W m−2 in the dry seasons and ±20 W m−2 in the wet season, with a systematic sampling error of about 10 W m−2. The random error is about 20% of the shortwave divergences shown later, which illustrates the difficulty in making definitive estimates of this quantity from observations.

[18] In the longwave, the GERB error of ±6 W m−2 is combined with the AMF error of ±5.1 W m−2 for each radiometer and the sampling error of ±3 W m−2 to produce an estimate of the uncertainty in the longwave divergences of ±10 W m−2, with a possible systematic error of about 4 W m−2. In this case, the random error is only about 6% of the longwave divergences shown later, which are thus known with much greater accuracy than the shortwave divergences. The comparisons shown by Bharmal et al. (submitted manuscript, 2009) between divergences calculated from the data and from radiative transfer simulations produce similar values, in support of these estimates.

2.3. Clouds

[19] Two cloud masks are available for the analysis of the fluxes. The first is based on the operational CLMK (CLoud MasK) product generated by the Nowcasting Software Application Facility using SEVIRI (Spinning Enhanced Visible and Infrared Imager) data on Meteosat-8 [Schmetz et al., 2002]. This is available throughout the period from the EUMETSAT on-line archive (http://archive.eumetsat.org). The CLMK data are at the same spatial and temporal resolution as the SEVIRI imagery (3 km and 15 min). They are derived from a cloud classification product, which uses a two step process to first identify clouds, and then classify them [Derrien and Le Gléau, 2005]. The cloud identification and classification processes both use a series of tests based on multispectral thresholds, where the threshold values are determined dynamically depending on illumination conditions and geographical location, climatological data, numerical weather prediction data and radiative transfer models. At night time only those tests that use the thermal infrared SEVIRI channels are available, with the result that the cloud classification is less reliable: for example, airborne dust is sometimes indistinguishable from low warm clouds. For the AMF Niamey airport site, regions of 17 by 17 pixels (covering a GERB ARG pixel) were extracted from the CLMK data, whose spatial average formed the time series of the TOA estimate of cloud cover.

[20] A second cloud mask was derived from the cloud fractions retrieved by P. Kollias et al. (Cloud and precipitation observations during the 2006 ARM Mobile Facility deployment in Niamey, Niger, Africa, submitted to Journal of Geophysical Research, 2009) from the W-band (95 GHz) cloud radar, micropulse lidar and ceilometer at the main AMF airport site. These are vertically pointing instruments with narrow fields of view, so the cloud mask is only representative of the atmosphere directly over the Niamey airport site. Unfortunately, the data do not cover the entire year, because the cloud radar only became operational in mid-March. Daily mean cloud fractions were calculated as vertical means of the daily mean values at all levels from Kollias et al. (submitted manuscript, 2009).

[21] The results shown later were not particularly sensitive to which cloud mask was used to sample the data for clear and cloudy conditions. However, since the SEVIRI mask relates directly to the GERB footprint and is available for the whole year, it was decided to us this exclusively in this paper. A daily average of the cloud mask was created in order to provide a binary flag for conditionally sampling the radiation data, and days were flagged as cloudy if this average value exceeded a threshold of 0.3.

2.4. Aerosols

[22] Retrievals of aerosol optical thickness were obtained from the AERONET site at Banizoumbou, approximately 60 km east of Niamey airport [Holben et al., 2001]. The data from this site represent the most comprehensive retrievals for this area that are available throughout the AMF deployment. A CIMEL sunphotometer, as used by AERONET, was also deployed at the AMF Niamey airport site, but was only present from August 2006 until the end of the deployment, so the Banizoumbou data are used here. The data from the two instruments show a high degree of correlation (r = 0.91 (Bharmal et al., submitted manuscript, 2009)), indicating that the Banizoumbou data are representative of conditions at the airport. The data used are the diurnally averaged AERONET version 2 direct sun inversions of aerosol optical thickness at 870 nm. These data have the advantage of being cloud screened with associated error estimates. Turner [2008] also uses AERONET data from this site to compare with optical thicknesses retrieved from the AERI infrared window radiances and confirms that the data correlate well with the infrared optical thicknesses.

3. Time Series of Nonradiative Variables

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Data Sources, Processing, and Errors
  5. 3. Time Series of Nonradiative Variables
  6. 4. Time Series of Radiative Fluxes
  7. 5. Radiative Flux Scatterplots
  8. 6. Radiative Divergences
  9. 7. Interpretation of the Longwave Fluxes and Divergences Using a Simple Model
  10. 8. Discussion and Conclusions
  11. Acknowledgments
  12. References
  13. Supporting Information

[23] In this and the subsequent sections, the wet season is defined as in part 1: from 5 May to 29 October. The period before 5 May is referred to as the first dry season and that after 29 October as the second dry season. Figure 1 shows time series of the variables used to analyze the radiative fluxes in the following sections. Figures 1a and 1b show the screen-level temperature and CWV data discussed in part 1, which are included here for reference in the subsequent analysis.

image

Figure 1. Time series of daily mean (a) screen-level temperature (°C) at the AMF Niamey airport site, (b) column water vapor (centimeters) derived from the microwave radiometer at the AMF Niamey airport site, (c) cloud fraction from the CLMK cloud mask derived from SEVIRI data for the GERB ARG pixel over Niamey, and (d) aerosol optical thickness at 870 nm from the AERONET site at Banizoumbou. The abscissa shows the day number in 2006, and the dashed vertical lines denote the boundaries between the calendar months, the first letters of which are indicated at the top of each plot.

Download figure to PowerPoint

3.1. Clouds

[24] Figure 1c shows the mean cloud fractions derived from the SEVIRI cloud mask. In the first dry season, cloud amounts were low, with mainly thin cirrus associated with the subtropical jet stream to the north or with convection to the south of Niamey, plus occasional thin midlevel clouds that formed at the top of the aerosol-laden boundary layer at around 5 km altitude. Cloud amounts increased with the arrival of the wet season and most cloud occurred in July, August and September, associated with the summer monsoon. The early and late monsoon periods were relatively clear in comparison, probably owing to the moist monsoon air layer being too shallow to support deep convection, although the onset and retreat of the monsoon were still apparent in the change in the amount and frequency of occurrence of cloud. The summer monsoon brought in the moisture necessary to support deep, moist convection, and so summer cloud types were more variable, consisting not only of thin cirrus, but also thicker low and midlevel convective cloud. After the retreat of the monsoon in October, cloud amounts returned to small amounts of cirrus, with fewer lower level clouds than in the first dry season.

3.2. Aerosols

[25] Aerosol loadings were significant throughout the year (Figure 1d), with several major events in addition to the previously documented dust storm in early March [Slingo et al., 2006]. Apart from these individual dust outbreaks, the background aerosol optical thickness remained at around 0.3 for the first half of the year, falling gradually during the late summer monsoon period to around 0.2, and reaching the lowest levels during the second dry season. The largest sustained values were in the wet season, particularly in June, when dust was raised by the gust fronts propagating away from deep convection. The reduction in background levels late in the monsoon may have been due to the wetting of the surface and the growth of vegetation following rainfall, both of which would have reduced the amount of dust available to be lifted by the wind. The lowest aerosol loadings occurred in the second dry season, providing good conditions for testing clear-sky radiation simulations [McFarlane et al., 2009; Bharmal et al., submitted manuscript, 2009].

4. Time Series of Radiative Fluxes

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Data Sources, Processing, and Errors
  5. 3. Time Series of Nonradiative Variables
  6. 4. Time Series of Radiative Fluxes
  7. 5. Radiative Flux Scatterplots
  8. 6. Radiative Divergences
  9. 7. Interpretation of the Longwave Fluxes and Divergences Using a Simple Model
  10. 8. Discussion and Conclusions
  11. Acknowledgments
  12. References
  13. Supporting Information

[26] This section presents the radiative fluxes measured at the surface by the AMF and at the TOA by GERB. The fluxes are presented first as time series to show the variation over the course of the year, and are then shown plotted against the dominant influences on this variability to illustrate the main controls on the radiation fields. These influences include screen-level temperature and CWV (presented in part 1 and repeated here in Figure 1), as well as clouds and aerosols. The CWV data used here are from the microwave radiometer (see part 1 for details). Understanding the controls on the radiative fluxes is essential for interpreting the radiative divergence results, to be discussed in section 6, as the divergences are calculated from the component fluxes.

4.1. Surface Longwave Fluxes

[27] Figure 2 shows the surface downwelling, upwelling and net longwave radiative fluxes measured by the AMF at Niamey airport. Downwelling longwave radiation (DLR, Figure 2a) increased overall through the first dry season by about 30 W m−2, but periodic increases above the base level of over 50 W m−2 occurred during this period, coinciding with dry season cloud and water vapor advection events, as well as the dust storm in early March. The onset of the monsoon can be seen clearly in mid-April, when the DLR showed a sudden increase with the arrival of more humid air. This was referred to as a “false onset” in part 1, as it was followed by a break, the main monsoon arriving only at the beginning of May. The DLR reached a maximum during the early part of the wet season, when cloud cover was increasing (Figure 1c) and the atmosphere was both warm and humid (Figures 1a and 1b). It then dropped by over 20 W m−2 in the later and cloudier part of the monsoon, when the increased cloud cover and hence lower shortwave insolation (see below), coupled with greater evaporation from the surface [Miller et al., 2009], led to lower surface and atmospheric temperatures. The end of the monsoon at the end of October was associated with a precipitous drop in the CWV (Figure 1b), which produced a corresponding reduction in the DLR of over 50 W m−2. In the second dry season, the strong signals from variations in CWV (Figure 1b) and in aerosol loadings (Figure 1d) are once again apparent. The lowest fluxes occurred at the end of the year, when the surface and air temperatures reached their minimum values.

image

Figure 2. Time series of daily mean surface (a) downwelling, (b) upwelling, and (c) net longwave flux (W m−2), measured at the AMF Niamey airport site. The abscissa shows the day number in 2006, and the dashed vertical lines denote the boundaries between the calendar months, the first letters of which are indicated at the top of each plot.

Download figure to PowerPoint

[28] While the behavior of the downward thermal fluxes described above appears straightforward, it actually results from a very unusual combination of changes in temperature and humidity. As discussed in part 1 and shown here in Figures 1a and 1b, surface and air temperatures increased during the first dry season to a maximum at the start of the monsoon in May. Temperatures then decreased to the peak of the monsoon in August, recovered in September/October, before dropping steadily toward the end of the year. In contrast, near-surface humidities and the CWV decreased through the first dry season, then increased as the monsoon arrived and reached their highest values at the peak of the monsoon. They decreased in September/October and dropped to low values again at the end of the year. Apart from in the second dry season, the seasonal changes in temperature and humidity were thus out of phase with each other. The reason is that humidity variations in this region are controlled by the wind direction and by the origin of the air mass (and hence by the large-scale atmospheric dynamics) more than by the temperature. This aspect of the climatology of the region is well known, but the impact on the longwave radiative fluxes and divergences is remarkable, as discussed later. As far as we are aware, this behavior has not been documented before.

[29] The upwelling longwave radiation (Figure 2b) is determined by the surface temperature and emissivity, with changes in the latter being less important [Settle et al., 2008]. The surface temperature was in general slightly higher than the air temperature, but the two were closely coupled, so that comparison of Figure 2b with the time series of air temperature (Figure 1a) shows that both the low- and high-frequency variations in the upward flux closely followed those of the temperature.

[30] The net longwave radiation at the surface (Figure 2c) is also dependent on temperature, as is shown later, but the most consistent influence appears to be from the CWV, as may be seen by comparing with Figure 1b. The largest net longwave thus occurred at the end of the first dry season, when the CWV was at its lowest value, which allowed the surface to cool more efficiently out to space. The net longwave dropped progressively through the wet season, reaching a minimum in August when the CWV reached its maximum. At this time, the atmosphere was at its most opaque and so the surface cooling was minimized. The net longwave increased again during the second dry season, as the CWV fell again. Cloud events always reduced the net longwave, particularly so during the dry seasons when the low CWV led to values of the cloud radiative effect that reached several tens of W m−2, even from the middle level and high clouds present at this time.

4.2. Surface Shortwave Fluxes

[31] The incoming shortwave radiation (Figure 3a) increased with the seasonal increase in insolation at the TOA through the first dry season until near the end of April, when the sun passed overhead at Niamey (Figure 4c). During the monsoon, solar radiation received at the surface decreased substantially, mainly because of the increased cloudiness but also, as will be shown later, because of water vapor absorption. The incoming shortwave was thus lower when the sun passed overhead for the second time in late August, compared with April, and continued to decrease thereafter as the sun moved further south. Comparison with Figure 1 shows that both clouds and aerosol caused large reductions all year round from the clear sky values, but that these became more frequent in the wet season, particularly during the cloud maximum between July and September. The impact of the dust storms evident in Figure 1d, such as that in early March and the extended feature in mid-June, can also be seen clearly. These are the main causes of reduced solar transmission to the surface during the dry season, apart from the impact of two cloud events during the first dry season. The onset and retreat of the monsoon were not so clearly marked in the solar than in the thermal, and the retreat appeared to occur almost a month sooner. This is because the main effect of the monsoon on solar fluxes is the impact of cloud, with a smaller impact of humidity. In the thermal region, the humidity of the monsoon air layer is much more important. The build up of monsoon cloud happened gradually, whereas the arrival of the monsoon air brings a sudden change in water vapor content. In addition, the area of convection moves south sooner than the moist air layer since it becomes shallower toward the north, therefore becoming too shallow to support cloud late in the monsoon season.

image

Figure 3. Time series of daily mean surface (a) downwelling, (b) upwelling, and (c) net shortwave flux (W m−2), measured at the AMF Niamey airport site. The abscissa shows the day number in 2006, and the dashed vertical lines denote the boundaries between the calendar months, the first letters of which are indicated at the top of each plot.

Download figure to PowerPoint

image

Figure 4. Time series of daily mean top of atmosphere (a) outgoing longwave radiation (OLR) measured by GERB, (b) reflected shortwave radiation measured by GERB, and (c) calculated incoming shortwave radiation (all in W m−2). The abscissa shows the day number in 2006, and the dashed vertical lines denote the boundaries between the calendar months, the first letters of which are indicated at the top of each plot.

Download figure to PowerPoint

[32] The reflected shortwave radiation at the surface (Figure 3b) shows the same basic pattern, with the additional effect of changes in the surface albedo [McFarlane et al., 2009]. The reduction in albedo during the wet season due to the higher average sun angle and the darkening of the surface due to rainfall and vegetation growth reduced the reflection at this time, amplifying the seasonal changes seen in the incoming shortwave.

[33] The net shortwave (Figure 3c) shows that the absorption of solar radiation by the surface was a maximum in the early period of the wet season, but was reduced substantially by cloud from July through September. This reduction in the insolation drove the drop in surface and air temperatures through the monsoon, documented in part 1. The signals of dust events and of cloud at all times of the year, apparent in the component fluxes, can also be seen in the net radiation.

4.3. Top of Atmosphere Fluxes Measured by GERB

[34] Figure 4a shows that the OLR increased during the first dry season and then dropped by over 50 W m−2 as the monsoon developed, recovering during the second dry season. These changes were driven in part by the same temperature and humidity variations discussed above. The OLR is most sensitive to humidity changes in the middle and upper troposphere, but Figure 8b in part 1 shows that these changes were correlated with changes in the CWV, which is used here. In the case of the OLR, these variations mostly acted in the same direction, so that OLR increased both because of increasing temperature and of decreasing CWV during the first dry season, and similarly decreased in the wet season as the CWV increased and the surface and lower tropospheric temperatures fell. This picture is complicated by clouds and to a lesser extent aerosol outbreaks, both of which acted to reduce the OLR. The cloud effect is apparent throughout the year, although the deep convective cloudiness during the monsoon consistently produced the lowest OLR values. Following the monsoon retreat, OLR rose as the atmosphere again became more transparent to thermal radiation. Toward the end of the year, several exceptionally clear and dry periods allowed the OLR to rise, countering the effect of the fall in surface temperature.

[35] The annual variation in the reflected solar radiation at the TOA (Figure 4b) is controlled by the insolation (Figure 4c), which was a minimum at the beginning of the year when the sun was furthest south and a maximum when the sun passed overhead in late April. The main cause of short-term albedo increases was the presence of bright, thick cloud. This caused large increases in reflection above the baseline curve, particularly in the later, cloudier phase of the monsoon when bright monsoon weather systems passed over the AMF site frequently, but also on one occasion in late January when unusually thick cloud was present.

5. Radiative Flux Scatterplots

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Data Sources, Processing, and Errors
  5. 3. Time Series of Nonradiative Variables
  6. 4. Time Series of Radiative Fluxes
  7. 5. Radiative Flux Scatterplots
  8. 6. Radiative Divergences
  9. 7. Interpretation of the Longwave Fluxes and Divergences Using a Simple Model
  10. 8. Discussion and Conclusions
  11. Acknowledgments
  12. References
  13. Supporting Information

[36] The discussion in section 4 shows that several relationships are apparent between the radiative fluxes and the meteorology from comparisons of the temporal variations. However, it is possible to look more closely at these relationships by plotting the radiation data against their controlling meteorological variables, and thereby gain more insight into the behavior of the fluxes that determine the radiative divergences. The longwave radiative fluxes are discussed first, followed by the shortwave.

5.1. Longwave Scatterplots

[37] In section 4.1, it was argued that temperature and CWV have opposing influences on the DLR at the surface. In this section, these influences are examined further, with guidance from the simple formula developed by Prata [1996], which parameterizes the clear-sky DLR in terms of these two variables. Prata's formula is based on an analytic approximation to represent the dependence of the effective emissivity of the atmosphere on the CWV. In the paper, it was tested using both radiative transfer calculations and data sets of simultaneous measurements of air temperature, vapor pressure and DLR. The formula reproduces the dependence of the clear-sky DLR on temperature and CWV found in various global data sets [Allan, 2006], and it therefore provides a useful background for the analysis of the Niamey data.

[38] Prata's formula for the DLR at the surface may be written as

  • equation image

where σ is the Stefan-Boltzmann constant and T is the screen-level temperature. The effective emissivity of the atmosphere ɛ is approximated as

  • equation image

where u is the CWV in units of g cm−2. The values of the DLR are thus calculated from two variables, the screen-level temperature T and the column water vapor u.

[39] The symbols on Figure 5 (top) show the values of the daily mean DLR measured at the Niamey airport site scatterplotted against the observed screen-level temperature (Figure 5a) and against the observed CWV (Figure 5c). The symbols are colored according to whether the observations were made during the dry or wet seasons. In addition, the dotted lines show the values of DLR predicted by equations (1) and (2), where in each plot the dependence on either temperature or CWV is shown for several fixed values of the other input variable. This provides a series of idealized “Prata curves” that illustrate the dependence of the DLR on only one variable, with the other held constant. They should not be interpreted as simulations of the actual data points, but merely as an aid to the analysis below. When the Prata formula is provided with both of the observed values of temperature and CWV, the simulated fluxes agree quite well with the data, as was shown in the analysis of the March 2006 dust storm [Slingo et al., 2006, Figure 3].

image

Figure 5. Scatterplots of daily mean surface downwelling longwave flux (W m−2) against (a) screen-level temperature (°C) and (c) column water vapor (centimeters). All data were measured at the AMF Niamey airport site. The data points are color-coded according to whether they were measured during the dry (red) or wet (blue) seasons. Open squares denote days with cloud, according to the SEVIRI cloud mask. The dotted lines show the fluxes given by the Prata formula (equations (1) and (2)) for several values of the column water vapor in Figure 5a and for several values of the screen-level temperature in Figure 5c. (b and d) Simplified representations of the trajectories of the measurements through the year (see text).

Download figure to PowerPoint

[40] Figure 5 (bottom) shows idealized versions of the trajectories of the observed values of the DLR through the year, which summarize the seasonal variations of the observations and their dependence on temperature and CWV.

[41] Figure 5a shows that the DLR increases with air temperature, as expected, but two complications are apparent. Firstly, the wet season fluxes are much higher than those during the dry season at the same air temperature, because the CWV (and hence the effective emissivity of the atmosphere) was much larger at this time of year. Secondly, the slope of the wet season fluxes is much lower than that during the dry season and the data points cross the Prata curves. This is because the lowest temperatures occurred in August, when the CWV was a maximum, whereas the highest temperatures in May were associated with somewhat lower CWV, as noted earlier. In addition, there was much more cloud late in the season, which further increased the emission to the surface at that time. Taken together, these factors lead to a much lower slope in the wet season, as the data points move across the Prata curves for constant values of CWV.

[42] Figure 5b illustrates the trajectory of the data points. At the beginning of the year, temperatures were around 25°C and increased through the first dry season to around 35°C whilst the DLR points moved onto a slightly lower Prata curve. With the monsoon onset, temperatures remained high but the DLR points jumped up to the 4-cm CWV Prata curve, after which the temperatures decreased. In late summer, the CWV was at its highest, but the temperature continued to fall, moving the DLR down the highest Prata curve. After the monsoon retreat, the temperature returned to around 30°C, but the CWV fell, moving the DLR back to the lower Prata curves. As the year came to an end, the falling temperatures and CWV brought the DLR to its lowest values by moving it further across the Prata curves. Cloud had additional effects on the DLR, but did not change this basic result.

[43] The dependence of the DLR on CWV is illustrated in Figure 5c. The Prata equations predict that the effective emissivity of the atmosphere, and hence the DLR, should increase with CWV. This is broadly true for the data points, but again these cluster into separate groups for the two seasons. In this plot, the trajectory of the points is very different in the two dry seasons, as shown in Figure 5d. During the first dry season from January to April, air temperatures increased by approximately 10°C as the CWV decreased, moving the DLR onto progressively higher Prata curves but backward on the CWV axis. With the monsoon onset, temperatures remained high, so the DLR followed the highest Prata curve in the direction of increased CWV, reaching about 4 cm in May. Thereafter, despite steadily increasing CWV, the DLR dropped as temperatures fell, moving the points onto a lower Prata curve in August. Following the monsoon retreat in October, the DLR dropped suddenly as both temperature and CWV fell, moving the data points to their lowest values and following a trajectory very different from that in the first dry season.

[44] Figure 5 provides valuable insights into how the opposing influences of changes in temperature and CWV led to the observed seasonal changes in the DLR. A similar analysis is now applied to the net longwave radiation at the surface and to the OLR, providing further background for the divergences presented in section 6 and for the analysis in section 7, in which the effects on the radiative divergences are interpreted within the framework of a simple model of the greenhouse effect.

[45] Figures 6a and 6b show scatterplots of the net longwave radiation at the surface against temperature and CWV. The effect of the very different behavior of the DLR in the dry and wet seasons seen in Figure 5a is apparent in Figure 6a. In the dry seasons, the dependence of the DLR on temperature shown in Figure 5a roughly follows the Prata curves, with the result that the net radiation becomes only slightly more negative as the air temperature increases (Figure 6a). In contrast, the very different slope in the wet season seen in Figure 5a translates into a strong increase of the surface longwave cooling with temperature. The cooling is largest at the start of the wet season, when temperatures are highest but the CWV is still relatively low, and is lowest in August when temperatures have dropped but the atmosphere is at its most opaque (see also Figure 2c).

image

Figure 6. (left) Scatterplots of daily mean surface net longwave flux (W m−2), against (a) screen-level temperature (°C) and (b) column water vapor (centimeters). All data were measured at the AMF Niamey airport site. (right) Scatterplots of daily mean top of atmosphere outgoing longwave radiation (OLR), measured by GERB (W m−2), against (c) screen-level temperature (°C) and (d) column water vapor (centimeters), measured at the AMF Niamey airport site. The data points are color-coded according to whether they were measured during the dry (red) or wet (blue) seasons. Open squares denote days with cloud, according to the SEVIRI cloud mask.

Download figure to PowerPoint

[46] The difference between the dry and wet seasons is less apparent when the net longwave radiation is plotted against CWV (Figure 6b), as was also the case with the DLR. Here, the behavior is very much as expected, despite the scatter, with a steady reduction in the surface cooling as the CWV increases and the smallest values associated with cloudy days during the wet season.

[47] The final component of the longwave radiation budget is the OLR, scatterplots for which are shown in Figures 6c and 6d. The dependence on the screen-level air temperature is weak, with considerable scatter, particularly in the wet season when the atmosphere was more cloudy (Figure 6c), but there is some evidence of an increase in OLR with air temperature that is most apparent in the dry season. The isolated group of points with the highest values of OLR correspond to early April, just before the false monsoon onset, when the CWV was smallest and so the transmission of the surface emission to space was at a maximum.

[48] Figure 6d also shows considerable scatter, but it does demonstrate that OLR decreases with increasing column water vapor, as would be expected if changes in the middle and upper tropospheric humidity to which the OLR is most sensitive occur in synchronism with those at lower levels, which appears to be the case (see Figure 8b in part 1). While it would be difficult to fit a reliable trend line through these data, it is important to note that the dependence of the OLR on CWV is roughly a mirror image of that for the surface net longwave shown in Figure 6b. This implies that the rate at which the atmosphere increases its net emission to the surface as CWV increases is roughly matched by the rate at which it traps the OLR. This compensation between the dependence of the surface net longwave and of the OLR on the CWV provides the key to understanding the behavior of the atmospheric longwave divergence, as is shown later.

[49] In many other environments of the world, one would not expect to find a significant dependence of the longwave fluxes on aerosol loadings, either because the mass loadings are too low or because the aerosol is too small to have an appreciable impact on thermal radiation. However, in the Sahel, where Niamey is located, the impact can be large [e.g., Slingo et al., 2006]. To investigate this, Figure 7 shows scatterplots of the DLR at the surface and of the OLR against the aerosol extinction, defined as

  • equation image

where τ is the aerosol optical thickness at 870 nm from the Banizoumbou AERONET site. Extinction is used here because of the large range of values of optical thickness encountered through the year. Figure 7a shows that there is a measurable dependence of the DLR at the surface on aerosol, particularly in the dry season when the effective emissivity of the atmosphere is low. In the wet season, the dependence is masked by the much larger changes in the other variables. The OLR is less sensitive, which is to be expected as the aerosol is mainly found in the lowest layers of the atmosphere. For comparison, the dotted lines show seasonal averages of fluxes computed using the methodology described by Bharmal et al. (submitted manuscript, 2009). These were produced by running the radiative transfer scheme for a range of prescribed aerosol loadings, using mean thermodynamic profiles for 10-day periods, and averaging the results according to season. The simulations are in good agreement with the observations and confirm the greater dependence of the DLR on the aerosol extinction, compared with the OLR. Overall, aerosol has a measurable, but modest, impact on the longwave fluxes, except for the events with high loadings during the dry seasons, when the effects of aerosols are larger as a result of the low amounts of water vapor.

image

Figure 7. Scatterplots of (a) daily mean surface downwelling longwave flux (W m−2), measured at the AMF Niamey airport site and of (b) daily mean top of atmosphere outgoing longwave radiation (OLR), measured by GERB (W m−2), against aerosol extinction at 870 nm (defined by equation (3)). The data points are color-coded according to whether they were measured during the dry (red) or wet (blue) seasons. Points are only plotted for days without cloud, according to the SEVIRI cloud mask. The dotted lines show simulated fluxes (see text for details).

Download figure to PowerPoint

5.2. Shortwave Scatterplots

[50] The main factors that control the shortwave fluxes are more apparent in the time series than is the case for the longwave fluxes, as was shown in section 4. For example, the modulation of the fluxes by the incoming solar radiation at the TOA and the signals of clouds and major aerosol events are very clear in Figures 3 and 4. One complication is the significant seasonal variation in the surface albedo [McFarlane et al., 2009], which was highest in the dry season and lowest in the wet season as the surface became wetter and vegetation grew [Settle et al., 2008]. Another complication is scattering by aerosols and clouds, which makes it difficult to perform as simple an analysis for the individual fluxes as in the longwave. Nevertheless, a few scatterplots are used here to illustrate the effects of the CWV and aerosol loadings in the shortwave. In these plots, the daily fluxes are normalized by the value of the insolation at the TOA on that day. To first order, this removes the external astronomical control on the solar fluxes, leaving the meteorological controls to show their influence more clearly in the scatterplots. The normalized downward shortwave radiative flux at the surface thus becomes the effective solar transmittance and the normalized reflected shortwave flux at the TOA becomes the albedo.

[51] Figure 8a shows a scatterplot of the normalized downwelling shortwave flux at the surface (i.e., the transmittance) against the CWV. The distribution of points is highly skewed, with low values mainly due to cloud but also to aerosol events. However, the upper envelope of the points follows a clear trend downward with increasing CWV. This trend is due to water vapor absorption, as may be seen by comparing the points with the dotted line, which shows the broadband transmittance predicted by the modified Lacis-Hansen scheme developed by Ramaswamy and Freidenreich [1992, section 2 and Table 2]. This line takes no account of scattering, either within the atmosphere or from the surface, but it does illustrate that the trend of the observations is consistent with increasing amounts of water vapor absorption.

image

Figure 8. Scatterplots of the ratio of the daily mean surface downwelling shortwave flux (W m−2), measured at the AMF Niamey airport site, to the calculated insolation at the top of the atmosphere (W m−2), against (a) column water vapor (centimeters) and (b) aerosol extinction at 870 nm (defined by equation (3)). In Figure 8a, the dotted line shows the dependence of the water vapor transmittance on column water vapor from a simple parameterization (see text for details). In Figure 8b, the dashed lines show the transmittances given by equations (4) and (5). For clarity, these lines have been offset in the vertical to lie above most of the data points. (c and d) Corresponding scatterplots for the ratio of the daily mean top of atmosphere reflected shortwave radiation measured by GERB (W m−2), to the calculated insolation at the top of the atmosphere (W m−2). In all four panels, the data points are color-coded according to whether they were measured during the dry (red) or wet (blue) seasons. In Figures 8a and 8c, open squares denote days with cloud, according to the SEVIRI cloud mask. In Figures 8b and 8d, points are only plotted for days without cloud.

Download figure to PowerPoint

[52] The corresponding scatterplot against the aerosol extinction at 870 nm, for clear skies only, is shown in Figure 8b. The data points again follow a downward trend and two curves are also plotted to assist the interpretation. The first is based on work by McFarlane et al. [2009], who performed a linear fit to the normalized radiative effect of aerosols on the downwelling shortwave flux at the surface for noncloudy periods, as a function of the aerosol optical thickness. When expressed as the transmittance, this gives (S. MacFarlane, personal communication, 2008)

  • equation image

where τ is the optical thickness at 523 nm, which has been adjusted to the value at 870 nm for plotting on Figure 8b using an Ångstrom coefficient of 0.326 [Turner, 2008].

[53] The second curve is given by the analytical expression from the two-stream equations for the transmittance of a slab composed of conservative scatterers (i.e., the single scatter albedo is 1),

  • equation image

where f is the fraction of the radiation that is forward scattered in a single event, given by (1 − f) = (1 − g)/2, where g is the asymmetry parameter. McFarlane et al. [2009] retrieved a mean value of 0.67 for g, giving f = 0.835, which is used in equation (5).

[54] The two curves are in good agreement with each other and are consistent with the trend of the data, except that most of the data points lie below the two curves at low values of the extinction (low optical thickness), particularly during the wet season. This is probably due to the fact that such clear conditions tend to be associated with higher values of the CWV, which Figure 8a shows will tend to have lower transmittances as a result of absorption by water vapor.

[55] Finally, Figures 8c and 8d show similar plots for the albedo at the TOA (calculated by normalizing the reflected shortwave flux measured by GERB by the calculated incoming shortwave flux), as a function of CWV and aerosol extinction. Figure 8c shows a negative relationship between the albedo and the CWV, although at higher water vapor loadings the atmosphere is more frequently cloudy, which greatly increases the albedo. The magnitude of the trend is consistent with that in Figure 8a, suggesting the impact of water vapor absorption, but seasonal changes in the surface albedo may also be contributing, since the highest albedo occurs in the dry season and the lowest at the end of the wet season.

[56] Figure 8d shows no clear dependence of the TOA albedo on the aerosol extinction at low values, but there is a marked trend toward higher albedos for extinctions above about 0.5 (optical thicknesses above 0.69). The data point on the extreme right is for 8 March, during the major dust storm, when the albedo was enhanced considerably by the dust [Slingo et al., 2006]. The other points on the right all correspond to significant dust events, which may also be seen by comparing Figure 1d with the reflected shortwave flux shown in Figure 4b. As with Figure 8c, there is a separation between the data points for the two seasons, which again points to an additional influence from changes in the surface albedo. Some data points that are well above the general trend are associated with days when the cloud mask failed to identify significant amounts of cloud that had a large impact on the GERB shortwave fluxes.

[57] The effects of aerosols on both the longwave and shortwave fluxes may be summarized as being measurable, but generally smaller than the signals from atmospheric temperatures, water vapor and clouds, except at the highest aerosol loadings. More detailed analyses are provided in the companion papers in this special section.

6. Radiative Divergences

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Data Sources, Processing, and Errors
  5. 3. Time Series of Nonradiative Variables
  6. 4. Time Series of Radiative Fluxes
  7. 5. Radiative Flux Scatterplots
  8. 6. Radiative Divergences
  9. 7. Interpretation of the Longwave Fluxes and Divergences Using a Simple Model
  10. 8. Discussion and Conclusions
  11. Acknowledgments
  12. References
  13. Supporting Information

[58] Having analyzed the component fluxes and discussed the factors that control them, this section presents estimates of the longwave, shortwave and net radiative divergences, obtained by combining the ARM surface data with the GERB data from space. This realizes one of the major goals of the RADAGAST project. Divergence is defined as the net radiation into a level, so that positive values imply heating and negative values imply cooling. Surface divergence is therefore defined as the downwelling minus the upwelling radiation, corresponding to the net radiation into the surface. The TOA divergence is defined as the incoming minus outgoing radiation, i.e., the net radiation into the system. The atmospheric divergences are obtained by subtracting those at the surface from those at the TOA. The sign convention means that the longwave divergences are all negative, corresponding to cooling, while the shortwave divergences are all positive, corresponding to heating. The net divergences are obtained by adding the shortwave and longwave components. Although some of the surface and TOA divergences have already been discussed in earlier sections (e.g., the OLR), plots are also shown here, not only for completeness but also to illustrate how their behavior determines that of the atmospheric divergences.

6.1. Time Series of Divergences

[59] Figure 9a shows the daily mean longwave radiative divergences through the year. As was found earlier, the seasonal variations in the TOA and surface divergences follow a very similar pattern, with maximum cooling both from the surface and out to space at the end of the first dry season, when temperatures were highest and the CWV (and hence the atmospheric emissivity) at a minimum. In contrast, the high values of CWV in August lead to the smallest values of the surface cooling and also the lowest values of the OLR. The effects at the TOA are strengthened by the cloudiness maximum at this time of year, although clouds also had a significant impact in the dry seasons.

image

Figure 9. Time series of daily mean (a) longwave, (b) shortwave, and (c) net (longwave plus shortwave) radiative divergences (W m−2), calculated from the component fluxes, for the Earth-atmosphere system as a whole (dotted line), for the surface (dashed line) and for the atmosphere (solid line). The abscissa shows the day number in 2006, and the dashed vertical lines denote the boundaries between the calendar months, the first letters of which are indicated at the top of each plot.

Download figure to PowerPoint

[60] The most remarkable feature in Figure 9a is the atmospheric divergence, which shows a much smaller variation through the year than the divergences at the TOA and at the surface. The small magnitude of this variation shows that most of the change in cooling at the TOA was made up of changes taking place at the surface, rather than in the atmosphere. From previous work [e.g., Stephens et al., 1994], one would expect the longwave cooling of the atmosphere to increase with the CWV, which does vary over a very large range in Niamey. However, while Figure 9a shows slightly higher cooling during the wet season than in the dry season, the magnitude of the difference is quite small. The reasons for this behavior are explored below.

[61] Figure 9b shows the corresponding divergences in the shortwave. The upper envelope of these curves is clearly controlled by the seasonal variation in the daily mean insolation at the TOA (compare with Figure 4c). As in the longwave, the shortwave divergences at the TOA and at the surface follow each other through the year, while the atmospheric divergence shows a smaller variation, although in the second dry season all three divergences show the same decline. Clouds produce the largest short-term signals, but the magnitudes at the TOA and at the surface are similar, so that the effect of clouds tends to cancel in the atmospheric absorption, consistent with the expectation that clouds do not make a major contribution to atmospheric absorption in the shortwave [Dong et al., 2008]. In contrast, comparison of Figure 9b with Figure 1d shows that aerosol events impact the surface divergence more than at the TOA, so that peaks in the atmospheric heating coincide closely with peaks in aerosol optical thickness. The decline in the atmospheric divergence through the second dry season is a result of the lower CWV and the generally lower aerosol loadings. In this season, the correspondence between the peaks in the divergence and the aerosol events seen in Figure 1d is particularly striking.

[62] The net radiative divergences, obtained by summing the longwave and shortwave components, are shown in Figure 9c. The net radiation at the TOA varies between the seasons, with the region gaining energy during boreal summer and experiencing a net energy loss during boreal winter. However, the surface gains radiative energy at all times of the year, but particularly during the summer monsoon, whilst the atmosphere continually loses radiative energy to space and to the surface at a steady rate of about 75 W m−2, increasing in the second dry season.

[63] Table 1 summarizes the annual means of all the daily mean fluxes and divergences, as well as their maximum and minimum values. No attempt has been made to interpolate over the extended periods when the GERB instrument was turned off (see section 2.2.2). Over the course of the year, there is a small positive net gain of radiative energy of 14.6 W m−2 at the TOA. Together with the surface turbulent fluxes presented by Miller et al. [2009], this table provides a complete description of the annual means of the vertical flux terms at the AMF Niamey airport site during 2006.

Table 1. Annual Means, and Maximum and Minimum Daily Mean Values, of the Radiative Fluxes and Divergences and of the Screen-Level Temperature, Column Water Vapor, and Aerosol Optical Thicknessa
 MinimumMeanMaximum
  • a

    The sampling is the same for all the radiative fluxes and is determined by the days when the OLR data are available.

OLR (W m−2)170.2266.3320.9
Surface LW down (W m−2)312.7392.7447.5
Surface LW up (W m−2)441.2496.1561.1
Surface LW net (W m−2)−155.1−103.4−37.1
Atmospheric LW divergence (W m−2)−216.5−162.8−99.7
TOA SW in (W m−2)338.3407.3444.2
TOA SW out (W m−2)85.8125.8269.6
TOA SW net (W m−2)170.6281.1342.5
Surface SW down (W m−2)59.3250.1351.2
Surface SW up (W m−2)9.357.584.6
Surface SW net (W m−2)50.0192.7267.7
Atmospheric SW divergence (W m−2)45.091.0201.9
TOA net divergence (W m−2)−62.514.6101.8
Surface net divergence (W m−2)7.089.2180.3
Atmospheric net divergence (W m−2)−124.7−75.09.1
Air temperature (°C)20.829.836.5
Column water vapor (cm)0.492.895.70
Aerosol optical thickness0.030.473.49

6.2. Scatterplots

[64] It was shown in Section 5 that the effects of temperature and CWV have opposing influences on the DLR and net longwave radiation at the surface, and also on the OLR. The dependence on temperature was quite complicated, with different behaviors in the dry and wet seasons. The dependence on CWV was broadly as would be expected: a reduction in both the surface and TOA cooling with increasing CWV. Consistent with these results, Figure 10a appears to show no discernible dependence of the atmospheric longwave divergence on temperature, but there is a weak dependence on the CWV (Figure 10b). Unfortunately, given the strong negative correlation between temperature and CWV, it is very difficult to interpret these scatterplots further, because they do not allow the two effects to be separated. A methodology that allows such a separation and which explains the behavior shown in Figure 9a is introduced in section 7.

image

Figure 10. Scatterplots of the longwave radiative divergence across the atmosphere (W m−2) against (a) screen-level temperature (°C), (b) column water vapor (centimeters), and (c) aerosol extinction at 870 nm (defined by equation (3)). In each panel, the data points are color-coded according to whether they were measured during the dry (red) or wet (blue) seasons. In Figures 10a and 10b, open squares denote days with cloud, according to the SEVIRI cloud mask. In Figure 10c, points are only plotted for days without cloud.

Download figure to PowerPoint

[65] While scatterplots cannot unravel the dependence of the longwave divergence on temperature and CWV, they are valuable for other variables. The dependence of the atmospheric longwave divergence on the aerosol extinction is shown in Figure 10c. This shows a discernible trend toward greater longwave cooling as the aerosol loading increases, consistent with the results in Figure 7, which show that the downward longwave at the surface is more sensitive to aerosol than the OLR. This dependence is somewhat stronger in the dry seasons.

[66] The atmospheric divergence in the shortwave clearly increases with CWV, particularly at the lowest humidities (Figure 11a). The shortwave absorption also increases with the aerosol loading (Figure 11b). This is mainly due to the reduced transmission to the surface (Figure 8b), because the TOA albedo actually shows an increase with the aerosol loading (Figure 8d), a characteristic feature of weakly absorbing aerosols.

image

Figure 11. Scatterplots of the shortwave radiative divergence across the atmosphere (W m−2) against (a) column water vapor (centimeters) and (b) aerosol extinction at 870 nm (defined by equation (3)). In both panels, the data points are color-coded according to whether they were measured during the dry (red) or wet (blue) seasons. In Figure 11a, open squares denote days with cloud, according to the SEVIRI cloud mask. In Figure 11b, points are only plotted for days without cloud.

Download figure to PowerPoint

7. Interpretation of the Longwave Fluxes and Divergences Using a Simple Model

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Data Sources, Processing, and Errors
  5. 3. Time Series of Nonradiative Variables
  6. 4. Time Series of Radiative Fluxes
  7. 5. Radiative Flux Scatterplots
  8. 6. Radiative Divergences
  9. 7. Interpretation of the Longwave Fluxes and Divergences Using a Simple Model
  10. 8. Discussion and Conclusions
  11. Acknowledgments
  12. References
  13. Supporting Information

[67] One of the most remarkable results from this study is the compensation between the effects of changes in temperature and column water vapor on the longwave fluxes and atmospheric divergences. These changes are negatively correlated for much of the year, so the scatterplots do not allow the effects to be separated unequivocally from each other. In this section, we examine this result further and provide such a separation with the help of a simple model of the greenhouse effect. This employs a version of the methodology of Stephens et al. [1994].

[68] Let the longwave atmospheric flux divergence be denoted by ΔF. Stephens et al. [1994] write this in terms of the component fluxes as

  • equation image

where the first term is the upward longwave flux from the surface, the second term is the outgoing longwave radiation (OLR) and the third term is the DLR at the surface.

[69] The problem with (6) in this form is that the second and third terms depend both on temperature and on the atmospheric emissivity (which for clear skies is largely controlled by the water vapor profile). To deal with this, Stephens et al. [1994] introduced two dimensionless parameters that normalize these two fluxes. Their choice of normalization was motivated by the desire to use observed quantities (in their case σTs4 and OLR) to estimate the surface fluxes and atmospheric divergence from readily available global sea surface temperature and OLR data sets. Here, since the main purpose is to separate the effects of temperature and column water vapor, we choose different normalizations that simplify the subsequent interpretation.

[70] The first dimensionless parameter is the normalized greenhouse effect, G, defined here as

  • equation image

This is always smaller than unity in Niamey, reflecting the fact that the greenhouse effect acts to reduce the emission to space, compared with that from the surface.

[71] Similarly, the second dimensionless parameter is defined as

  • equation image

This is also smaller than unity, because the downward longwave radiation from the atmosphere is always lower than the emission from the surface in Niamey. Comparison with equation (1) shows that F is similar to ɛ in the Prata formula, the difference being that here the surface temperature is used (this does not affect the following argument).

[72] Substituting (7) and (8) into (6) leads to

  • equation image

where the value of ΔF is generally negative, because on average the net effect of longwave radiation is to cool the atmosphere.

[73] To first order, equation (9) separates the effects of temperature (the first term) from those of atmospheric emissivity (the term in parentheses), although this is only approximate as in reality both F and Fg depend on the full vertical profiles of both temperature and humidity.

[74] The study by Stephens et al. [1994] concentrated on clear-sky data over the oceans. They showed that the magnitudes of the greenhouse effect and of the clear-sky atmospheric longwave cooling increased with both sea surface temperature and column water vapor. This behavior is largely controlled by the Clausius-Clapeyron equation: over the oceans, the column water vapor (and hence the effective emissivity of the atmosphere) is lowest at cold, high latitudes and highest at warm, low latitudes. As a result, column water vapor and temperature are positively correlated when global data are considered, as shown by Stephens [1990]. In contrast, over Niamey the column water vapor is controlled much less by the air temperature than by the atmospheric dynamics, so that the lowest values occur when the hot, dry Harmattan wind blows from the deserts, while the highest values occur during the wet season, when the air blows from the Gulf of Guinea. The net effect is that temperature and column water vapor are negatively correlated for most of the year, leading to the unusual impact on the longwave fluxes and heating rates found in this study.

[75] To illustrate the impact on the longwave atmospheric divergence, scatterplots of G, F and 1 − GF against CWV are shown in Figure 12. Although there is some scatter, Figure 12 demonstrates that the dependence on the CWV of the normalized greenhouse effect and of the atmospheric emissivity seen from the surface are virtually mirror images of each other, so that the 1 − GF term in equation (9) shows only a weak dependence on CWV. As the CWV increases, the atmosphere thus loses longwave radiation to the surface with an increasing efficiency similar to that with which it traps the OLR. The time series of these terms shown in Figure 13 illustrates the partial compensation between G and F through the year and the smaller changes in 1 − GF. The relatively small variation in the atmospheric longwave divergence through the year is thus due to the partial cancellation between the normalized greenhouse effect G and the atmospheric emissivity seen from the surface F, which reduces the dependence of the atmospheric cooling on the CWV, so that the remaining variations through the year of the divergence come from the relatively smaller changes in the σTs4 and 1 − GF terms in equation (9).

image

Figure 12. Scatterplots of the dimensionless parameters (a) G, (b) F, and (c) 1 − GF (see text for definition) against the column water vapor (centimeters). The data points are color-coded according to whether they were measured during the dry (red) or wet (blue) seasons. Open squares denote days with cloud, according to the SEVIRI cloud mask.

Download figure to PowerPoint

image

Figure 13. Time series of the dimensionless parameters (a) G, (b) F, and (c) 1 − GF (see text for definition). The abscissa shows the day number in 2006, and the dashed vertical lines denote the boundaries between the calendar months, the first letters of which are indicated at the top of each plot.

Download figure to PowerPoint

8. Discussion and Conclusions

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Data Sources, Processing, and Errors
  5. 3. Time Series of Nonradiative Variables
  6. 4. Time Series of Radiative Fluxes
  7. 5. Radiative Flux Scatterplots
  8. 6. Radiative Divergences
  9. 7. Interpretation of the Longwave Fluxes and Divergences Using a Simple Model
  10. 8. Discussion and Conclusions
  11. Acknowledgments
  12. References
  13. Supporting Information

[76] This paper has presented the broadband shortwave and longwave radiative fluxes measured from space and from the AMF site at Niamey airport during the RADAGAST experiment in 2006. The fluxes at the TOA and at the surface were combined to provide estimates of the divergence of radiation across the atmosphere. The radiation data were interpreted with the help of the surface air temperature and CWV data presented in part 1, together with additional information on the cloud cover and aerosol optical depths from other data sets. These data were used to investigate the factors that control the radiative fluxes and divergences. The shortwave fluxes are controlled by the annual cycle of the insolation at the top of the atmosphere, by clouds, aerosols and water vapor and by seasonal changes in the surface albedo. While clouds strongly influence the fluxes, their effect on the atmospheric shortwave divergence is much smaller. The longwave fluxes are controlled by the surface temperature, atmospheric temperatures, humidities and clouds and also by aerosols, which is a result of the unusually high loadings of dust aerosols present throughout the year. A remarkable compensation was found between the influence of the air temperature and CWV on the longwave fluxes, which was interpreted with the aid of the simple model of Prata [1996]. In addition, the atmospheric longwave divergence was found to be surprisingly constant through the year. A simple model of the greenhouse effect, based on the work of Stephens et al. [1994], provided helpful insights that suggested that this result is a consequence of a partial cancellation between the dependence of the greenhouse effect seen from space and from the surface on the CWV.

[77] There are several issues arising from this work that merit further investigation. The lack of sensitivity of the atmospheric longwave divergence to the large changes in CWV that take place through the year appears to be related to changes in humidity that are correlated through the whole depth of the troposphere, particularly during the wet season, as suggested by Figure 8b in part 1. In the present paper, CWV is used as a surrogate for the changes in middle and upper tropospheric humidities, but more work is needed to quantify this relationship. A more detailed analysis of the sonde humidities is certainly required, coupled with controlled numerical experiments with radiation codes.

[78] Clearly, it is also important to establish whether these results are unique to 2006 and how representative they are of the wider area beyond Niamey. Unfortunately, the paucity of surface radiative flux data in this region outside of the period when the AMF was deployed in Niamey means that the particular approach taken here would not be possible. However, it would be interesting to see whether numerical weather prediction (NWP) analyses, or reanalyses, might be used to investigate some of these issues. For example, the analyses could be evaluated against the RADAGAST data for 2006 to establish their strengths and weaknesses and then used to investigate other years and the applicability of the results to a wider geographical region. One limitation is that most NWP analyses do not explicitly include aerosols, some representation of which is clearly necessary. The analyses could also provide estimates of the atmospheric heat transport that must balance the nonzero net radiation at the top of the atmosphere shown in Table 1. It would also be interesting to determine whether climate models can reproduce the radiative climatology found here, as well as the relationships between the radiative and other variables.

[79] No attempt has been made here to derive information on the vertical profiles of the radiative divergences, but in principle techniques that use data from instruments at the permanent ARM sites [e.g., McFarlane et al., 2007] could be applied to the RADAGAST data. This would also complement the investigation of middle and upper tropospheric humidities mentioned above.

[80] Further analysis of the small impact of clouds on the atmospheric shortwave divergence found here would also be valuable, in view of the continuing uncertainty as to the existence of excess or anomalous cloud absorption [Ramana et al., 2007]. It would certainly be interesting to see whether the RADAGAST data could be used to establish limits on the magnitude of any absorption that could not be reproduced by contemporary radiation codes, although the significant relative error found in section 2.2.4 would clearly be a restriction.

Acknowledgments

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Data Sources, Processing, and Errors
  5. 3. Time Series of Nonradiative Variables
  6. 4. Time Series of Radiative Fluxes
  7. 5. Radiative Flux Scatterplots
  8. 6. Radiative Divergences
  9. 7. Interpretation of the Longwave Fluxes and Divergences Using a Simple Model
  10. 8. Discussion and Conclusions
  11. Acknowledgments
  12. References
  13. Supporting Information

[81] This paper is dedicated to the memory of Anthony Slingo, without whose inspiration the RADAGAST deployment and this subsequent work would not have been possible. We thank Didier Tanré for his efforts in providing the Banizoumbou AERONET site. The AMF data were obtained by the Atmospheric Radiation Measurement (ARM) Program, which is funded by the Office of Biological and Environmental Research, Office of Science, U.S. Department of Energy. The Edition 1 GERB data were obtained from the GERB Ground Segment Processing System (GGSPS) at the Rutherford Appleton Laboratory, UK. We thank Richard Allan and David Turner for their comments on the manuscript. The analysis in section 5.1 builds upon the Ph.D. at ESSC of Peter Henderson, awarded in 2006. Helen White is funded by a UK Natural Environment Research Council studentship, and Nazim Ali Bharmal is funded by NERC grant NE/D002370/1. During the preparation of this work, Anthony Slingo and Gary Robinson were supported by the NERC-funded National Centre for Earth Observation.

References

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Data Sources, Processing, and Errors
  5. 3. Time Series of Nonradiative Variables
  6. 4. Time Series of Radiative Fluxes
  7. 5. Radiative Flux Scatterplots
  8. 6. Radiative Divergences
  9. 7. Interpretation of the Longwave Fluxes and Divergences Using a Simple Model
  10. 8. Discussion and Conclusions
  11. Acknowledgments
  12. References
  13. Supporting Information

Supporting Information

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Data Sources, Processing, and Errors
  5. 3. Time Series of Nonradiative Variables
  6. 4. Time Series of Radiative Fluxes
  7. 5. Radiative Flux Scatterplots
  8. 6. Radiative Divergences
  9. 7. Interpretation of the Longwave Fluxes and Divergences Using a Simple Model
  10. 8. Discussion and Conclusions
  11. Acknowledgments
  12. References
  13. Supporting Information
FilenameFormatSizeDescription
jgrd14987-sup-0001-t01.txtplain text document1KTab-delimited Table 1.

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.