Geophysical Research Letters

Influence of aerosol climatology on forecasts of the African Easterly Jet



[1] The European Centre for Medium-Range Forecasts (ECMWF) recently implemented a revised annually varying aerosol climatology in their forecasting system. The new climatology greatly reduces the aerosol optical depth over the Sahara compared to the previous time-invariant climatology. Using high resolution dropsonde data in addition to the model analyses, it is demonstrated that the direct radiative effect resulting from the aerosol climatology modification significantly improves the ECMWF 5-day forecasts of the African Easterly Jet, the central dynamical feature in this region.

1. Introduction

[2] Aerosols directly affect the atmospheric radiation budget, and their inclusion in climate models significantly modifies estimates of future climate (see the review by Haywood and Boucher [2000]). Due to their relatively short residence time in the atmosphere, aerosol concentrations tend to be very heterogeneous, both spatially and temporally, and therefore their direct radiative effect can manifest itself in significant modifications of the atmospheric general circulation. Aerosols also have an indirect effect on the atmosphere through their role in cloud microphysics [e.g., Twomey, 1974] but this will not be addressed in this article.

[3] Here the influence of the aerosol direct radiative forcing on the representation of the main dynamical features of the African Sahel region is examined, using the ECMWF integrated forecast system (IFS) model. While this model faithfully reproduces many aspects of the global circulation, both Kamga et al. [2000] and Thorncroft et al. [2003] show inadequacies in the prediction of the 700 hPa African Easterly Jet (AEJ) at the 5-day range when compared to the analyses in the former study and to high resolution dropsonde data in the latter. As the forecast progresses, the jet generally weakens substantially, especially to the east, and is usually too zonal in structure. This is despite the fact that the analysis replicates the observed AEJ with reasonable fidelity [Thorncroft et al., 2003]. The accurate prediction of the jet structure is important, since it is a central component of the African monsoon system in the Sahel region and plays an important role in the initiation of both African Easterly waves and mesoscale convective complexes [Houze and Betts, 1981] as well as tropical cyclogenesis [Karyampudi and Pierce, 2002].

[4] Over the last decade satellite observations and improved modelling studies have led to a more accurate picture concerning aerosol distributions. This includes the high concentrations of mineral aerosols that are observed over the African Sahara, which can be advected far out into the Atlantic, often reaching the Americas [e.g., Perry et al., 1997; Prospero and Lamb, 2003]. In response, the ECMWF model has recently incorporated a new aerosol climatology, significantly altering the aerosol optical depth in this region. The consequences for the forecast of the AEJ are presented here.

[5] High resolution dropsonde data from the JET2000 campaign are used for validation purposes. These are supplemented by three months of analyses (the assessment of the atmospheric state used for forecast initialization), which are considered a reasonable proxy for the ‘truth’ since the relatively sparse observations in this region are still able to effectively constrain the analysis system, relative to the much larger medium range forecast errors [Tompkins et al., 2004] (available at

2. Aerosol Climatologies

[6] The two aerosol climatologies used in this study correspond to those used operationally in the ECMWF forecast system before and after the 7th October, 2003. The old climatology was originally designed by Tanre et al. [1984]. It provides annual mean geographical distributions for aerosol types of maritime, continental, urban and desert aerosol, in addition to a uniformly distributed tropospheric and stratospheric ‘background’ aerosol loading. Radiative properties (extinction coefficient, single scattering albedo, asymmetry factor) are consistently derived for each aerosol type and the various spectral intervals of the ECMWF radiation schemes following Tanre et al. [1984].

[7] More recently, chemical transport models have addressed the life cycles of various aerosol types. A climatology for the annual cycle of the distribution of various aerosol types has been compiled by Tegen et al. [1997]. This has been implemented in the ECMWF forecast system, with the annual cycle described by monthly mean aerosol optical depth distributions (aerosol data files are available at The radiative properties are derived following Hess et al. [1998]. Table 1 compares the maximum optical thicknesses in the old and new climatologies. In particular, the old climatology was dominated by desert aerosols, with a spatial maximum optical thickness of 1.9 over the Saharan region. This annual mean figure is replaced by a spatially moving peak value of between 0.18 (December) and 1.01 (July) in the dust-like aerosol category of the new climatology, with the peak again over the Sahara. This reduction is also confirmed by more recent in situ studies [Haywood et al., 2003]. Note that the vertical distributions for all aerosol components are similar in the two climatologies, and that the new climatology has no requirement for a tropospheric background climatology, while the stratospheric aerosol loading remains unchanged. Further details are in the work of Tompkins et al. [2005].

Table 1. Maximum Optical Thickness in the Previous and Revised Aerosol Climatologiesa
  • a

    Previous optical thicknesses are from Tanre et al. [1984], and revised optical thicknesses are from Tegen et al. [1997]. The Tanre climate is divided into geographical regions. The right two columns give the January and July monthly mean value for the specific aerosol categories into which the Tegen climatology is divided.

Urban0.1black carbon0.0390.039
Background Aerosol Distribution

[8] The large diminution of the total aerosol optical thickness, mainly linked to that of the dust-like aerosols, is expected to increase the available solar radiation at the surface. The concurrent decrease in downward longwave radiation at the surface is much smaller (<10% of the shortwave signal). In the ECMWF model, the aerosol concentration does not impact the cloud microphysics, and thus any change in the dynamical circulation must arise from the direct radiative effect.

3. Results

3.1. Comparison to JET2000

[9] Two north-south sorties were conducted by the aircraft to measure transects through the AEJ on the 28th and 29th August at an approximately longitude of 2.3E. The latter date was shown by Tompkins et al. [2004] to be strongly affected by nearby convection. Due to the lack of predictability of such an event only the 28th August flight is used. The dropsondes from the longest continuous transect are used for comparison to a pair of 5 day forecasts conducted with the T511 resolution model (approximately equivalent to a 39 km horizontal resolution at the equator). For each sonde, the nearest model grid point to each drop location is identified, and all gridpoints in a transect of ±5 degrees in the cross-flight direction are averaged (as Tompkins et al. [2004]). This averaging, the implicit near-grid diffusion of the model and the lower vertical resolution, all lead to smoother model fields relative to the data.

[10] Figure 1a shows the mean bias of the zonal wind for the old and new aerosol climatologies, averaged across the transect from 8°N to 19°N. For this date the new aerosol climatology appears to have a beneficial impact. In agreement with previous analysis the model suffers from positive zonal wind biases implying the jet is too weak. The new aerosol climatology produces a net acceleration of the flow throughout the lower and mid troposphere in the 5-day forecast, with a mean acceleration across the transect of around 1 m s−1. Root mean square errors are also reduced (not shown), dropping at the jet level of 700 hPa from approximately 6 to 4 m s−1.

Figure 1.

(a) Zonal wind mean bias of 5-day forecast verifying at 12 UTC on the 28th August 2000 using the old and new aerosol climatologies compared to the dropsonde data. Bias is calculated across the transects shown in panels (b), (c) and (d) which show the zonal wind from the dropsondes (locations marked with white triangles at panel bottom, UTC times marked at top), and the forecasts with old and new aerosol climatologies, respectively. For clarity, the data in panel (b) is vertically smoothed with a 40 hPa box car window.

[11] Examination of the model and dropsonde transects (Figures 1b–1d) reveals that the structure of the jet in the forecast has also slightly improved. In addition to increasing the core velocities in excess of 2 m s−1, the greatest acceleration (>3 m s−1) is at the southern end of the transect, where the errors were greatest. However, neither forecast captures the north-south tilt of the zonal wind, where the peak winds occur at lower altitudes to the south, also seen in the NCEP July climatology of Cook [1999]. It is the failure to capture this structure that leads to the peak bias occurring in the strong shear zone at 800 hPa. The structure of the jet is important since it determines the strength and location of the strong wind shear zones crucial for initiation of mesoscale organized convective systems, and thus the under-prediction of low-level shear is critical.

3.2. Seasonal Impact

[12] The JET2000 aircraft observations appear to show a positive impact of the new aerosol climatology on the 5-day model forecast of zonal winds in the region of the AEJ. However, it is inadvisable to generalize conclusions drawn from a single forecast. To see if the trends observed for this single date hold more generally, four months of 5 day forecasts from June to September (the peak AEJ season) in 2003 using two versions of the ECMWF model are examined, again at T511 resolution. The two model versions used are referred to as ‘cycles’ 26r1 and 26r3 (the former was operational at this time). In terms of the physical parameterizations used in the nonlinear forecast model, the principal difference between these cycles was the introduction of the new aerosol climatology in the latter.

[13] The four month average of the 5-day forecast zonal wind (Figure 2) shows that the new aerosol climatology improves many attributes of the AEJ. There is a net acceleration of the zonal flow at the southern flank of the jet axis by a 2.4 m s−1, which is maximized between −5°W and 10°E (Figure 2d). This acceleration extends the region of maximum jet winds much further eastwards (Figure 2c), bringing the forecast into much closer agreement with the 26r3 (and 26r1) analyses (in Figure 2a). It is also clear that the 4 month average forecast reproduces the position of the jet, with the jet further south towards the east. Previous model versions were similar to cycle 26r1, with a zonal jet structure [Cook, 1999]. The improvements are not restricted to the jet; the over strong 700 hPa westerlies at 5°N are also reduced. Overall, the five day forecast of zonal wind is significantly improved compared to the analysis (compare panels e and f), and all the changes are consistent with the comparison to the high resolution dropsonde data. August mean zonal cross sections through the jet confirm this, with a well formed jet only occurring at the correct altitude and latitude when the new aerosol climatology is used (not shown) [see Tompkins et al., 2005].

Figure 2.

Mean 700 hPa zonal wind from 4 months (June–September 2003) of (a) Daily 12Z Analyses (b) 5-day 26r1 forecast (FC) using old aerosol climatology and (c) 5-day 26r3 FCs using new climatology, (d) FC difference; panel c - panel b, (e) 26r1 FC error; panel b - panel a, (f) 26r3 FC error; panel c - panel a.

[14] As stated above, the new aerosol climatology only impacts the forecast via its direct effect on the radiative forcing of the atmosphere. Figure 3 quantifies this by contrasting the 5-day mean total columnar solar absorption, averaged for the 4 months. The impact is significant, with total absorption differences exceeding 45 W m−2. The atmospheric cooling induced by the introduction of the new aerosol climatology (which has significantly less solar absorption) is concentrated in the planetary boundary layer (PBL) and lower troposphere, thus stabilizing the atmosphere. This is partially offset by increases in surface sensible heat flux ranging from 15 to 20 W m−2 across the region (not shown).

Figure 3.

Difference (new minus old aerosol climatology) in total columnar solar absorption for 4 months of 5 day forecasts, showing reductions of up to 50 W m−2 using the new aerosol climatology.

[15] The increase in stability increases atmospheric subsidence and suppresses deep convection. Miller and Tegen [1999] already emphasized that the vertical extent and magnitude of the aerosol forcing can influence the circulation through its feedback with convection. Deep convection can self-aggregate, for example through positive feedbacks with the land surface or coldpool activity [Taylor and Lebel, 1998; Simpson, 1980]. The largest change in subsidence occurs in the regions that undergo deep convection in the control model, indicating that such a feedback is operating. The Intertropical Convergence Zone (ITCZ) thus migrates to the south. One consequence is that the low level south-westerly monsoon flow is less strong and low level moisture is advected less far north. This is beneficial since comparison to low altitude aircraft data by Thorncroft et al. [2003] indicated that the monsoon flow was too strong in the ECMWF model. Reductions of up to 5K mean equivalent potential temperature (θe) at the lowest model level occur in association with the southerly displacement of the ITCZ (Figure 4).

Figure 4.

As Figure 3 but for θe at the lowest model level at 10 m.

[16] The over-strong monsoon flow and associated northward migration of the ITCZ in the mid-range forecast resulted in the weak geostrophic jet winds in earlier model versions [see Tompkins et al., 2004, Figure 1], which depend on the contrast between the dry and moist convective regions to the north and south, respectively [Thorncroft and Blackburn, 1999]. The more accurate radiative forcing associated with the improved aerosol climatology prevents the northward ITCZ migration and thus allows the AEJ to be sustained in the forecast.

[17] Further analysis of the tropospheric deep temperature and moisture profiles (not shown) reveals that the most significant differences are restricted to the PBL monsoon flow described above, although above 800 hPa there is a reduction in relative humidity associated with an increase in the northerly meridional wind component. This, combined with a reduction in stability at 20°N below 600 hPa, indicates that the increase in surface sensible heat fluxes strengthens the heat low circulation with the new aerosol climatology. The consequential mid-level drying could further regulate the deep convection to the south. See Tompkins et al. [2005] for additional discussion.

4. Conclusions

[18] Aerosols can alter the atmospheric circulation through their direct radiative forcing. Here, the impact of an improved aerosol climatology on 5-day forecasts of the African Easterly jet is examined. A pair of 5-day forecasts were compared to high resolution dropsonde data from the JET2000 campaign. The first forecast used the annually fixed aerosol climatology of Tanre et al. [1984]. The second forecast instead used an updated climatology described by Tegen et al. [1997] with reduced aerosol loading over Africa. The new aerosol climatology significantly improves some aspects of the jet structure and strength. In addition, 4 months of 5-day forecasts were compared using the contrasting aerosol distributions. This demonstrated a clear improvement with the new climatology, with the jet strengthened, elongated to the east, and less zonal, in agreement with the analyses.

[19] It is proposed that the modification suppresses deep convection by stabilizing the atmosphere, preventing the ITCZ from progressively migrating north during the forecast. A strong reduction in PBL θe is noted, and thus a feedback between deep convection and low level moist advection is also likely to play a role. Future work will investigate the wider implications for this change on the large-scale circulation through teleconnections.