Impacts of convection schemes on simulating tropical-temperate troughs over southern Africa
This study examines southern African summer rainfall and tropical temperate troughs (TTTs) simulated with three versions of an atmospheric general circulation model differing only in the convection scheme. All three versions provide realistic simulations of key aspects of the summer (November–February) rainfall, such as the spatial distribution of total rainfall and the percentage of rainfall associated with TTTs. However, one version has a large bias in the onset of the rainy season. Results from self-organizing map (SOM) analysis on simulated daily precipitation data reveals that this is because the occurrence of TTTs is underestimated in November. This model bias is not related to westerly wind shear that provides favorable conditions for the development of TTTs. Rather, it is related to excessive upper level convergence and associated subsidence over southern Africa. Furthermore, the model versions are shown to be successful in capturing the observed drier (wetter) conditions over the southern African region during El Niño (La Niña) years. The SOM analysis reveals that nodes associated with TTTs in the southern (northern) part of the domain are observed less (more) often during El Niño years, while nodes associated with TTTs occur more frequently during La Niña years. Also, nodes associated with dry conditions over southern Africa are more (less) frequently observed during El Niño (La Niña) years. The models tend to perform better for La Niña events, because they are more successful in representing the observed frequency of different synoptic patterns.
KeywordsTropical-temperate trough El Niño-Southern Oscillation Southern Africa Convection scheme Atmospheric general circulation model
Tropical-temperate troughs (TTTs) provide a substantial portion of summer rainfall over southern Africa (Africa south of 12.5°S). During a TTT event, tropical convection is linked with a transient system in the mid-latitudes (e.g. Vigaud et al. 2012), and a band of cloud and rain extending from the northwest to the southeast is formed (Harrison 1984; Todd and Washington 1999; Washington and Todd 1999; Todd et al. 2004; Ratna et al. 2012). The positioning of the Angola Low or related troughs over the northwestern part of the subcontinent plays an important role in the formation of TTTs over southern Africa (Lyon and Mason 2007, 2009; Vigaud et al. 2008).
The interannual variation of rainfall in this region has been shown to be influenced by El Niño-Southern Oscillation (ENSO) (e.g. Lindesay and Vogel 1990; Richard et al. 2000; Cook 2000, 2001; Manhique et al. 2011) and sea surface temperature (SST) anomalies in the surrounding oceans (e.g. Mason 1995; Rouault et al. 2003; Washington and Preston 2006; Williams et al. 2008; Vigaud et al. 2012), including those associated with the subtropical dipole modes (Reason 2002; Fauchereau et al. 2009; Morioka et al. 2010, 2011, 2012). To mitigate impacts of the above-mentioned interannual variations, skillful predictions are required (Behera and Yamagata 2001; Reason et al. 2006; Landman et al. 2009). Landman and Beraki (2012) conducted retroactive multi-model forecasts over southern Africa, and found that their forecasts had relatively good skill during El Niño and La Niña years, but performed poorly during neutral years (years without either El Niño or La Niña events). Also, Yuan et al. (2013) showed that a coupled general circulation model (CGCM) with high skills in predicting ENSO and the subtropical dipole modes had relatively high skills in predicting southern African precipitation anomalies in a broad region south of 10°S. Although these studies have illustrated some useful skill in forecasting summer rainfall over southern Africa, the simulation and prediction of rainfall over this region still faces numerous deficiencies. For example, Kataoka et al. (2012) showed that almost all CGCMs that participated in the third phase of the Coupled Model Intercomparison Project (CMIP3; Meehl et al. 2007) failed to simulate the relationship between the precipitation anomaly over southern Africa and global SST anomalies. Also, Lyon and Mason (2009) showed that both atmospheric general circulation models (AGCMs) forced by the observed SST and hindcast seasonal forecasts from CGCMs were unable to reproduce atmospheric circulation anomalies over southern Africa during the strong El Niño event of January–March 1998.
Realistic simulations of summer rainfall are important to obtain plausible projections of future climate change over southern Africa, which may in turn be helpful for adaptation (e.g. Thomas et al. 2007). The projection of Engelbrecht et al. (2009) suggested a general decrease in rainfall over southern Africa, but with more frequent occurrence of TTTs over the southeastern part of the subcontinent during mid-summer. The latter resulted from the intensification of the Mascarene High over the southwestern Indian Ocean under global warming. On the other hand, Shongwe et al. (2009) have shown that in the CMIP3 models, the rainfall onset over southern Africa is delayed under global warming, owing to a significant reduction in moisture supply from the southwestern Indian Ocean. Also, Lyon (2009) showed a future drying trend in austral summer rainfall, although this was found to be a model-dependent result, and the experiments of Tadross et al. (2005) indicated that choice of cumulus convection scheme may be regarded as an important source of uncertainty in regional projections of future rainfall over southern Africa. The identification of model biases associated with a particular convection scheme, and the eventual improvement or optimal selection of schemes, may contribute to a reduction in uncertainties associated with the projection of future climate change over this region.
Realistic modeling of the basic climatic state is the first step towards the realistic simulation of interannual variations, accurate seasonal prediction, and more reliable projections of future climate change. However, realistic simulations of the southern African rainfall climatology remain a big challenge, partly because of the interaction of tropical and extra-tropical processes over this region. In this regard, van den Heever et al. (1997) used a regional atmosphere model and successfully simulated many aspects of two TTT events. More recently, several studies have attempted to improve the simulation of the rainfall over southern Africa (Crétat et al. 2012; Ratnam et al. 2012). Crétat et al. (2012) conducted 27 sensitivity experiments using three different kinds of parameterizations for cumulus convection, planetary boundary layer, and microphysics in a regional atmospheric model. Ratnam et al. (2012) compared results from the same regional model, which was forced by observed SSTs or coupled with an ocean mixed-layer model. However, these regional models depend heavily on the lateral boundary conditions provided by global models or reanalysis data, making it somewhat difficult to determine the relative contribution of convection schemes in causing model biases. Therefore, we here analyze three versions of the same AGCM differing only in the convection scheme, in light of obtaining more realistic simulations of precipitation over the southern African region. Such an approach was useful for understanding of precipitation in other regions such as in India (e.g. Singh et al. 2011; Sinha et al. 2012).
This paper is organized as follows. A brief description of the model, convection schemes, data, and methodology is given in the next section. In Sect. 3, we compare seasonal variations in precipitation over the southern African region and TTTs simulated by three versions of our AGCM, and discuss possible causes of model biases. We further evaluate model performances in simulating interannual variations, with a special focus on the relation with ENSO, in Sect. 4. Summary and discussions are provided in Sect. 5.
2 Model, data, and methodology
2.1 Model and data
The AGCM used in this study is the Frontier Atmospheric General Circulation Model (FrAM; Guan et al. 2000). Influences of climate variability related to Indian Ocean Dipole and ENSO on regional climate is relatively well captured by the FrAM (Chakraborty et al. 2005; Yuan et al. 2012). It is the atmospheric component of the University of Tokyo Coupled general circulation model (Tozuka et al. 2006, 2011; Doi et al. 2010). The model equations are solved on 28 hybrid levels in the vertical, from the surface up to 10 hPa level, by using the spectral transform method with triangular truncation at wavenumber 42 (T42). The longwave radiation scheme is based on the multiple parameter random model of Shibata and Aoki (1989) and Shibata (1989). In this scheme, H2O, CO2, and O3 are considered as absorbers of the longwave radiation and the cloud emissivity is estimated as a function of temperature. The shortwave radiation scheme is based on Lacis and Hansen (1974), except for the calculation for partially cloudy skies. Here, H2O and O3 are considered as absorbers of the shortwave radiation. The cloud fraction is assumed to be a function of relative humidity and calculated following Slingo and Slingo (1991). The assumption of random overlapping is used for both longwave and shortwave radiation. For the land surface model, we used that of Viterbo and Beljaars (1995). The surface eddy fluxes of momentum, heat, and moisture are calculated using bulk formula (Louis et al. 1982), and the effect of subgrid-scale orography induced by the gravity wave drag is parameterized after Palmer et al. (1986).
For the parameterization of cumulus convection, schemes developed by Kuo (1974), Emanuel (1991), and Tiedtke (1989) are used in this study (see Stensrud 2007 for a review). Briefly, Kuo (1974) formulated a parameterization in which convective precipitation is proportional to total column moisture convergence and it is regarded as a deep-layer control scheme. The parameterization proposed by Tiedtke (1989) is a mass flux scheme with updraft plume, downdraft plume, and environmental subsidence. Entrainment of the updraft plumes is assumed to be proportional to the large-scale moisture convergence, while downdraft plumes are assumed to start at the level of free sink and proportional to the upward mass flux. The precipitation rate is equal to condensed liquid water in the above plume model. Emanuel (1991) developed a parameterization categorized as a mass flux scheme that takes into account the collective effects of the various subparcels in the cloud. A specified fraction of condensed water from the subparcels falls as precipitation. We call these three experiments FrAM_Kuo, FrAM_Emanuel, and FrAM_Tiedtke, respectively. We note that we do not intend to discuss superiority of a particular scheme in this study. Rather, the three experiments should be considered as sensitivity experiments of a single AGCM. Also, their performance depends on the resolution of the model, and the parameterization of Kuo (1974) tends to perform better with larger grid size (Singh et al. 2011).
This model is integrated from 1981 to 2008 using monthly SST and sea ice cover data from Hurrell et al. (2008). This dataset has been used in the Atmospheric Model Intercomparison Project (AMIP) simulations. For each experiment, five different initial conditions are used to generate five ensemble members, and outputs after 1982 are used for the present analysis. To generate initial conditions, we have spun up the model from a calm and isothermal atmosphere for about 3 years (the spin-up time varies slightly for the five different ensemble members, being 3 years for one member with the others 5, 10, 15, and 20 days shorter, respectively), using the monthly climatologies of SSTs as a lower boundary forcing. The CO2 concentration was set to the AMIP-specified value of 348 ppmv, and the solar constant was set to AMIP-specified 1,365 W m−2.
We also use the Global Precipitation Climatology Project (GPCP) data (Adler et al. 2003) for precipitation, and the National Centers for Environmental Prediction (NCEP)/National Center for Atmospheric Research (NCAR) reanalysis data (Kalnay et al. 1996) for wind, sea-level pressure (SLP), temperature, and specific humidity, to validate the model simulations. We note that we obtained qualitatively similar results even when we used the NCEP-DOE reanalysis 2 data (Kanamitsu et al. 2002) and ECMWF reanalysis data (Uppala et al. 2005), instead of the NCEP/NCAR reanalysis data.
2.2 Methodology: self-organizing maps
The input data is first prepared from both observed and simulated daily rainfall anomalies, by interpolating the observed data into the T42 grid of the AGCM over the southern African region (43.254–12.558°S, 0°–50.625°E). Since the daily rainfall data of the GPCP is available only from 1997 to 2008, we focus on the rainy seasons (November to February) from 1997/98 to 2007/08. As a result, the input matrix consists of 228 grids points with 19 grids in the zonal direction and 12 grids in the meridional direction, and 21,120 days of data (1,320 days of data for the observations and five ensembles of 1,320 days of data for each version of the AGCM). We note that simulated daily precipitation data are used without taking the ensemble mean when we perform the SOM analysis. Then, the dimension of the two-dimensional SOM array is chosen to be 5 nodes × 4 nodes. The topology of the array is selected to be rectangular, and the reference vectors are initialized to random values. We have chosen to use a “bubble” function for the neighborhood function. The training is undertaken in two steps; we use a larger initial learning rate and a neighborhood radius for the first phase to put reference vectors in an order, and a smaller rate and radius to tune the values of reference vectors in the second phase. As a result, we have obtained 20 different daily precipitation patterns, which will be discussed in Sects. 3 and 4.
2.3 Methodology: equitable threat score
3 Seasonal variation
4 Interannual variation
During El Niño years, nodes associated with TTTs in the southern (northern) part of the domain appear less (more) frequently in the observations; nodes D1–D5 (A1–C1) have negative (positive) anomalies. This is well captured by FrAM_Tiedtke, as is also evident from the fact that this version has the highest phase synchronization among the three versions. However, all versions fail to capture positive anomalies in nodes A2–A5 that show dry conditions over southern Africa. This is one of the reasons why the phase synchronization remains around 50 for the all three versions.
On the other hand, nodes representing TTTs are observed to occur more frequently during La Niña years (note the positive anomalies for nodes A1, B1 and D2–D5 in Fig. 18). FrAM_Emanuel best captures positive anomalies in these nodes with three nodes showing positive anomalies. In contrast to the situation during El Niño years, FrAM_Emanuel and FrAM_Tiedtke tend to perform better in capturing the negative anomalies in Nodes A2–A5. Because these dry patterns appear less frequently, the southern African region experiences more rainfall during La Niña years in general. As a result, FrAM_Emanuel and FrAM_Tiedtke have a high phase synchronization of 70 and 65, respectively, whereas FrAM_Kuo has a low phase synchronization of 45.
As expected in the absence of strong influences from ENSO, the frequency anomaly is within ±1 % in more than 80 % of the nodes during the normal years. FrAM_Tiedtke (FrAM_Kuo) has the highest (lowest) skill with the phase synchronization of 65 (35).
In summary, the model versions perform better in simulating the interannual variations in the precipitation pattern for La Niña years compared to El Niño or normal years, and FrAM_Emanuel and FrAM_Tiedtke have higher skills in general compared with FrAM_Kuo.
5 Summary and discussions
Using three versions of the same AGCM differing only in the convection scheme, we have evaluated skills of models in simulating southern African rainfall and TTT attributes. All three versions have relatively good capabilities in simulating the summer precipitation, although one version (FrAM_Tiedtke) has a serious bias in the onset. This version simulates excessive upper level convergence and associated subsidence over southern Africa. As a result, development of TTTs is suppressed and connection of tropical and extra-tropical precipitation is delayed by about 1 month. It is interesting to note that for all three versions, the ability to represent the climatology of monthly rainfall patterns is lowest in November. Since the onset of rainy season is very important for subsistence farming in the southern African region, this model bias is potentially a limiting factor to the skill of early-season seasonal forecasts over the region.
Regarding the simulation of interannual variation, all three versions have relatively good skill, particularly in equatorial East Africa and South Africa. In addition, they are successful in capturing negative (positive) precipitation anomalies over southern Africa in El Niño (La Niña) years, although the exact location of peak precipitation anomalies is slightly shifted. When synoptic precipitation patterns are examined using SOMs, we have found that nodes associated with TTTs in the southern (northern) part of the domain are observed less (more) often during El Niño years. In contrast, nodes associated with TTTs occur more frequently during La Niña years. Also, nodes associated with dry conditions over southern Africa appear more (less) frequently during El Niño (La Niña) years.
Interestingly, the models have better skill in simulating precipitation anomalies during La Niña years, and this may explain why forecast skills have been found to be higher during La Niña years (Landman and Beraki 2012). Because of limitation in the length of daily precipitation data, we note that there are only two (three) El Niño (La Niña) events in the composites, and the analysis should be repeated after the accumulation of observation data.
However, this study is the first to illustrate that the usage of different convection schemes in an AGCM can have pronounced effects on the simulation of southern African rainfall in austral summer. In fact, the study shows that the simulation of upper level circulation and TTT attributes are sensitive to the choice of cumulus convection scheme. We therefore expect that the results presented in this study may shed new light on simulation and prediction of the precipitation over the southern Africa region.
Constructive comments from two anonymous reviewers helped us to improve our manuscript. The AGCM was run on the supercomputers of Information Technology Center, the University of Tokyo under the cooperative research with Atmosphere and Ocean Research Institute, the University of Tokyo. The SOM_PAK software was provided by the Neural Network Research Centre at the Helsinki University of Technology and is available at http://www.cis.hut.fi/research/som_pak. The present research is supported by Japan Science and Technology Agency and Japan International Cooperation Agency through Science and Technology Research Partnership for Sustainable Development (SATREPS). The first author is supported by the Japan Society for Promotion of Science through Grant-in-Aid for Exploratory Research 24654150. The second and third authors acknowledge the support from the Applied Centre for Climate and Earth System Studies (ACCESS, South Africa), and National Research Foundation (NRF, South Africa) in performing this research.
- Adler RF, Huffman GJ, Chang A, Ferraro R, Xie P-P, Janowiak J, Rudolf B, Schneider U, Curtis S, Bolvin D, Gruber A, Susskind J, Arkin P, Nelkin E (2003) The version 2 global precipitation climatology project (GPCP) monthly precipitation analysis (1979-present). J Hydrometeor 4:1147–1167CrossRefGoogle Scholar
- Guan Z, Iizuka S, Chiba M, Yamane S, Ashok K, Honda M, Yamagata T (2000) Frontier atmospheric general circulation model version 1.0 (FrAM1.0): model climatology. Tech Rep 1:27Google Scholar
- Kalnay E, Kanamitsu M, Kistler R, Collins W, Deaven D, Gandin L, Iredell M, Saha S, White G, Woollen J, Zhu Y, Leetmaa A, Reynolds R, Chelliah M, Ebisuzaki W, Higgins W, Janowiak J, Mo KC, Ropelewski C, Wang J, Jenne R, Joseph D (1996) The NCEP/NCAR 40-year reanalysis project. Bull Am Meteorol Soc 77:437–471CrossRefGoogle Scholar
- Louis J, Tiedtke M, Geleyn JF (1982) A short history of PBL parameterization at ECMWF. Workshop on Planetary Boundary Layer Parameterization, ECMWF, pp 59–80Google Scholar
- MacKellar N, Tadross M, Hewitson B (2010) Synoptic-based evaluation of climatic response to vegetation changes over southern Africa. Int J Climatol 30:774–789Google Scholar
- Nicholls N, Uotila P, Alexander L (2010) Synoptic influences on seasonal, interannual and temperature variations in Melbourne, Australia. Int J Climatol 30:1372–1381Google Scholar
- Shibata K (1989) An economical scheme for the vertical integral of atmospheric emission in longwave radiation transfer. J Meteorol Soc Jpn 67:1047–1055Google Scholar
- Tozuka T, Miyasaka T, Chakraborty A, Mujumdar M, Behera SK, Masumoto Y, Nakamura H, Yamagata T (2006) University of Tokyo coupled general circulation model (UTCM1.0). Ocean-Atmos Res Rep 7:44Google Scholar
- Van den Heever SC, D’Abreton PC, Tyson PD (1997) Numerical simulation of tropical-temperate troughs over southern Africa using the CSU RAMS model. S Afr J Sci 93:359–365Google Scholar