Preparation of the simulations
The initial and boundary conditions need to be defined prior to running a simulation with Méso-NH. There are different possibilities for the lateral bound-ary conditions. They can be cyclic/rigid wall/open; coupled with the large-scale fields given by the ARPEGE (Action de Recherche Petite Echelle Grande Echelle) or ECMWF (European Centre for Medium-range Weather Forecasts) models, forecast or analysis data, or with the higher resolution fields obtained from limited domain models like ALADIN (Aire Limitée, Adaptation dynamique, Développement InterNational — ARPEGE version on a domain located in Euro-pean region) or AROME (Applications of Research to Operations at MEsoscale); or defined by the interactive grid-nesting. In this work, the ECMWF meteorolog-ical analyses are used to prescribe the prognostic model variables at the initial and boundary conditions for the variables existing in ECMWF, otherwise the variables are set as zero. The initialisation of the model is done at (1) 0000 UTC 6 April 2009, (2) 0000 UTC 23 November 2011, and (3) 0000 UTC 9 February 2013, where the ECMWF horizontal resolution is Δx=25 km for the case of 2009 and Δx=16 km for the cases of 2011 and 2013. During the simulation, files are coupled with the ECMWF analyses every 6 h, the boundary conditions for other time values are calculated as the linear approximation between two consecutive ECMWF analyses used as coupling files. No large-scale forcing is applied.
Before the start of a simulation, a physiographic file is produced con-taining the surface cover type which is obtained from the global ECOCLIMAP database (Masson et al., 2003). There are 250 available surface cover types which describe not only the landscape, but the whole ecosystem. There ex-ist 15 land cover types e.g. different types of forest, soil, snow, wetlands, ur-banized areas, and 16 climate types e.g. different types of tropical, marine, desert climate (sometimes different ecosystems can be aggregated into one rep-resentative system), thus giving vegetation parameters for areas all around the world. Orography and land cover are obtained from the GTOPO30 databases (https://lta.cr.usgs.gov/GTOPO30) with the resolution 30 arc second (∼1 km). As we consider a domain close to the equator, we chose the Mercator projection. For using in the model, the surface data are interpolated over the grid mesh. A good representation of surface features is important for a realistic reproducing of meteorological phenomena, as they can be often influenced or generated by the topography: precipitation, mesoscale circulations, blocking of synoptic systems. The locations of land and water as well as different types of land cover are necessary to calculate the surface fluxes of moisture, heat and momentum. Four different types of the surface (natural landscape, urban zone, sea and inland water) are deduced from the cover types. During a Méso-NH simulation, a specific surface scheme is used for each of these types. Then, the physiographic data together with information on the vertical grid and with the interpolated ECMWF analyses are used to start a simulation or for the boundary conditions.
In the vertical, we use the Gal-Chen and Sommerville coordinate following the surface elevation. In our case, the vertical grid spacing varies from 60 to 600 m. In the all simulations we used 72 vertical levels. The upper boundary is supposed to be a rigid horizontal lid with a free slip condition on the atmosphere. To damp the gravity waves generated by convection, a sponge layer is usually in-serted near the upper boundary and the prognostic variables are relaxed towards the large-scale values in this sponge layer. In our case, the model top was set to 30 km with the upper 3 km being a sponge layer. An additional absorbing zone is located close to the lateral boundaries, it allows to damp outward propagating waves or to slowly incorporate inward propagating larger scale waves.
Simulations with the Méso-NH model
While low-resolution simulations (with the horizontal resolution of 10–100 km) give the possibility to explore the large-scale motions, in such simulations the processes that are not resolved within grid spacings, need to be parameter-ized. Thus a question of the quality of these simulations arises (e.g. the review of Prein et al. (2015)). The convection parameterization schemes used in low-resolution simulations have been identified as a main reason of model errors and uncertainties (Ellingson et al., 1991; Henderson-Sellers et al., 1993; Peder-sen and Winther, 2005; Déqué et al., 2007). One especially important subgrid scale process in climate models is the deep convection that is a major source of precipitation in many regions, i.e. the tropics. The triggering of deep convec-tion depends on the interaction of processes at scales from the microscale to the synoptic scale, thus making the parameterizing of deep convection very compli-cated. The convection parameterization schemes can thus introduce errors at different scales, from the diurnal cycle of convective precipitation (Brockhaus et al., 2008) to intraseasonal or interannual variability (Song and Zhang, 2009; Zhang and Song, 2010; Chikira, 2010). High-resolution simulations no longer need convective parameterizations, and moreover, the orography in such simu-lations is more accurately represented. The added value of the high-resolution simulations exists thus in regions where deep convection is important and in re-gions with a strongly heterogeneous orography, making the high-resolution sim-ulations useful for studying atmospheric processes that are difficult to observe directly.
Using the Méso-NH model, we can compare the simulations with the dif-ferent resolutions. To estimate the impact of the explicit-convection modeling i.e. to analyze the effects of the resolution and representation of convection on precipitation, low-resolution simulations were also performed. In total, eight simulations were performed to investigate three different episodes of the MJO passage over the Indian Ocean and the Maritime Continent. All simulations were performed with the Méso-NH version 5-1-3 at the supercomputers Occigen We consider three time periods when the MJO was active over the Indian Ocean and the Maritime Continent: (1) 6–14 April 2009 (9 days), (2) 23–30 November 2011 (8 days), and (3) 9–28 February 2013 (20 days). We concen-trate on the Indian Ocean and the Maritime Continent, where the MJO propaga-tion is generally the most visible. We performed simulations for a large domain 26.7◦S–26.7◦N, 44.6◦–155.4◦E which is shown in Fig. 2.5 (different subdomains of the simulation domain are used for different types of analysis). This domain was chosen to describe the MJO propagation in different regions and compare the MJO properties in active and suppressed phases of the MJO during its pas-sage over the same location. The choice of a large simulation domain was possi-ble thanks to to the parallel computing capability of Méso-NH model (Pantillon et al., 2011).
The HiRes simulations
For every time period (2009, 2011, and 2013), a high-resolution simulation is performed where convection is calculated explicitly. In this thesis, these sim-ulations will be referred to as HiRes-2009, HiRes-2011, and HiRes-2013. In all these simulations, the 1D turbulence scheme is used with the mixing length of Bougeault and Lacarrère (1989). For the 2011 period, we have also con-ducted two different sensitivity tests. In the first one, a HiRes simulation (HiRes-2011a) was performed using “all or nothing” method of cloud paramerization instead of the statistical subgrid scale cloud scheme described in Chapter 2.1.3. In the second sensitivity test, a HiRes simulation (HiRes-2011b) was performed using the 3D turbulence scheme with the mixing length of Deardorff (1980).
MJO indices and NCEP/NCAR/NOAA data
MJO indices. Different MJO indices can be used to identify the MJO phase and to determine the location of the MJO on the globe, as explained previ-ously (Chapter 1). In this work, they are used to evaluate the ability of the simulations to reproduce the MJO propagation and to explain the differences between the different MJO episodes. Here, the OLR MJO Index (OMI) (Ki-ladis et al., 2014), the Velocity Potential MJO index (VPM) (Ventrice et al., 2013), and Real-Time Multivariate MJO series (RMM) (Wheeler and Hendon, 2004) are used. OMI is constructed by the projection of 20–96 day filtered OLR onto the daily spatial EOF patterns of 30–96 day eastward filtered OLR. (For the comparison of OMI and RMM at the same diagram, the sign of OMI PC1 and the PC ordering need to be reversed, as OMI(PC2) is analogous to RMM(PC1) and –OMI(PC1) is analogous to RMM (PC2).) VPM index is cal-culated similarly to RMM, but in the combined EOF calculation 200-hPa ve-locity potential is used instead of OLR, together with U200 and U850. OMI and VPM indices were obtained from the NOAA/ESRL Physical Sciences Divi-sion (PSD) Web site: https://www.esrl.noaa.gov/psd/mjo/mjoindex/ and the RMM index was obtained from the Bureau of Meteorology (Australia) Web site: http://www.bom.gov.au/climate/mjo/. NCEP/NCAR datasets. To obtain the wind climatology fields necessary to calculate the wind anomalies, the datasets of the US National Centers for Environmental Prediction (NCEP)/National Center for Atmospheric Research (NCAR) 40-year Reanalysis Project are used. In the NCEP/NCAR 40-year Reanal-ysis Project, the atmospheric fields were recovered and later assimilated with a data assimilation system that is kept unchanged over the reanalysis period (Kalnay et al., 1996). Different sources of observation (land stations, aircraft, satellites etc.) were used for developing these datasets, including the data that were not available in real time and were provided later by many countries. In this work, NCEP/NCAR reanalyses for the zonal wind velocity “U, NCEP/NCAR Reanalysis 1” are used (https://www.esrl.noaa.gov/psd/). The spatial reso-lution of the zonal wind velocity is 2.5◦x2.5◦ (144×73) (Kalnay et al., 1996) for the global domain 90◦N–90◦S, 0.5◦E–359.5◦E.
For the data analysis, we use the NCAR Command Language (NCL) (Version 6.1.2): Boulder, Colorado: UCAR/NCAR/CISL/VETS, http://dx.doi.org/10. 5065/D6WD3XH5.
Initial conditions: distribution of surface pressure and SST
The initial distributions for the surface pressure and the SST for all cases of study is shown in Fig. 3.2. The surface pressure (Fig. 3.2a,c,e) has lower values (620–1000 hPa) over the land, while over the ocean the pressure is up to 1020 hPa. In all cases, there is a zone of lower pressure over the equatorial Indian Ocean (1000–1005 hPa). For the case of 2011, the surface pressure over the Indian Ocean and the Maritime Continent is lower than for the other two cases. The pressure is higher over the ocean part in the north and the south of the simulation domain.
The SST (Fig. 3.2b,d,f) has lower values at higher latitudes in the south and the north of the simulation domain. The highest values are located at the north-ern and the equatorial part of the Indian Ocean, the ocean part of the Maritime Continent, and near the northern Australia.
Distribution of precipitation
The spatial distribution of the mean precipitation for TRMM, LowRes, and HiRes for every episode is shown in Figs. 3.3–3.5. For better visualization, HiRes and LowRes simulations were adapted to TRMM grid.
For the 2009 event (Fig. 3.3), TRMM shows the most intensive precipitation over the central and eastern Indian Ocean with amplitudes up to 30 mm day−1. Strong precipitation exists also over the Maritime Continent, especially in its central region (Sumatra and Borneo). The Severe Tropical Storm Jade (3–14 April) is seen that was located around (50◦E, 16◦S) on 4 April and then moved towards the southern boundary of the simulation domain. Also, the initiation of the Cyclonic Storm Bijli (14–17 April) is visible that started to form on 14 April from an organized area of rains and thunderstorms over the southeast Bay of Bengal.
In LowRes, there exists a deficit in precipitation over the central Indian Ocean and an excess in the southwestern part of the simulation domain. Over the Mar-itime Continent, the precipitation is mostly concentrated over the islands. HiRes reproduces the mean precipitation pattern over the Indian Ocean better, even though it still gives a precipitation deficit near the equator over the Indian Ocean and over the ocean part of the Maritime Continent. In both simulations, the Se-vere Tropical Storm Jade is also visible and gives stronger precipitation than in TRMM, while the formation of the Cyclonic Storm Bijli is not seen. Another com-mon feature of the two simulations is the presence of a large amount of weak rain (less than 1 mm day−1) which is absent in TRMM. Nevertheless, TRMM precipitation product might be not able to well reproduce weak rain rates due to such factors as instrument measurement thresholds, calibration and combina-tion of different instruments, or the approximations used in the calculating of 3B42 product, thus it is not possible to estimate the quality of the simulations for weak precipitation.
For the 2011 event (Fig. 3.4), strong precipitation in TRMM is located along the equator over the Indian Ocean extending to the south and north up to 10◦–15◦ latitudes. Precipitation is also intensive over the Maritime Continent though its amplitude is weaker. Also, the path of the Deep Depression ARB 04 (26 November – 1 December) is visible moving along the southwestern side of the Indian subcontinent. In both LowRes and HiRes, the areas where the inten-sive precipitation is concentrated are fairly well reproduced. LowRes however shows weaker precipitation over the Maritime Continent and an excess of pre-cipitation in the northeast and the west of the simulation domain. HiRes gives more scattered precipitation but the spatial patterns are well reproduced. The path of the Deep Depression ARB 04 is clearly visible in LowRes while having higher intensity than in TRMM. In HiRes, it is visible but much weaker.
For the 2013 event (Fig. 3.5), the strongest precipitation in TRMM is located in the eastern part of the simulation domain. It is weaker over the Indian Ocean. In both simulations, there is a deficit of precipitation near the equator over the Indian Ocean. The precipitation over the Maritime Continent is located mostly over the islands, contrary to TRMM. The simulations also overestimate the rain-fall associated with cyclones, located around 15◦–20◦S: Tropical Cyclone Haruna (14–28 February, 45◦E), Tropical Cyclone Gino (11–15 February, 80◦E), and Se-vere Tropical Cyclone Rusty (22 February – 5 March, 120◦E). The cyclones are less noticeable in TRMM data.
To provide a quantitative assessment, precipitation accumulated over the sim-ulation period is calculated over the whole domain (Table 3.1). The both sim-ulations have lower values of the total accumulated precipitation in the 2009 and 2013 episodes than TRMM. LowRes shows an excess in the total accumu-lated precipitation as compared to TRMM, in the 2011 episode. HiRes gives the lowest values for all three episodes.
When examining the average total precipitation values per day (Table 3.2), the episode of 2011 had the highest average total precipitation values per day as compared to 2009 and 2013 episodes. The simulations show less precipitation than TRMM with the exception of LowRes in 2011.
Table of contents :
Introduction (en français)
1 The Madden–Julian oscillation
1.1 Overview of the MJO
1.2 Processes governing the evolution of the MJO
1.3 Observations and simulations of the MJO
2 Model, data and methods
2.1 The Méso-NH model
2.2 Simulations with the Méso-NH model
2.3 Other datasets
3 Assessment of the Méso-NH simulations
3.1 Overview of the MJO events
3.2 Simulations of three MJO events
3.3 Sensitivity of the 2011 simulation to the cloud fraction and turbulence parameterizations
4 Atmospheric overturning during the MJO 103
4.2 Article: The three atmospheric circulations over the Indian Ocean and the Maritime Continent and their modulation by the passage of the MJO
Conclusions and perspectives
Conclusions et perspectives (en français)