1. Introduction
This paper is a sequel to Noble et al. (2014, hereafter Part I), which analyzed the sensitivity of the National Center for Atmospheric Research’s (NCAR) Weather Research and Forecasting (WRF) Model daily circulation over West Africa to alternative model parameterizations. The WRF is a numerical weather prediction (NWP) model used for atmospheric science research. WRF incorporates a community model infrastructure that facilitates scientific collaboration, whereby the operational and research communities work jointly toward the development of next-generation NWP capabilities by giving the research community better access to operational codes for investigating modeling successes and failures (Bernardet et al. 2008). The infrastructure has also allowed rapid adaptation of the WRF for regional climate model (RCM) applications (Liang et al. 2012). An RCM includes coding to compute atmospheric processes plus soil hydrology, and accounts for the boundary forcing of high-resolution terrain, land–sea contrasts, surface characteristics, and other components of the Earth system. RCMs are driven by time-variable conditions along their lateral atmospheric boundaries and are thus used to downscale global retrospective analyses (or reanalyses) or multidecadal global climate models (GCMs) to provide information for climate, climate variability, and change with regional refinements (Giorgi and Mearns 1991).
Despite the WRF Model’s worldwide popularity, the model was designed for short-range weather prediction over North America and was not intended for long-term climate simulations (Dudhia 2014). Recent studies have investigated its capabilities for North American RCM applications (Caldwell et al. 2009; Liang et al. 2012). WRF RCM applications in tropical regions have encountered numerous problems, including deficient rainfall at the equator and precipitation overestimation over West Africa (Druyan et al. 2009). To remedy unsatisfactory simulation skill for West, East, and South African precipitation, several studies have tested model sensitivities to different physics parameterizations (Flaounas et al. 2011; Pohl et al. 2011; Crétat et al. 2012). This paper focuses on WRF simulation skill for West Africa, and takes the optimization step emphasized by Flaounas et al. (2011) further, by analyzing the multiple combinations of five parameterizations—in effect considering 64 different model configurations.
West Africa includes the Sahel, a semiarid transition zone at 10°–20°N bounded by the Sahara Desert to the north and a rainy region to the south that includes Senegal, Mali, Mauritania, Niger, Burkina Faso, Chad, and Sudan. The rainy season occurs during boreal summer when the West African monsoon (WAM) brings the largest portion of the annual rainfall to the Sahel. The Sahel remains dry during the winter and spring months until rains arrive in late June from a northward inland surge of moist monsoon air from the Gulf of Guinea, marking the onset of the WAM (Sultan and Janicot 2000). Scientific attention has focused on the Sahel because the area suffered a series of severe droughts during the late 1960s to the 1980s. To better understand its weather and climate variability, the Sahel has been the site of major field experiments, including the JET2000 Experiment (Thorncroft et al. 2003) and the African Monsoon Multidisciplinary Analysis (AMMA), which included comprehensive field measurements during the 2006 boreal summer (Redelsperger et al. 2006). This paper evaluates WRF precipitation simulations during the AMMA Special Observing Period 3 (SOP3).
Although model-based climate studies relate to monthly or seasonal mean fields, confidence in simulations of the seasonal climate improves if the model realistically captures the characteristics of relevant daily weather phenomena. The current study aims to examine the sensitivity of WRF skill for downscaling reanalysis at daily time scales, a first step in evaluating its performance for monthly to seasonal time scales. In particular, the current study evaluates its WRF simulations of spatial precipitation patterns associated with African easterly waves (AEWs). AEWs are westward-propagating synoptic atmospheric waves, best monitored at 850, 700, or 600 hPa, with wavelengths of 3000–5000 km and two distinct period bands: 2–6 and 6–9 days. AEWs are a major source of synoptic-scale precipitation variability over West Africa. (Diedhiou et al. 1999; Druyan et al. 2008; Burpee 1972). Accordingly, WRF performance should be evaluated in terms of how well AEW-associated circulation and precipitation are represented.
There is considerable literature that relates precipitation to AEWs. Several studies find that convection and precipitation are more prevalent within and ahead of the trough. Composites of AEWs show rainfall maxima southwest of the 700-hPa wave apex (Reed et al. 1977; Druyan et al. 2008). Satellite infrared imagery shows that a preferred location for convective rainfall over West Africa is ahead of AEW troughs (Payne and McGarry 1977). Rain gauge investigations note positive rainfall anomalies in and ahead of AEW troughs (Diedhiou et al. 1999).
Other studies found that some convection occurred east of the trough, in the southerly flow. AEWs over the northern Sahel (17.5°N) are more likely to have moist mesoscale convection east of troughs, while AEWs at more southerly latitudes (7.5°N) seem to be rainy more often at and west of troughs (Fink and Reiner 2003), linked to convection and rainfall over the Sahel region (Ruti 2010). Rainfall totals can be greatest just prior to AEW trough passage within AEW troughs, or east of troughs within the strongest 700-hPa southerlies (Druyan et al. 2008). In a study based on TRMM precipitation radar and ERA-I, Janiga and Thorncroft (2016) determine that high summertime precipitation rates over the Sahel favor the northerly phase of AEWs (west of troughs), while AEW-related rainfall over the western coast of Africa and the eastern Atlantic shifts toward the trough where the air is moist. They also suggest that climate simulation models with parameterized convection tend to underestimate rain rates west of the troughs and overestimate rain rates within the troughs and to their east. The most frequent preference for heavy precipitation southwest of the apex relates to the broad area of lower-tropospheric convergence created by AEWs west of the 700-hPa trough (Druyan et al. 1996).
AEWs are not always accompanied by precipitation, as found by Taleb and Druyan (2003). Analysis of observations suggests that only about one-third of seasonal precipitation totals is recorded on days affected by statistically significant wavelet amplitudes of 700-hPa meridional wind in the 3–5-day period band.
GCM simulations show westward propagating rainfall “footprints” presumed to be associated with AEWs on time–longitude Hovmöller plots (Xue and Shukla 1993). However, squall lines and mesoscale convective systems (MCSs) can often propagate at different speeds relative to their associated AEWs (Fortune 1980). Westward-propagating precipitation bands, many, presumably, embedded within AEWs, appear as diagonal swaths on time–longitude (Hovmöller) diagrams of TRMM daily observations over West Africa (Druyan et al. 2006).
MCSs, which produce most of the precipitation during the summer monsoon in West Africa, move more quickly than AEWs on average and less than a half are associated with AEWs (Lebel et al. 2003). Radar-based studies of the same period as this study found that MCS precipitation in West Africa is more influenced by conditional stability and vertical wind shear than by AEW phase (Guy et al. 2011). Berry and Thorncroft (2012) show in WRF simulations that organized deep convection embedded within an AEW makes a large contribution to the synoptic-scale mean potential vorticity and therefore to the energetics of the AEW. Their WRF experiment demonstrates that convection is vital for the maintenance of the AEW over West Africa and suggests that AEWs require active convection to persist for an extended length of time.
Multiple alternative WRF parameterizations are available to represent physical processes: convection parameterization schemes (CPSs), PBL schemes, land surface models (LSMs), microphysics parameterization schemes (MPSs), and radiation-transfer schemes (RADs). A configuration of WRF parameterizations must be selected for each application and perhaps for each geographic region. Testing and validating WRF parameterizations is especially important because time mean precipitation simulations are sensitive to model configuration (Lynn et al. 2009). Thus, the motivation here is to test the sensitivity of WRF simulations of West Africa weather and climate to model configuration. Suggestions of skillful WRF configurations resulting from this evaluation should benefit other investigators planning WRF applications over West Africa. Part I reported on configurations for circulation simulations, while this paper discusses WRF skill in simulating precipitation.
Summary of WRF circulation performance
Part I compared 12-day simulations of circulation and 700-hPa vorticity over West Africa from 64 alternative WRF configurations to MERRA, NCEP–DOE AMIP-II reanalysis, and radiosonde soundings at two locations. The largest favorable impact on WRF vorticity simulation is realized by selecting the Grell–Devenyi (GD) cumulus convection scheme (Grell and Dévényi 2002). This preference may reflect the more versatile skill of that scheme’s ensemble approach to computing the two-way interaction between convection and large-scale circulation within the very complex WAM regime. Vorticity simulations using GD achieve higher correlations with reanalysis than the remaining simulations using the Kain–Fritsch (KF) convection scheme (Kain 2004). Impacts of other model parameterizations are more ambiguous. Selected WRF configurations have skill in simulating vorticity attributed to AEWs up to 9 days into the simulation for 2006, but skill drops sharply thereafter.
Berry’s (2009) analysis of selected WRF simulations over West Africa on a 125-km grid for September 2004 is relevant. He tests the sensitivity of simulations to three cumulus parameterization schemes: Betts–Miller–Janjić (Vaidya and Singh 2000), KF, and GD. Berry finds that GD simulates the most realistic representation of the westward-propagating AEWs that cross the African coast on 2, 4, 7, and 9 September 2004. Moreover, using GD also produces the most realistic representation of precipitation. Of the three schemes, only GD simulates all four AEWs since the KF and BMJ schemes dissipate one of the AEWs and propagate another one too slowly. Premature dissipation of AEWs can result from underestimates of deep convection that inhibit the generation of potential vorticity. This, in turn, limits the dynamics that strengthen AEWs (Berry and Thorncroft 2012).
Validation statistics from Part I for selected WRF configurations simulating the parallel period during 10 additional years indicate that results for 2006 are much more favorable than for other years. In fact, during most years, even the best WRF configurations fail to produce 700-hPa vorticity distributions with any correlation to reanalysis after only a day or two. Discrepancies between simulated and observed trajectories of AEW vorticity centers negatively impact model simulations of precipitation variability since significant precipitation maxima during the summer are associated with AEWs.
2. Methods and data
a. Simulation period
The current study examines simulations for 2–13 September 2006, the same 12-day period analyzed in Part I. This period had four well-defined AEW troughs with embedded growing and decaying convective activity of various sizes, durations, and intensities, including the development of the pre-Helene tropical storm at the end of the period (Franklin and Brown 2008; Cifelli et al. 2010). The selected time period also occurs during the combined AMMA and NASA AMMA (NAMMA) field campaigns (Redelsperger et al. 2006; Zipser et al. 2009) during August and September 2006. Previous WRF modeling studies have examined this period (2011; Druyan et al. 2009; Chiao and Jenkins 2010; Flaounas et al. 2011, 2012; Cook and Vizy 2009) but none examined the sensitivity of WRF simulations to the wide variety of parameterization schemes available for WRF, save for Part I. The short evaluation period allows the analysis to be based on some 104 simulations.
b. WRF configurations
This paper evaluates version 3.2.1 of the Advanced Research version of WRF (WRF-ARW; Skamarock et al. 2008, henceforth WRF), which employs the ARW dynamics solver, or “Eulerian mass” solver. The WRF is a modular, nonhydrostatic, and fully compressible model that uses the sigma vertical coordinate to better simulate airflow over complex terrain. WRF in these experiments uses a 20-km horizontal grid at 30 terrain-following sigma layers between the earth’s surface and the 50-hPa model top.
Figure 1 shows the model domain and topographic features used for simulations. Elevations are derived from the global 30 arc s elevation dataset from the U.S. Geological Survey. The red box designates the region for which validation scores are compiled—an area noted for AEWs and precipitating deep convection. The initial conditions and lateral boundary conditions (LBCs), including sea surface temperatures, for all simulations in this study were specified from the National Centers for Environmental Prediction–U.S. Department of Energy Atmospheric Model Intercomparison Project Reanalysis II NCEP–DOE II (hereafter referred to as NCEP2) (Kanamitsu et al. 2002). The NCEP2 data are interpolated from pressure levels on a 2.5° Gaussian grid and 6-hourly temporal availability to the WRF grid using the WRF preprocessing system. Reanalysis over sparse data regions like West Africa provides LBC and validation data that are not always a perfect representation of actual conditions, but the gridded interpolation is nevertheless anchored to and consistent with the observed meteorology. Therefore, initializing and driving the WRF with NCEP2 should represent a high level of potential skill of WRF simulations. No nudging or interactive nesting was used in any of the experiments.
WRF Model computational domain with terrain elevation contours at 250-m intervals. The red box outlines the region validated for precipitation variability.
Citation: Monthly Weather Review 145, 1; 10.1175/MWR-D-15-0294.1
Numerical experiments are designed to evaluate WRF performance for different combinations of the WRF parameterizations that are identified in Table 1. This study tests 64 configurations for simulation using the following parameterizations: CPS from KF and GD; PBL physics from Yonsei University (YU), the Mellor–Yamada–Janjić (MJ), and Mellor–Yamada–Nakanishi–Niino (MN) schemes; surface physics from the five-layer thermal diffusion LSM (5L), the Noah (NO) LSM, the Rapid Update Cycle (RU) LSM, and the Pleim–Xue (PX) LSM; radiation physics from the Rapid Radiative Transfer Model (Rt) and the Community Atmosphere Model (CM); and microphysics from the WRF single-moment 5-class microphysics scheme (W5). Of course, given the multiplicity of parameterizations available for WRF, it is impossible to test every possible combination (Liang et al. 2012).
WRF parameterizations considered for all experiments, with abbreviated code used throughout this study and source citation.
Table 2 shows the 64 numerical experiments tested using data from 2006, configured by changing one parameterization at a time, starting with the first option and its acronym listed in Table 1. Experiments are labeled consecutively, 1–64. Although numerous, the combinations tested here represent a small subset of testable possibilities. Note that WRF experiment 2 is the recommended WRF reference configuration for the WRF version 3.2.1 NWP (Wolff 2011; Harrold 2012) and for tropical simulations (J. Dudhia 2008, personal communication). Hereafter, WRF experiment 2 is referred to as WRF2 and each WRF experiment is referred to in parallel fashion (e.g., WRF experiment 27 = WRF27). It is possible that no single WRF configuration is optimal under all circumstances (location, times of day, season, etc.). The reader is referred to Skamarock et al. (2008) for further details of all parameterizations in Tables 1 and 2.
The 64 different WRF Model combinations of parameterizations included in the sensitivity analysis.
The MPS explicitly resolves water vapor, cloud, and precipitation processes. Pohl et al. (2011) and Crétat et al. (2012) find that MPSs exert a minor influence on the location, intensity, and number of rainy events over eastern and southern Africa. A number of single-moment MPS schemes can be used in WRF: 3-class (WSM3), 5-class (WSM5), and 6-class (WSM6; Hong et al. 2004). Both the WSM5 and WSM6 are suitable for domain resolutions less than 25 km (Hong et al. 2006). This study adapts the computationally less-expensive WSM5. Table 2 does not show MPS since the WSM5 is used in all WRF configurations. However, the discussion refers to a sensitivity test for the impact of alternative MPSs.
c. Validation datasets
To assess the skill of the simulations, model results are compared to satellite rainfall estimation products (SREPs) described in Table 3. The conventional rain gauge network in West Africa is inadequate for validating spatial and temporal details of the precipitation simulations. SREPSs are useful alternatives to gauge measurements because of their better spatial and temporal coverage. Several independently derived SREPs—albeit based on similar measurements—are also referenced.
Validation data.
Differences among SREPs reflect the degree of confidence that we can place in the verification analysis. The TRMM algorithm 3B42 product is one of the best sources for merged high quality precipitation estimates, particularly because it provides 3-hourly high spatial resolution (0.25°) estimates (Huffman et al. 2007). TRMM does not include any ground validation in its submonthly products. However, the alternative NASA Global Precipitation Climatology Project 1° Daily (GPCP 1DD) “final” analysis SREP is calibrated with rain gauge data and is, perhaps, the best historical precipitation product available for this purpose (Huffman et al. 2009). Results are also compared with the Climate Prediction Center morphing technique (CMORPH; Joyce et al. 2004) and the Precipitation Estimation from Remote Sensed Information Using Artificial Neural Networks (PERSIANN; Behrangi et al. 2009). Modeled precipitation data are also compared to MERRA and the NCEP2 reanalysis, which provide independent estimates at Δx, Δy = 1.25° and 2.5°, respectively.
To better assess WRF’s potential to characterize West African weather patterns, results from another model are presented as a benchmark. The NASA Center for Climate Systems Research Regional Model version 3 (NASA GISS/CCSR RM3, hereafter RM3) is run over the same domain on a 0.5° grid for the same 12-day period and forced by the same LBCs. The RM3 has previously been used to study AEWs (Druyan et al. 2006, 2009, 2010a) and longer-term climate simulations in the West African Monsoon Modeling Evaluation (WAMME) initiative (Druyan et al. 2010b). Druyan et al. (2010b) summarize WAMME results from five RCMs simulating the WAM with both reanalysis and GCM forcing, but that study does not include the WRF Model.
3. Results
a. Total-accumulated precipitation
Distributions of accumulated precipitation (ptot) for 2–13 September 2006 are constructed for each of the 64 WRF experiments in Table 2 and then compared to SREP observations. All modeled and observed SREP ptot are compared on a 0.25° × 0.25° grid by interpolation with the first-order conservative-remapping method (Jones 1999). Figures 2a–j show the 12-day ptot results for GPCP, TRMM, WRF2, WRF27, WRF32, WRF59, WRF60, WRF64, and RM3, respectively. WRF2 is the default configuration, while the other WRF configurations validate best against 2006 SREP ptot results, as explained in section 3c. Furthermore, WRF27 and WRF32 12-day simulations of circulation and 700-hPa vorticity over West Africa validate best against reanalysis (Part I). The GPCP ptot (Fig. 2a) shows two subregions: the southern box outlines the tropical rain-belt region that includes most ptot (5°–15°N, 25°W–10°E), while the northern box (10°–20°N, 25°W–10°E) outlines the Sahel region (Nicholson 2013). The combined subregions make up the region shown in Fig. 1, for which all results are validated.
Precipitation totals ptot, for the 2–13 Sep 2006 period: (a) GPCP; (b) TRMM; (c) WRF2, the default configuration; (d) WRF27; (e) WRF32; (f) WRF59; (g) WRF60; (h) WRF64; and (i) RM3. The plot for GPCP shows the Sahel region in red and the tropical rainband region in black.
Citation: Monthly Weather Review 145, 1; 10.1175/MWR-D-15-0294.1
Differences between GPCP and TRMM ptot (Figs. 2a,b) mean that WRF validation scores against GPCP will differ from scores validated against TRMM. Moreover, the differences are a humble reminder that the actual precipitation characteristics may be elusive. The GPCP shows slightly more ptot maxima than TRMM over central Nigeria (8°–12°N, 4°–10°E), over the central savanna plains, and along the West African coast (10°–13°N, 16°–12°W), as well as a slightly wetter Sahel region. The WRF configurations produce too much rain, especially WRF2 in Fig. 2c, whereas WRF27, WRF32, WRF59, WRF60, and WRF64 in Figs. 2d–i show less exaggerated rain than the default. All WRF experiments show a very wet rainband region and a drier Sahel region compared with the observations, implying excessive simulated low-level moisture convergence along the West African coast. The RM3 in Fig. 2h produces ptot maxima similar to GPCP and TRMM over all regions (resulting in correspondingly favorable RM3 validation scores, discussed below in section 3c). None of the WRF experiments produces the observed orographic precipitation maxima over the Cameroon Highlands. WRF32 and WRF64 are the best WRF simulations in representing the maximum values in the east, the relative minima near 0° longitude and east of the Guinea Highlands, and the secondary peak along and offshore the west coast. Both WRF versions use the GD convective scheme, the MN PBL, and the PX surface model. Differences resulting from the different parameterizations are explored in section 3c.
Figure 3 shows the 2–13 September 2006 ptot ratios (WRF/observed, area averaged over each of three regions shown in Fig. 2a) for each of the 64 WRF experiments, relative to GPCP (green bars) and TRMM (red bars). Figure 3a shows the ratios for the Sahel subregion, Fig. 3b for the rainband subregion, and Fig. 3c for the combined region. The x axis shows each WRF version from Table 2 along with the RM3 (labeled R3). The plots show that the WRF wet biases are common to all configurations and are slightly smaller in the comparisons to TRMM. Note that ratios for the rainband region (Fig. 3b) are similar to ratios for the combined regions (Fig. 3c), whereas the ratios for the Sahel region (Fig. 3a) are generally higher. The combined subregional domain represented in Fig. 3c (see Fig. 1) is chosen to monitor WRF precipitation performance in subsequent discussions.
Ratios of total precipitation between WRF and GPCP (red bar) and TRMM (green bar) for (a) the Sahel region, 10°–20°N, 20°W–10°E; (b) the rainband region, 5°–15°N, 20°W–10°E; and (c) both regions combined, 5°–20°N, 20°W–10°E.
Citation: Monthly Weather Review 145, 1; 10.1175/MWR-D-15-0294.1
A Taylor diagram provides a visual framework for a statistical summary of how well patterns match each other in terms of their correlation r, a normalized standard deviation σn normalized by dividing by the corresponding SREP σ value, and a normalized rmse-centered difference between each point concurrently in one plot (Taylor 2001). Figure 4 graphically summarizes the skill scores between the modeled and SREP ptot distributions between 5° and 20°N in a Taylor diagram. The radial distance from the origin measures the magnitude of each σn, normalized by dividing each WRF σ by the corresponding TRMM σ value. The normalized rmse-centered difference between each point and TRMM is measured by the values of the dashed semicircles. The spatial r, WRF versus TRMM, is given by the azimuthal position of each point, labeled along the outer arc. Figure 4b makes the same comparison against GPCP ptot. Figures 4a and 4b include the statistical scores achieved by each of the 64 WRF configurations, and secondary reference points of the other SREP, MERRA and NCEP2, and RM3. The positions of the WRF default and the five WRF configurations from Fig. 2 are labeled. Figure 4a shows that the majority of the WRF ptot distributions achieve r scores between 0.25 and 0.45 versus TRMM, while Fig. 4b shows that they get somewhat higher r scores versus GPCP (r = 0.40–0.70). In fact, WRF64 ptot (also shown in Fig. 2g) scores the highest r versus GPCP (0.70). RM3 outperforms all the WRF configurations by achieving the most favorable r and σn versus TRMM and GPCP (r = 0.78 and 0.90 and σn =1.10 and 1.13, respectively), consistent with the ptot shown in Fig. 2i.
Taylor diagrams displaying validation scores of 64 WRF simulations of 12-day precipitation accumulations over the area outlined in Fig. 1. The corresponding scores of other datasets and of the RM3 simulation are also shown. Selected WRF configurations are identified. Spatial correlation coefficients are designated along radial spokes while values of concentric arcs marked along the x and y axes indicate σ, normalized by the reference data standard deviation, thus σn. The dashed semicircles measure the normalized rmse-centered differences between each point and the reference data against the (a) TRMM reference data and (b) GPCP reference data.
Citation: Monthly Weather Review 145, 1; 10.1175/MWR-D-15-0294.1
b. Simulation of meridional moisture flux
Modeled moisture advection in the lower troposphere is crucial because the flux of summertime moisture into West Africa regulates precipitation rates. Druyan et al. (2010b) analyze the strength of the 950-hPa meridional moisture advection (qυ) for June–September as a diagnostic of model performance. Here, we consider qυ at the standard atmospheric level of 925 hPa as a lower troposphere diagnostic of the water cycle over West Africa during the period of this study. Figure 5a shows MERRA 925-hPa mean vector winds (V925) for 2–13 September 2006, superimposed over the 12-day mean qυ. Figures 5b–e show four versions of the spatial distribution of WRF minus MERRA V925 superimposed over the WRF qυ bias for each configuration relative to MERRA.
(a) MERRA 925-hPa resultant vector winds (υ9) for 2–13 Sep 2006 and mean meridional moisture flux (qυ) (g kg−1) (m s−1). (b) WRF2 minus MERRA υ9 and qυ, (c) WRF27 minus MERRA υ9 and qυ, (d) WRF32 minus MERRA υ9 and qυ, and (e) WRF60 minus MERRA υ9 and qυ.
Citation: Monthly Weather Review 145, 1; 10.1175/MWR-D-15-0294.1
There is no objective way to determine whether the WRF configurations have collectively produced a more realistic representation of lower-tropospheric moisture advection than MERRA since it is also model dependent. WRF2, WRF32, and WRF60 show a reduction in qυ in the lee of the Guinean Highlands. The three nondefault WRF configurations in Figs. 5c–e reproduce the monsoonal circulation and qυ patterns similar to MERRA over the Atlantic Ocean and Gulf of Guinea. However, they include a sizeable positive qυ bias relative to reanalysis in the Guinean region near 5°–10°N and 10°W–10°E and over the southeastern North Atlantic. These differences indicate that WRF imports excess moisture into WA in the lower troposphere. The exaggerated northward and eastward moisture transports across the Guinean coast region are consistent with high precipitation accumulations discussed in section 3a and Fig. 2. Negative WRF qυ biases over the eastern Gulf of Guinea may contribute to the aforementioned negative precipitation biases near the observed Cameroon orographic precipitation maximum.
Further analysis shows that WRF-simulated q results validate well against the reanalysis while WRF-simulated υ results are too strong (in the area of positive qυ bias). All the WRF configurations (not shown) show similar V925 and qυ features to those presented in Figs. 5b–e. In summary, WRF produces enhanced near-surface circulation toward WA. Similar deficiencies in MM5 seasonal simulations are reported by Druyan et al. (2010b), attributed to edge effects from the western boundary of their model domain at 35°W, the same western domain boundary of the domain used in this study. The impact on WRF simulations of moving the western boundary westward is not explored here.
c. Daily accumulated precipitation
Time–longitude Hovmöller distributions (Martius et al. 2006; Hovmöller 1949) of daily precipitation (pdaily) for the boxed region in Fig. 1 are constructed for each of the 64 WRF experiments in Table 2 and then compared to the corresponding GPCP observations. The validation of the Hovmöller time–longitude distributions of simulated versus observed pdaily provides statistics that evaluate temporal and longitudinal variability. Figure 6 shows nine time–longitude Hovmöller diagrams of pdaily for 2–13 September 2006, in addition to three Hovmöller diagrams of 700-mb (1 mb = 1 hPa) relative vorticity (RV). The x axis indicates longitude, and the y axis indicates the elapsed time. The data represent averages over latitudes 5°–20°N, plotted against each longitude, 20°W–10°E. Figure 6a shows the GPCP Hovmöller. Figures 6b and 6c show TRMM and CMORPH; Figs. 6d–h show WRF27, WRF32, WRF59, WRF60, and WRF64, respectively; and Fig. 6i does the same for RM3. Figures 6d–h show each WRF configuration generating four westward-moving pdaily maxima tracks roughly corresponding to GPCP. The pdaily tracks are related to transient AEWs and therefore to the tracks on Hovmöller distributions of meridional wind and vorticity discussed in Part I. To observe if the parameterized convection has a tendency to occur in the areas of a 700-mb trough (Part I), time–longitude Hovmöller distributions of 700-mb RV are shown for NCEP2 (Fig. 6j), WRF27 (Fig. 6k), and WRF32 (Fig. 6l).
Time–longitude Hovmöller distributions of daily accumulated precipitation, pdaily, (mm day−1). Precipitation is plotted at 0000 UTC on the designated day (y axis), 2–13 Sep, and averaged over 5°–20°N: (a) GPCP, (b) TRMM, (c) CMORPH, (d) WRF27, (e) WRF32, (f) WRF59, (g) WRF60, (h) WRF64, and (i) RM. Time–longitude Hovmöller distributions of relative vorticity (m2 s−2 × 105) at 700 mb are shown for (j) NCEP2, (k) WRF27, and (l) WRF32.
Citation: Monthly Weather Review 145, 1; 10.1175/MWR-D-15-0294.1
The comparison of GPCP to NCEP2 RV tracks present in Fig. 6j shows that precipitation maxima occur in areas of significant RV (orange). The comparison to GPCP indicates exaggerated WRF pdaily maxima, several position displacements, and time lags. For example, Figs. 6d–h show each WRF configuration simulating tracks in rather straight lines, implying constant propagation speeds, unlike GPCP, TRMM, and CMOPRH, which show tracks jumping from one maximum to another. The largest noticeable discrepancy between the WRF models and SREP observations occurs with track 2 during 3–8 September. All of the WRF configurations miss the appearance of track 2 on 3–4 September 2006, because of a propagation time lag. Only WRF32 and WRF60 (Figs. 6e,g) show the track correctly reaching 20°W on 8 September. Perhaps this difference is related to the trajectories of two AEW circulation patterns merging during these days (Part I). The comparison of GPCP to NCEP2 RV tracks present in Fig. 6j shows that precipitation maxima occur in areas of substantial RV (orange). However, the RV track 2 in NCEP2 shows a trough occurring over 7°W on 5–6 September; the GPCP and the TRMM results show pdaily maxima occurring earlier. All the WRF simulations show pdaily maxima occurring within the 700-mb RV trough. Both WRF27 and WRF32 pdaily maxima (Figs. 6d,e) occur over 7°W on 5–6 September at the same time as substantial 700-mb RV (Figs. 6k,l) and then proceed to 10°–15°W on 7–8 September. WRF simulations in Figs. 6d–h simulate a distinct track 3, which is almost in spatiotemporal agreement with GPCP, TRMM, and CMORPH, except that the SREPs do not show pdaily maxima for 9 September at 3°W, whereas the WRF shows continuous pdaily maxima from 3° to 5°W for that day. Schwendike and Jones (2010) attribute the interruptions of the rainfall events to the growth and decay of MCSs ahead of the AEW. The MCSs underwent stages of decay and regeneration, formed into a mesoscale convective vortex that enhanced the AEW vorticity, and interacted with another vortex generated by the topography near the coast; the WRF simulations do not capture these complex interactions. These WRF model pdaily simulation errors help explain some of the low skill score described below. The WRF27 and WRF32 pdaily maxima for track 3 occur in the presence of modeled 700-mb RV troughs. Janiga and Thorncroft (2016) suggest that models need to appropriately partition between convective versus stratiform precipitation in order to successfully simulate spatial precipitation patterns organized by AEWs. For example, placing the peak precipitation mainly in the AEW troughs and underestimating rainfall ahead of troughs may indicate that dynamically forced ascent plays too great a role in the model (Crétat et al. 2012). Crétat et al. (2012) find that WRF can exaggerate the magnitude of AEW-related rainfall events and overestimate their longitudinal extent as a result of excessive coupling between wave activity and convection.
Although the RM3 barely captures the movement of tracks 1 and 2, it correctly simulates the positions of the other pdaily maxima in GPCP. These results confirm that the RM3 requires an adjustment period of several days for the model physics to accommodate the initial moisture distribution, before it shows skill in simulating daily variability of precipitation, as noted by Druyan et al. (2006). The concept of considering a similar adjustment period in the WRF simulations is further explored in section 3d.
Figure 7 graphically summarizes the skill scores of the modeled WRF Hovmöller pdaily distributions against GPCP in Taylor diagrams. Figures 7a–d not only allow us to evaluate model performance by characterizing the statistical relationship between modeled and observed pdaily, but they also help us investigate modeled pdaily sensitivity to alternative WRF components. The Taylor distribution of statistics is shown four times: in each panel a different WRF parameterization from Tables 1 and 2 is highlighted to examine its influence on the scores. Figure 7a does this for CPS, Fig. 7b does this for RAD, Fig. 6c does this for PBL, and Fig. 7d does this for LSM. First, the Taylor diagram shows that TRMM scores the best r (r = 0.86) and σn (σn = 0.98), implying that comparisons between the WRF configurations and GPCP should show some similarities to comparisons with TRMM. Next, the diagrams show that MERRA scores a significant r (r = 0.43) but low σn (σn = 0.43) and NCEP2 scores poorly (r = 0.05), illustrating its significant difference from the observational evidence of pdaily based on SREP. Third, RM3 scores the highest r among all the regional models (r = 0.60). WRF experiments validate with a range of scores, several experiments achieving statistically significant r between r = 0.30 and 0.40 and a σn close to unity. The five WRF configurations that score some of the most favorable σn (σn < 1.25) and with high r against GPCP are labeled in each diagram (WRF27, WRF32, WRF59, WRF 60, and WRF64). Changing the CPS parameterization from KF to GD (Fig. 7a) produces an unambiguous improvement in the scores of many WRF configurations, achieving most of the highest r and best σn results. This is the same conclusion reached in Part I for the validation of simulated vorticity. In addition, three out of the four best scorers use the Rt RAD rather than the CM (Fig. 7b). Otherwise, some high scoring configurations incorporate alternative parameterizations of RAD, PBL, and LSM, implying that their influence is ambiguous. However, no version with the MJ PBL (Fig. 7c) and neither the 5L nor NO LSM (Fig. 7d) is found in the five selected WRF configurations. In fact, all configurations with the MJ PBL and NO–5L LSM yield poor performance. Five WRF experiments are shown as promising and are notated in Figs. 7a–d. Note that all five of the WRF experiments share the same GD CPS in Fig. 6a and that they use paired selection of A2–MN PBL in Fig. 7c and PX–RU LSM in Fig. 7d.
Validation scores for the time–longitude Hovmöller distributions of pdaily (as in Fig. 5), for each of 64 WRF experiments compared to GPCP. In these Taylor diagrams, values along the radial spokes indicate r and values of concentric arcs indicate σ, normalized by GPCP standard deviations, thus σn. The dashed semicircles measure the normalized rmse-centered differences between each point and GPCP. Colors (see legend) differentiate between major WRF Model parameterizations used in each experiment (see Table 2): (a) CPS, (b) RAD, (c) PBL, and (d) LSM. The best WRF scores are highlighted with their version number listed in Table 2. (NCEP2 reanalysis is labeled as R2.)
Citation: Monthly Weather Review 145, 1; 10.1175/MWR-D-15-0294.1
To investigate the causes of the poor performance scores, Fig. 8 examines the daily changes in validation scores of five WRF configurations selected as promising from Fig. 7 and the RM3. It shows daily cumulative r, validating modeled pdaily up to and including the day indicated, versus GPCP and TRMM, for WRF27, WRF32, WRF59, WRF60, WRF64, and RM3. It also shows daily r, validating the longitudinal variability of 5°–20°N pdaily means for each day for the same WRF configurations versus GPCP and TRMM. The daily r specific to each day shows whether the simulation for an individual day was skillful or not, which cannot be deduced from the cumulative r. Each WRF configuration panel in Fig. 8 shows a high cumulative r at initialization that drops rapidly during the first 2 days and then proceeds with a rather low cumulative r for the rest of the simulation. The daily r trends show that the low cumulative r scores occur for two reasons. First, the SREPs indicate pdaily maxima between 7°and 13°W on 4 September 2006 (Figs. 8a–c) while the WRF configurations miss this event. Second, the SREPs indicate a disruption of pdaily associated with AEW3 on 9 September 2006 near 3°W, while the WRF models show more continuity. The daily r results also show that the WRF configurations get relatively higher scores against GPCP. WRF Model simulation skill of pdaily is strongly influenced by the temporal error. The peaks and the minima for each WRF configuration do occur on the same day, indicating that the configurations have similar performance characteristics. Significant, positive daily r results imply that the WRF configurations use information from the LBCs to create precipitation events with similar timing and trajectories as SREPs. However, the performance of WRF may be sensitive to the synoptic-scale configuration of a particular day and its representation by the forcing data, in addition to model physics. Figure 8f shows that for RM3 versus GPCP (or TRMM), the cumulative r is higher than for any of the WRF models, and that the scores improve after the fifth day into the simulation, confirming RM3 model skill improvement after a 5-day spinup.
Time series showing two aspects of correlation r of the WRF modeled pdaily against the GPCP and TRMM pdaily results valid at 0000 UTC. The cumulative r between the models and GPCP (solid blue line) and TRMM (solid green line) validates results up to and including the day indicated. The daily r between the models and GPCP (dashed red line) and TRMM (dashed orange line) validates results specific to that day.
Citation: Monthly Weather Review 145, 1; 10.1175/MWR-D-15-0294.1
d. Spinup and multiyear validation
Druyan et al. (2006) report that RM3 pdaily simulations undergo an initial 5-day adjustment period before results compare well with TRMM-observed variability. The benefit of a spinup period is clearly visible in the RM3 Hovmöller diagram of pdaily in Fig. 6b and in the correlation scores in Fig. 8f. The five WRF configurations, WRF27, WRF32, WRF59, WRF60, and WRF64, are additionally integrated from initial conditions on 27 August and validated against GPCP for the 2–13 September study period. The 6-day spinup period does not, however, improve their precipitation validation scores (not shown).
The WRF27, WRF32, WRF59, WRF60, and WRF64 simulations are repeated with NCEP2 boundary conditions for the remaining years, 2000–2010. Figure 9 shows cumulative r results versus elapsed time for the Hovmöller pdaily distributions for simulations of all 11 years. Figure 9a compares TRMM against GPCP and Figs. 9b–f show the same for the selected WRF configurations. CMORPH observations are excluded here because they only start in 2003. Figure 9a shows that the TRMM and GPCP simulations are similar, although there is slight disagreement during the first two days. The largest drop from high to low correlations occurs during the first 48 h in each WRF simulation, which indicates that NCEP2 forcing data do not greatly influence the precipitation produced in the simulations. The WRF configurations show that 2006 (red-filled circles) does not perform well, although it gives significant scores from 10 to 13 September. Indeed, the simulations for 2005 and 2008 give the best r scores and the simulations for 2001, 2002, and 2004 give the worst performing scores. Figure 9 shows that the validation of 2005 maintains high correlations throughout the period for all WRF configurations. The interannual differences in correlation scores are greater than the differences between configurations, suggesting that the precipitation variability during some years is more difficult to simulate than during others. These results differ slightly from those of the prequel study (Part I), which showed that 2006 maintained high correlations throughout the 12-day period and 2005 achieved the best scores for 700-hPa vorticity when validated against the reanalysis. The similar correlation scores between the precipitation results of this study and the vorticity results of Part I support the assumption that skillful simulations of daily precipitation over West Africa require skillful simulations of AEW-related vorticity centers.
An 11-yr comparison of correlations (cumulative r) vs elapsed time for modeled Hovmöller pdaily distributions vs GPCP for 2–13 Sep 2000–10 valid at 0000 UTC: (a) TRMM (b) WRF27, (c) WRF32, (d) WRF59, (e) WRF60, and (f) WRF64.
Citation: Monthly Weather Review 145, 1; 10.1175/MWR-D-15-0294.1
e. Impacts of alternative microphysics
The foregoing evaluation of WRF performance combines the roster of parameterizations listed in Table 2 with the WRF single-moment 5-class MPS scheme. Impacts of using alternative MPS schemes are now tested by repeating the 12-day simulations of versions WRF32 and WRF27 with each of two alternative microphysics schemes: the new Thompson graupel (Thompson 2013) and the Goddard cumulus ensemble (Wolff 2011; Harrold 2012). Berry and Thorncroft (2012) also consider these schemes but favor the new Thompson MPS scheme after testing 32 WRF configurations for AEW simulation. Here, the horizontal distribution of 12-day precipitation accumulations is compared to GPCP data and validation statistics are compared with results based on using the WSM5. Results for 2006, listed in Table 4 (top), show that substituting the new Thompson MPS in place of the WSM5 in WRF32 increases the spatial correlation of simulated precipitation with GPCP-observed 12-day accumulations from 0.54 (Fig. 2) to 0.61, and the correlation with the GPCP observed time–longitude Hovmöller pdaily distribution of precipitation (as in Fig. 6) from 0.32 to 0.40 [Table 4 (bottom)]. There are also concomitant improvements in the variance and bias. The Goddard scheme did not improve the performance of WRF32, and neither MPS improved WRF27. This suggests that the new Thompson graupel MPS has some advantages over the WSM5 for selected WRF versions. However, the apparent advantage of using the Thompson graupel MPS for the 2006 Hovmöller pdaily distribution was not evident in the simulations for the other years (see Fig. 9).
Correlation, RMSE, and the ratio of standard deviation (Ratio std dev) scores from the comparison of WRF modeled precipitation estimates to NASA GPCP estimates. The WRF models utilize the new Thompson graupel (Th) and the Goddard cumulus ensemble (Gd) microphysics. Shown are (top) 12-day accumulations and (bottom) time–longitude Hovmöller distribution of pdaily precipitation.
4. Discussion and conclusions
More than 100 twelve-day WRF simulations, run on a 20-km grid over West Africa in September during 11 consecutive years, are evaluated to examine the sensitivity of WRF daily precipitation simulations over West Africa to model configuration. This study focuses on the simulation of 12-day accumulations and daily precipitation over West Africa during September, where much of the precipitation variability is linked to transient AEWs. Evaluation of the corresponding circulation, mean and daily, is reported in a prequel paper (Part I). Some 64 WRF configurations are tried by using different combinations of available WRF parameterizations: CPS, LSM, PBL, and RAD physics. Simulated precipitation data are compared to NASA’s GPCP and TRMM satellite precipitation estimates, which provide independent estimates of actual conditions at a relatively high horizontal resolution (Δx, Δy = 0.25°).
The WRF Model simulates the summer monsoon precipitation maximum over West Africa, but with somewhat higher values than observed. Precipitation simulations are analyzed for 2–13 September 2006, a period that coincides with the NAMMA field campaign, where four major precipitation events associated with transient AEWs occur within the zonal belt 5°–20°N. Six WRF configurations, including five of the better-performing configurations, produce 12-day accumulations that range from 110% to 186% (105%–180%) of the GPCP (TRMM) values, and show the south-to-north gradients from rainy to dry conditions over the Sahel. The most excessive accumulation is simulated by the WRF Model’s default configuration (2014), which is 180% (175%) of the GPCP (TRMM) values. Excessive WRF accumulations are consistent with the WRF’s exaggerated northward and eastward moisture flux across the West African and Guinean coast regions in these simulations. The best WRF performers in this analysis are WRF27, WRF32, and WRF64, which all share GD CPS. WRF32, and WRF64 share the MN PBL and the PX surface model but differ in their radiation parameterizations. Compared to the performance of other regional models over West Africa though, the discrepancies in the mean state are serious deficiencies. Excessive precipitation accumulations are consistent with the WRF’s exaggerated northward and eastward moisture transports across the West African and Guinean coast regions in these simulations.
Propagation of rain maxima associated with the four westward-propagating AEWs appears as diagonal swaths in time–longitude Hovmöller charts of daily precipitation. Corresponding tracks of the 700-hPa meridional wind and vorticity on Hovmöller distributions are discussed in Part I. Time–longitude Hovmöller distributions of daily accumulated precipitation from WRF were objectively validated against GPCP, TRMM, and CMORPH. The WRF-generated tracks favor rather constant propagation speeds whereas the observations show irregular swaths. Alternative parameterizations influence the simulation of precipitation maxima and tracks, which is borne out by the range of validation scores for the 64 WRF configurations summarized in the Taylor diagrams. The best WRF performers achieve correlations (against GPCP) of between 0.35 and 0.42 and spatiotemporal variability amplitudes only slightly higher than SREP. A parallel simulation by the benchmark RM3 achieves a higher correlation (r = 0.52) and realistic spatiotemporal variability amplitudes against GPCP. It should be noted that none of the RCMs show the propagating precipitation maxima shifting with the formation of MCSs and subsequent regeneration of convection along a new phase line in the Hovmöller diagram that eventually became Tropical Cyclone Helene (Schwendike and Jones 2010).The WRF simulations in this study were unable to capture those complex interactions, which are best attempted using higher-resolution models with explicit convection.
The largest favorable impact on WRF precipitation simulation is realized by selecting the GD CPS. Part I reports a similar success for the simulation of 700-hPa vorticity variability. This preference may reflect the more versatile skill of that scheme’s ensemble approach to computing the two-way interaction between convection and large-scale circulation (Grell and Dévényi 2002). Simulations using this scheme achieve higher correlations with SREP than do the remaining simulations using the KF CP (Kain 2004), and GD also simulates less extreme daily precipitation maxima than KF CPS (not shown). As in the current study, Berry (2009) finds that using GD CPS in WRF produces the most realistic representation of the precipitation field and the most realistic representation of the westward-propagating AEWs that cross the African coast. Rather than using a single closure criterion to determine how the cumulus parameterization scheme feeds back on the model atmosphere, GD employs the ensemble mean of multiple closure schemes using CAPE removal, destabilization effects, moisture convergence, and low-level vertical velocity. The trigger is a measure of buoyancy that can be modified by the low-level moisture convergence or active convection. The closure schemes are also perturbed by variations of sensitive parameters related to the strength of the convection and feedback on the model atmosphere. The current study and that of Berry (2009) suggest that this ensemble approach is particularly suited to the complex summer WAM conditions that generate precipitation within MCSs and AEWs, and along the intertropical discontinuity rainband. Berry (2009) suggests that the presence of dry Sahara air overlying the moist monsoon layer keeps mean layer humidity below the triggering threshold used by the Betts–Miller–Janjić scheme, which requires a moist vertical profile. Moreover, in Berry’s experiments, KF produces broad areas of light precipitation for propagating systems instead of the observed narrow swaths of substantial rainfall, which may be a result of deep saturated layers in postconvective soundings that produce unrealistic stratiform rainfall, and GD produces the most realistic propagating precipitation maxima and the most realistic orographic maximums.
Impacts of other model parameterizations are more ambiguous. More consistently favorable results are obtained using the RU and PX LSMs. Configurations incorporating the NO LSM did not perform well, but there are no unambiguous impacts attributed to using the remaining RAD, PBL, and LSM schemes.
Examination of the cumulative and noncumulative daily correlation scores versus elapsed time for the selected WRF configurations for 2006 shows some WRF skill in simulating precipitation during the first two day and for 9–11 September, while the rest of the daily scores were more disappointing. Our comparison of each SREP to data from 20 NAMMA rain gauge stations confirms that GPCP and TRMM are consistent with ground truth.
The WRF limitations for simulating daily meteorological fields do not preclude the model producing realistic seasonal mean climate fields, especially if many of the high-frequency errors are random. While 64 model configurations cover a wide range, the testing reported here is not exhaustive. Results in this study favor WRF27, WRF32, WRF60, and WRF64 as the model configurations to consider; yet WRF27 (see Table 2) is the configuration that achieves the best overall performance. This is sustained by the circulation validation reported in a prequel paper (Part I), which describes the evaluation of WRF–AEW circulation variability for the same 64 configurations. Future work should explore the performance of new WRF parameterizations for simulating West African weather and climate.
Acknowledgments
The authors gratefully acknowledge the inspiration and encouragement for this project of the late Professor Thomas Warner. EUN was supported by NASA Cooperative Agreement NNX11AR61G. LMD and MF were supported by National Science Foundation Grant AGS-1000874 and NASA Cooperative Agreement NNX11AR63A. MERRA data were obtained from NASA’s GMAO web site (http://gmao.gsfc.nasa.gov/merra), NCEP Reanalysis-2 data were provided by the NOAA-ESRL Physical Sciences Division, Boulder, Colorado, from their website (http://www.esrl.noaa.gov/psd).
REFERENCES
Behrangi, A., K.-L. Hsu, B. Imam, S. Sorooshian, G. J. Huffman, and R. J. Kuligowski, 2009: PERSIANN-MSA: A precipitation estimation method from satellite-based multispectral analysis. J. Hydrometeor., 10, 1414–1429, doi:10.1175/2009JHM1139.1.
Benjamin, S. G., and Coauthors, 2002: RUC20—The 20-km version of the Rapid Update Cycle. NWS Tech. Procedures Bull. 490, 30 pp. [Available online at http://ruc.noaa.gov/ppt_pres/RUC20-tpb.pdf.]
Bernardet, L., and Coauthors, 2008: The Developmental Testbed Center and its winter forecasting experiment. Bull. Amer. Meteor. Soc., 89, 611–627, doi:10.1175/BAMS-89-5-611.
Berry, G. J., 2009: African easterly waves and convection. Ph.D thesis, State University of New York at Albany, 215 pp.
Berry, G. J., and C. D. Thorncroft, 2012: African easterly wave dynamics in a mesoscale numerical model: The upscale role of convection. J. Atmos. Sci., 69, 1267–1283, doi:10.1175/JAS-D-11-099.1.
Burpee, R. W., 1972: The origin and structure of easterly waves in the lower troposphere of North Africa. J. Atmos. Sci., 29, 77–90, doi:10.1175/1520-0469(1972)029<0077:TOASOE>2.0.CO;2.
Caldwell, P., H.-N. S. Chin, D. C. Bader, and G. Bala, 2009: Evaluation of a WRF dynamical downscaling simulation over California. Climatic Change, 95, 499–521, doi:10.1007/s10584-009-9583-5.
Chen, F., and J. Dudhia, 2001: Coupling an advanced land surface–hydrology model with the Penn State–NCAR MM5 modeling system. Part II: Preliminary model validation. Mon. Wea. Rev., 129, 587–604, doi:10.1175/1520-0493(2001)129<0587:CAALSH>2.0.CO;2.
Chiao, S., and G. S. Jenkins, 2010: Numerical investigations on the formation of Tropical Storm Debby during NAMMA-06. Wea. Forecasting, 25, 866–884, doi:10.1175/2010WAF2222313.1.
Cifelli, R., T. Lang, S. A. Rutledge, and N. Guy, 2010: Characteristics of an African easterly wave observed during NAMMA. J. Atmos. Sci., 67, 3–25, doi:10.1175/2009JAS3141.1.
Collins, W. D., P. J. Rasch, B. A. Boville, J. J. Hack, J. R. McCaa, D. L. Williamson, J. T. Kiehl, and B. Briegleb, 2004: Description of the NCAR Community Atmosphere Model (CAM 3.0). NCAR Tech. Rep. NCAR/TN-4641 STR, 214 pp. [Available online at http://www.cesm.ucar.edu/models/atm-cam/docs/description/description.pdf.]
Cook, K. H., and E. K. Vizy, 2009: Tropical storm development from African easterly waves in the eastern Atlantic: A comparison of two successive waves using a regional model as part of NASA AMMA 2006. J. Atmos. Sci., 66, 3313–3334, doi:10.1175/2009JAS3064.1.
Crétat, J., B. Pohl, Y. Richard, and P. Drobinski, 2012: Uncertainties in simulating regional climate of southern Africa: Sensitivity to physical parameterizations using WRF. Climate Dyn., 38, 613–634, doi:10.1007/s00382-011-1055-8.
Diedhiou, A., S. Janicot, A. Viltard, P. de Felice, and H. Laurent, 1999: Easterly wave regimes and associated convection over West Africa and tropical Atlantic: Results from the NCEP/NCAR and ECMWF reanalyses. Climate Dyn., 15, 795–822, doi:10.1007/s003820050316.
Druyan, L. M., P. Lonergan, and M. Saloum, 1996: African wave disturbances and precipitation at Niamey during July–August 1987 and 1988. Climate Res., 7, 71–83, doi:10.3354/cr007071.
Druyan, L. M., M. Fulakeza, and P. Lonergan, 2006: Mesoscale analyses of West African summer climate: Focus on wave disturbances. Climate Dyn., 27, 459–481, doi:10.1007/s00382-006-0141-9.
Druyan, L. M., M. Fulakeza, and P. Lonergan, 2008: The impact of vertical resolution on regional model simulation of the West African summer monsoon. Int. J. Climatol., 28, 1293–1314, doi:10.1002/joc.1636.
Druyan, L. M., M. Fulakeza, P. Lonergan, and E. Noble, 2009: Regional climate model simulation of the AMMA Special Observing Period #3 and the pre-Helene easterly wave. Meteor. Atmos. Phys., 105, 191–210, doi:10.1007/s00703-009-0044-5.
Druyan, L. M., M. Fulakeza, P. Lonergan, and R. Worrell, 2010a: Regional model nesting within GFS daily forecasts over West Africa. Open Atmos. Sci. J., 4, 1–11.
Druyan, L. M., and Coauthors, 2010b: The WAMME regional model intercomparison study. Climate Dyn., 35, 175–192, doi:10.1007/s00382-009-0676-7.
Dudhia, J., 2014: A history of mesoscale model development. Asia-Pac. J. Atmos. Sci., 50, 121–131, doi:10.1007/s13143-014-0031-8.
Fink, A. H., and A. Reiner, 2003: Spatiotemporal variability of the relation between African easterly waves and West African squall lines in 1998 and 1999. J. Geophys. Res., 108, 4332, doi:10.1029/2002JD002816.
Flaounas, E., S. Bastin, and S. Janicot, 2011: Regional climate modelling of the 2006 West African monsoon: Sensitivity to convection and planetary boundary layer parameterization using WRF. Climate Dyn., 36, 1083–1105, doi:10.1007/s00382-010-0785-3.
Flaounas, E., S. Janicot, S. Bastin, and R. Roca, 2012: The West African monsoon onset in 2006: Sensitivity to surface albedo, orography, SST and synoptic scale dry-air intrusions using WRF. Climate Dyn., 38, 685–708, doi:10.1007/s00382-011-1255-2.
Fortune, M., 1980: Properties of African squall lines inferred from time-lapse satellite imagery. Mon. Wea. Rev., 108, 153–168, doi:10.1175/1520-0493(1980)108<0153:POASLI>2.0.CO;2.
Franklin, J. L., and D. P. Brown, 2008: Atlantic hurricane season of 2006. Mon. Wea. Rev., 136, 1174–1200, doi:10.1175/2007MWR2377.1.
Giorgi, F., and L. O. Mearns, 1991: Approaches to the simulation of regional climate change: A review. Rev. Geophys., 29, 191–216, doi:10.1029/90RG02636.
Grell, G. A., and D. Dévényi, 2002: A generalized approach to parameterizing convection combining ensemble and data assimilation techniques. Geophys. Res. Lett., 29, doi:10.1029/2002GL015311.
Guy, N., S. A. Rutledge, and R. Cifelli, 2011: Radar characteristics of continental, coastal, and maritime convection observed during AMMA/NAMMA. Quart. J. Roy. Meteor. Soc., 137, 1241–1256, doi:10.1002/qj.839.
Harrold, M., 2012: The Developmental Testbed Center’s final report on the inter-comparison of the WRFv3.3.1. AFWA operational and RRTMG-replacement configurations. DTC, Boulder, CO, 40 pp. [Available online at http://www.dtcenter.org/eval/meso_mod/afwa_test/wrf_v3.3.1/WRFv3.3.1_AFWA_RRTMG_final_report.pdf.]
Hong, S.-Y., J. Dudhia, and S.-H. Chen, 2004: A revised approach to ice microphysical processes for the bulk parameterization of clouds and precipitation. Mon. Wea. Rev., 132, 103–120, doi:10.1175/1520-0493(2004)132<0103:ARATIM>2.0.CO;2.
Hong, S.-Y., Y. Noh, and J. Dudhia, 2006: A new vertical diffusion package with an explicit treatment of entrainment processes. Mon. Wea. Rev., 134, 2318–2341, doi:10.1175/MWR3199.1.
Hovmöller, E., 1949: The trough-and-ridge diagram. Tellus, 1, 62–66, doi:10.1111/j.2153-3490.1949.tb01260.x.
Huffman, G. J., and Coauthors, 2007: The TRMM Multisatellite Precipitation Analysis (TMPA): Quasi-global, multiyear, combined-sensor precipitation estimates at fine scales. J. Hydrometeor., 8, 38–55, doi:10.1175/JHM560.1.
Huffman, G. J., R. F. Adler, D. T. Bolvin, and G. Gu, 2009: Improving the global precipitation record: GPCP version 2.1. Geophys. Res. Lett., 36, L17808, doi:10.1029/2009GL040000.
Iacono, M. J., J. S. Delamere, E. J. Mlawer, M. W. Shephard, S. A. Clough, and W. D. Collins, 2008: Radiative forcing by long-lived greenhouse gases: Calculations with the AER radiative transfer models. J. Geophys. Res., 113, D13103, doi:10.1029/2008JD009944.
Janiga, M. A., and C. D. Thorncroft, 2016: The influence of African easterly waves on convection over tropical Africa and the East Atlantic. Mon. Wea. Rev., 144, 171–192, doi:10.1175/MWR-D-14-00419.1.
Janjić, Z. I., 2002: Nonsingular implementation of the Mellor–Yamada level 2.5 scheme in the NCEP Meso model. NCEP Office Note 437, 61 pp. [Available online at http://www.emc.ncep.noaa.gov/officenotes/newernotes/on437.pdf.]
Jones, P. W., 1999: First- and second-order conservative remapping schemes for grids in spherical coordinates. Mon. Wea. Rev., 127, 2204–2210, doi:10.1175/1520-0493(1999)127<2204:FASOCR>2.0.CO;2.
Joyce, R., J. Janowiak, A. Phillip, and P. Xie, 2004: CMORPH: A method that produces global precipitation estimates from passive microwave and infrared data at high spatial and temporal resolution. J. Hydrometeor., 5, 487–503, doi:10.1175/1525-7541(2004)005<0487:CAMTPG>2.0.CO;2.
Kain, J. S., 2004: The Kain–Fritsch convective parameterization: An update. J. Appl. Meteor., 43, 170–181, doi:10.1175/1520-0450(2004)043<0170:TKCPAU>2.0.CO;2.
Kanamitsu, M., W. Ebisuzaki, J. Woollen, S.-K. Yang, J. J. Hnilo, M. Fiorino, and G. L. Potter, 2002: NCEP–DOE AMIP-II Reanalysis (R-2). Bull. Amer. Meteor. Soc., 83, 1631–1643, doi:10.1175/BAMS-83-11-1631.
Lebel, T., A. Diedhiou, and H. Laurent, 2003: Seasonal cycle and interannual variability of the Sahelian rainfall at hydrological scales. J. Geophys. Res., 108, 8389, doi:10.1029/2001JD001580.
Liang, X.-Z., and Coauthors, 2012: Regional Climate–Weather Research and Forecasting Model. Bull. Amer. Meteor. Soc., 93, 1363–1387, doi:10.1175/BAMS-D-11-00180.1.
Lim, J. O. J., and S.-Y. Hong, 2005: Effects of bulk ice microphysics on the simulated monsoonal precipitation over East Asia. J. Geophys. Res., 110, D24201, doi:10.1029/2005JD006166.
Lynn, B. H., R. Healy, and L. M. Druyan, 2009: Quantifying the sensitivity of simulated climate change to model configuration. Climatic Change, 92, 275–298, doi:10.1007/s10584-008-9494-x.
Martius, O., C. Schwirez, and H. C. Davies, 2006: A refined Hovmöller diagram. Tellus, 58A, 221–226, doi:10.1111/j.1600-0870.2006.00172.x.
Nakanishi, M., and H. Niino, 2006: An improved Mellor–Yamada level-3 model: Its numerical stability and application to a regional prediction of advection fog. Bound.-Layer Meteor., 119, 397–407, doi:10.1007/s10546-005-9030-8.
Nicholson, S. E., 2013: The West African Sahel: A review of recent studies on the rainfall regime and its interannual variability. ISRN Meteor., 2013, 453521, doi:10.1155/2013/453521.
Noble, E., L. M. Druyan, and M. Fulakeza, 2014: The sensitivity of WRF daily summertime simulations over West Africa to alternative parameterizations. Part I: African wave circulation. Mon. Wea. Rev., 142, 1588–1608, doi:10.1175/MWR-D-13-00194.1.
Payne, S. W., and M. M. McGarry, 1977: The relationship of satellite inferred convective activity to easterly waves over West Africa and the adjacent ocean during phase III of GATE. Mon. Wea. Rev., 105, 413–420, doi:10.1175/1520-0493(1977)1052.0.CO;2.
Pleim, J. E., 2007: A combined local and nonlocal closure model for the atmospheric boundary layer. Part I: Model description and testing. J. Appl. Meteor. Climatol., 46, 1383–1395, doi:10.1175/JAM2539.1.
Pleim, J. E., A. Xiu, and A. Xiu, 2003: Development of a land surface model. Part II: Data assimilation. J. Appl. Meteor. Climatol., 42, 1811–1822, doi:10.1175/1520-0450(2003)042<1811:DOALSM>2.0.CO;2.
Pohl, B., J. Crétat, and P. Camberlin, 2011: Testing WRF capability in simulating the atmospheric water cycle over equatorial East Africa. Climate Dyn., 37, 1357–1379, doi:10.1007/s00382-011-1024-2.
Redelsperger, J.-L., C. D. Thorncroft, A. Diedhiou, T. Lebel, D. J. Parker, and J. Polcher, 2006: African Monsoon Multidisciplinary Analysis: An international research project and field campaign. Bull. Amer. Meteor. Soc., 87, 1739–1746, doi:10.1175/BAMS-87-12-1739.
Reed, R. J., D. C. Norquist, and E. E. Recker, 1977: The structure and properties of African wave disturbances as observed during phase III of GATE. Mon. Wea. Rev., 105, 317–333, doi:10.1175/1520-0493(1977)105<0317:TSAPOA>2.0.CO;2.
Ruti, P. M., 2010: The twentieth century African easterly waves in reanalysis systems and IPCC simulations, from intra-seasonal to inter-annual variability. Climate Dyn., 35, 1099–1117, doi:10.1007/s00382-010-0894-z.
Schwendike, J., and S. C. Jones, 2010: Convection in an African easterly wave over West Africa and the eastern Atlantic: A model case study of Helene (2006). Quart. J. Roy. Meteor. Soc., 136, 364–396, doi:10.1002/qj.566.
Skamarock, W. C., and Coauthors, 2008: A description of the Advanced Research WRF version 3. NCAR Tech. Note NCAR/TN-475+STR, 113 pp., doi:10.5065/D68S4MVH.
Sultan, B., and S. Janicot, 2000: Abrupt shift of the ITCZ over West Africa and intra-seasonal variability. Geophys. Res. Lett., 27, 3353–3356, doi:10.1029/1999GL011285.
Taleb, E. H., and L. M. Druyan, 2003: Relationships between rainfall and West African wave disturbances in station observations. Int. J. Climatol., 23, 305–313, doi:10.1002/joc.883.
Taylor, K. E., 2001: Summarizing multiple aspects of model performance in a single diagram. J. Geophys. Res., 106, 7183–7192, doi:10.1029/2000JD900719.
Thompson, G., 2013: High-resolution winter simulations of winter precipitation over the Colorado Rockies. Proc. ECMWF Workshop on Parametrization of Clouds and Precipitation, Reading, United Kingdon, ECMWF, 46 pp. [Available online at http://www.ecmwf.int/en/elibrary/12672-high-resolution-winter-simulations-winter-precipitation-over-colorado-rockies.]
Thorncroft, C. D., D. J. Parker, R. R. Burton, and M. Diop, 2003: The JET2000 project: Aircraft observations of the African easterly jet and African easterly waves. Bull. Amer. Meteor. Soc., 84, 337–351, doi:10.1175/BAMS-84-3-337.
Vaidya, S. S., and S. S. Singh, 2000: Applying the Betts–Miller–Janjic scheme of convection in prediction of the Indian monsoon. Wea. Forecasting, 15, 349–356, doi:10.1175/1520-0434(2000)015<0349:ATBMJS>2.0.CO;2.
Wolff, J., 2011: The Developmental Testbed Center WRF reference configuration implementation plan. DTC, Boulder, CO, 5 pp. [Available online at www.dtcenter.org/config/RC_Implementation_Plan.pdf.]
Xue, Y., and J. Shukla, 1993: The influence of land surface properties on Sahel climate. Part I: Desertification. J. Climate, 6, 2232–2246, doi:10.1175/1520-0442(1993)006<2232:TIOLSP>2.0.CO;2.
Zipser, E. J., and Coauthors, 2009: The Saharan air layer and the fate of African easterly waves—NASA’s AMMA field study of tropical cyclogenesis. Bull. Amer. Meteor. Soc., 90, 1137–1156, doi:10.1175/2009BAMS2728.1.