1 Introduction

From weather forecasting perspective to radiative energy balance standpoint, gathering accurate information on cloud properties is important on a day-to-day basis. Clouds are also principal modulators of the earth’s hydrological cycle connecting a critical link between precipitation efficiency and radiative budget of the climate system. They also play a crucial role in the formation of severe storms by condensing water vapor from warm ocean to form large cumulonimbus clouds.

The cloud properties such as thermodynamic phase of clouds (e.g., liquid/ice) and their quantities can generally be measured by ground-based or satellite-borne remote-sensing instruments. In particular, good efforts have been made to derive cloud quantities from the measurements on-board various satellite platforms. Both passive and active remote-sensing techniques have been successful in providing a good insight into the distribution of clouds over both ocean and land surfaces. Some of the prior studies related to satellite observation of clouds are documented in Otkin and Greenwald (2008), Minnis et al. (2011), Islam et al. (2014a), and Islam et al. (2014d), and can be referred therein.

Nevertheless, satellite data are scarce and have temporal and spatial resolution limitations. On the other hand, cloud quantities change rapidly and vary with respect to time and space. Therefore, satellite-based estimate of cloud quantities cannot provide a full picture of cloud information of our entire globe. The mesoscale numerical weather prediction (NWP) models can complement the space-based technologies for a variety of atmospheric and surface observations, including clouds. With the advancements in computing powers, it is now possible to simulate weather events in very high resolution, and as such, a wide range of NWP studies is comprehensively documented in the literature (Dai et al. 2013; Ishak et al. 2013; Islam et al. 2013; Srivastava et al. 2013; Islam et al. 2014b; Srivastava et al. 2014a; Srivastava et al. 2014b). However, the NWP models have their own limitations itself, for instances, the limitations associated with the physics parameterizations used. In order to accurately simulate the cloud morphology in mesoscale models, it is essential to take into account of all complex interactions between cloud species with precise particle size distribution. However, in reality, bulk microphysics schemes are used in mesoscale models with some approximations of particle size distribution. Besides, simulation errors could also be introduced from model initialization, lateral boundary conditions, model resolution, and other physics parameterizations such as planetary boundary layer (PBL) and cumulus schemes.

A tropical cyclone has recently been hit Philippines in November 2013 and has caused a catastrophic damage. The cyclone, named as Haiyan, has had minimum central pressure of 895 hPa, making it one of the strongest typhoons in the history (super typhoon). In the present study, a mesoscale model, known as the Weather Research and Forecasting (WRF), has been used in simulating the clouds over the super typhoon Haiyan. The goal is to evaluate the effect of physics parameterizations on cloud simulation and if they differ with each other. As part of the assessment, the simulated cloud phase (e.g., liquid or ice) and important cloud quantities such as ice water path are compared against MODIS optical cloud information. It is worth mentioning that the simulated track and intensity of the typhoon have been extensively validated in our previous study with the best track data (Islam et al. 2014c). Based on our previous study, it is found that the WRF simulation has been successful in predicting the track propagation of the typhoon. Nevertheless, the model has substantially underestimated the intensity prediction. In this study, we are rather interested in investigating the capability of the WRF model to realistically simulate cloud properties.

The remainder of this paper is structured as follows: The datasets, WRF model configuration, and simulation setup are outlined in Sect. 2. Section 3 provides the results and discussions. The paper is concluded in Sect. 4.

2 Data and simulation setup

2.1 WRF simulation

The Weather Research and Forecasting (WRF) model used in this experiment is the Advanced Research WRF (ARW) dynamic solver, developed by NCAR/MMM (Mesoscale and Microscale Meteorology Division). The equation set for ARW is Eulerian Mass dynamical core with terrain following vertical coordinates. For a comprehensive description of the WRF-ARW dynamic solver, referring the article of Skamarock and Klemp (2008) is recommended to the readers.

In this study, the WRF model is configured to simulate super typhoon Haiyan with two domains, as shown in Fig. 1. Two-way nested run option has been allowed in the simulation between the mother and child domains. The domain 1 (coarse domain) is comprised of 169 × 135 grid points, 25-km grid size, and 28 vertical levels, covering a size of 4225 × 3375-km2 area. The domain 2 is configured using high-resolution grid spacing (5 km), such that small-scale feature of the system can be captured by the model. It covers the area of 2180 × 2155 km2, and also uses 28 vertical levels. Note that vertical grids are variable and stretched.

Fig. 1
figure 1

Map showing the WRF domain settings for the super typhoon Haiyan simulation

Full size image

The model topography and other static geo-fields are initialized from very high-resolution global datasets (http://www2.mmm.ucar.edu/wrf). The simulation is performed without assimilating any observation data. The ERA-Interim analyses produced by the ECMWF global forecast model are used for model initialization and lateral boundary conditions. The simulation is started at 0000 UTC 6 November 2013 and run for 48 h. By this time, the system has gained the strength of category 5 equivalent tropical storm.

A variety of physics combinations have been investigated to assess the simulation of cloud properties. Summarizing, a total of 10 simulation experiments have been carried out by employing different physics parameterizations, as tabulated in Table 1. The investigated experiments include a combination of different microphysics and PBL parameterization schemes, including GGCE (Tao et al. 1989), MY2 (Milbrandt and Yau 2005a, b), THOM (Thompson et al. 2008), WSM3 (Hong et al. 2004), WSM6 (Hong and Lim 2006) microphysics schemes, and YSU (Hong et al. 2006), ACM2 (Pleim 2007), BouLac (Bougeault and Lacarrere 1989), MYNN (Nakanishi and Niino 2006), and QNSE (Sukoriansky et al. 2005) PBL schemes. Each simulation uses Kain–Fritsch (KF) cumulus parameterization scheme (Kain 2004), the rapid radiative transfer model (RRTM) longwave radiation scheme, Dudhia shortwave radiation scheme, MM5 similarity surface layer, and Noah land surface model (Chen and Dudhia 2001). Additionally, a simulation with no cumulus parameterization scheme (THOM-YSU-NCu) is included for comparison with the other simulations.

Table 1 Physics options used for different simulation experiments
Full size table

In the latter section (Sect. 3), we will be comparing the results between different simulations. Furthermore, the simulated cloud properties and cloud quantities will be compared against MODIS-derived outputs.

2.2 MODIS data

The moderate-resolution imaging spectroradiometer (MODIS) is a payload instrument on board the Terra and Aqua satellites. It acquires data in 36 spectral bands ranging from 0.4- to 14.4-μm wavelength at varying spatial resolutions. In the present study, the MODIS cloud datasets are acquired from MOD06_L2 atmospheric product. A variety of cloud properties including cloud-particle phase (ice vs. water, clouds vs. snow), effective cloud particle radius (r e ), and cloud optical thickness (τ) are stored in the MOD06_L2 product at 1-km pixel resolution. They are derived using the MODIS visible, near-infrared, and shortwave infrared channel radiances. Cloud water path is then indirectly converted from r e and τ information. The cloud water path is defined as

$$ CWP=\frac{4\rho \tau {r}_e}{3{Q}_e} $$
(1)

where CWP can either be liquid water path (LWP) or ice water path (IWP) depending upon the cloud top phase, Q e is the extinction efficiency, and ρ is the density of water or ice.

Some other cloud properties such as cloud top temperature and cloud top pressure are produced at 5 × 5 1-km pixel resolution. A complete description of the MODIS cloud algorithms can be found in Platnick et al. (2003).

During our WRF simulation run, a good MODIS overpass scene has been captured by NASA’s Terra satellite at 0120 UTC 07 November 2013 over the super typhoon Haiyan. It passed directly over the typhoon region and provided excellent coverage of the model domain. This gives us opportunity to compare the WRF simulated fields against MODIS-derived cloud properties on this particular match-up scene.

Figure 2 highlights some cloud properties as seen by the MODIS instrument for this typhoon overpass. The eye of the cyclone is well depicted in the figure. Generally, in extreme events, such as the event of this tropical cyclone, cloud top pressure should be considerably lower than the vicinity, should have a cold temperature on top of the cloud, and should contain a good amount of ice particles. Such phenomenon is quite evident in this typhoon case, as can be seen from the figure.

Fig. 2
figure 2

MODIS cloud top pressure (top left), cloud top temperature (top right), cloud optical thickness (bottom left), and cloud phase optical properties (bottom right) from the 0120 UTC 7 November 2013 overpass

Full size image

3 Results

In this section, we will investigate the variations in simulated cloud properties between the different simulation experiments through the ARW solver. We will also be comparing the simulated cloud information against MODIS overpass data.

Figure 3 depicts the contour diagrams of cloud mixing ratio for the 10 simulation experiments, shown as a function of simulation hours (0–48 h). It is quite certain that the simulation of cloud mixing ratio varies notably among the experiments. Another similar figure is constructed in Fig. 4, but for the rain-mixing ratio. More detailed quantitative comparisons can be made by providing accumulative LWP and IWP amounts for different simulation runs (Fig. 5). As can be seen, the accumulative cloud water path amounts can vary substantially from one physics combination to another one. In particular, the use of different microphysics options can lead to completely different LWP amounts over the entire simulation period. For instance, at the end of 48-h simulation period, more than 2-mm LWP difference is seen between THOM-YSU and MY2-YSU simulation runs. Note that only dissimilarity between these two runs has been the use of different microphysics options, while other physics options are kept the same. It is also important to outline that the choice for using or not using cumulus parameterizations can play very important role in the simulation results. In our experiments, the THOM-YSU-NCu option, which does not use any cumulus parameterization scheme, has overestimated the LWP quantities than any of the physics combination used. This highlights the importance of using a cumulus parameterization scheme for the cyclonic event simulation. By contrary, the differences in the simulations due to PBL options are relatively small. For the IWP, similar tendencies are seen, demonstrating the lesser variability of IWP amounts between different PBL options (ACM2, BouLac, MYNN, QNSE, YSU with WSM6) than the different microphysics options (GGCE, MY2, THOM, WSM3, WSM6 with YSU). Obviously, WSM3 microphysics parameterization does not use the ice mass variable (Q i ), so IWP cannot be calculated.

Fig. 3
figure 3

Evolution of cloud mixing ratio (kg/kg) during the 48-h WRF simulation with different combinations of physics parameterizations from domain 2

Full size image
Fig. 4
figure 4

Same as Fig. 3, but for rain mixing ratio (kg/kg)

Full size image
Fig. 5
figure 5

Cumulative LWP (left) and IWP (right) simulation as a function of simulation hour with different combinations of physics parameterizations

Full size image

In Fig. 6, we provide the WRF simulated IWP fields obtained from the 10 simulation experiments over domain 2 and compare them against MODIS-derived IWP fields. The WRF simulated fields are taken from the closest hours (0100 UTC) to the MODIS overpass time (0120 UTC) on 7 November 2014. For a fair comparison, the MODIS IWP fields are averaged to the WRF domain. All simulation runs reveal a good dynamic structure and location of the cyclone, and, in fact, very close to the field observed by the MODIS. The exception is in the WSM3-YSU case, which has not produced any IWP field due to lack of ice mass variable in the scheme. Apart from this, the WRF model simulates the maturity of the cyclone quite well with realistic coverage of ice clouds. The eye and eyewall of the cyclone are also well produced. Nevertheless, the intensity of the cyclone seems overestimated in terms of IWP amounts, if one attempts to compare the simulation results with MODIS fields. Remarkably, the simulations those used the same microphysics option (WSM6) but different PBL options (ACM2, BouLac, MYNN, QNSE, and YSU) expose very similar IWP fields. Nonetheless, the microphysics options have very strong influence on the spatial coverage of the IWP field. For example, compared to WSM6-YSU simulation, MY2-YSU simulation exerts larger area of geographical coverage. The revealed IWP intensities between these two are also different. Furthermore, the THOM-YSU-NCu simulation has produced very high amount of IWP fields, as compared to the other simulation runs. The discrepancies in the magnitudes of the IWP fields can further be visualized through constructing empirical cumulative distribution histograms between the WRF simulated and MODIS-derived IWP fields (Fig. 7). The figure also supports the indication that the greater variability of producing a wide range of IWP magnitudes is mainly associated with the microphysics options, rather than PBL schemes.

Fig. 6
figure 6

Simulated IWP (mm) at 0100 UTC on 7 November 2013 with different combinations of physics parameterizations from domain 2. Interpolated MODIS IWP field on the domain is also included for comparison

Full size image
Fig. 7
figure 7

Empirical cumulative distribution functions of the WRF simulated and the MODIS observed IWP from domain 2

Full size image

Referring to Fig. 2, one can confirm that the selected typhoon event has mainly resulted in ice cloud phase properties in the vicinity of the MODIS overpass granule. Very small spatial precipitation cells are denoted as liquid clouds. Now, we would like to investigate the capability of the WRF model for simulating accurate ice and liquid cloud phase optical properties in comparison with MODIS observations. Table 2 exhibits the percentage of ice and liquid cloud phase cells from the WRF and MODIS datasets over domain 2. It can be seen that 75.93 % pixels are distinguished as ice clouds by the MODIS algorithm in the domain datasets. Only 10.59 % pixels are denoted as liquid pixels. Remaining pixels are associated with clear scenes. In comparison to the MODIS, WRF simulations contain fewer ice clouds but more liquid clouds. Nevertheless, the frequency range of ice and liquid clouds as simulated by the WRF model over domain 2 by different physics options has appeared to be very comparable. Except for the WSM3-YSU simulation, around 65–68 % pixels are calculated as ice clouds, while 29–37 % pixels are calculated as liquid clouds. In general, the difference between the WRF and the MODIS cloud phase properties can be considered as somewhat reasonable.

Table 2 The WRF and MODIS-derived ice and liquid cloud phase frequencies over the domain 2
Full size table

4 Conclusions

This manuscript examines the capability of a mesoscale NWP model (WRF-ARW) for accurately simulating cloud properties on a tropical cyclonic event over the Northwestern Pacific Ocean. A total of 10 high-resolution simulation experiments have been carried out to investigate the sensitivity of different combinations of physics parameterizations for such simulations. The simulation experiments reveal a strong influence of physics parameterizations on cloud simulation. The variations between the simulated cloud water path amounts have been found very substantial. In particular, such differences are mainly associated with the choice of microphysics parameterizations rather than PBL schemes.

The simulated cloud properties are further compared against MODIS level 2 cloud products. It is found that, in general, the WRF dynamic solver is able to duplicate the realism of cloud fields over the typhoon system. Moreover, the WRF simulated cloud frequencies have agreed reasonably well with the MODIS datasets. Additionally, the use of cumulus parameterization schemes is also essential for good cloud simulation. Nevertheless, the model has been unable to accurately simulate the intensity of the cloud information (e.g., IWP), if one takes the MODIS optical cloud properties as a reference. However, while interpreting the results, one should also remember the obvious difference between MODIS and WRF simulations due to the mismatch of horizontal spatial resolutions. Furthermore, MODIS cloud properties are certainly not true. They are taken as a reference in this work, but one should consider the uncertainties associated with the MODIS retrievals as well.

Data assimilation can be very useful for improving the WRF simulation and prediction of cloud properties. In the future, we will be investigating whether assimilating satellite radiances could improve the results.