**4. Numerical modelling of tropical cyclones**

A significant number of studies regarding TC propagation, track prediction, time and place of landfall and intensity of the storm are carried out for several ocean basins including NIO. Considerable improvements in predicting the TCs are also achieved till date. In view of these, this section highlights the recent developments regarding TC predictability over NIO region and the current scenario.

#### **4.1. Model predictability**

two vortices, if they are of equal strength. In the presence of the β effect, the two vortices rotate around each other relative to the centre of rotation. This centre of rotation is not fixed and, instead, moves northwest ward in response to the 'β effect'. 'Fujiwhara effect' is noticed over other basins of the world including Atlantic, but is not applicable for TCs formed over NIO.

The most common way of dissipation of a TC is its landfall. When the storm moves over land, it deprives itself from warm water and the available moisture over ocean. Consequently, it is deprived from the energy source and the warm core with thunderstorms near the centre turns into a remnant low-pressure area due to quick loss of energy. Weakening can also occur if it encounters a vertical wind shear that causes the heat engine and convection shift away from

> <sup>3</sup> *E Cv D D* = r

where *E*D is the rate of energy dissipation per unit time per unit horizontal surface area, *v* defines the wind speed, *ρ* is for air mass density, and '*C*D' is the drag coefficient that depends upon the surface irregularities. Since the power dissipation in TCs is proportional to the cube of its wind velocity, the severity can be computed as the cumulative sum of the cube of the

There are two sources which are capable of changing the TC intensity, one is internal variability and other one is environmental interaction. One important aspect of later source is the interaction between the ocean and the storm system. Usually TC is regarded as the most forceful case in air-sea interaction studies where energy from the warm ocean waters is delivered via surface heat flux [30]. The ocean response is quite sensitive to the surface drag coefficient. Emanuel [31] used a simple numerical model to establish the progress of hurricane intensity. Their findings advocate that in most cases, the intensity depends on three factors, viz. initial intensity of cyclone, thermodynamic state of atmosphere through which the cyclone propagates and finally the heat exchange with the upper layer of the ocean underlying the core of the cyclone. Rapid intensification of TC is noticed when it passes over the deep upper ocean mixed layer and that upper ocean thermal structure plays a significant role in the intensification process [32–34]. Sutyrin [35] performed simulations with a coupled model of the oceanic and atmospheric boundary layers and concluded that the interaction is strong enough to change the supply of heat and moisture fluxes from the ocean into the atmosphere significantly within

few hours of the formation of the storm and consequently, influence the TC intensity.

The intensity of TC increases with increase in SST and upper ocean heat content [36]. The positive feedback occurs when genesis and intensification happens. During this phase, the

the centre. The rate of power dissipation of TCs can be computed [29] as

wind velocity over time according to the above equation.

200 Recent Developments in Tropical Cyclone Dynamics, Prediction, and Detection

**3. Role of ocean in genesis and intensification**

**2.4. Dissipation**

Various regional models such as GFDL (USA), ALADIN (France), Quasi-Lagrangian Limited Area Model or QLM (India), MM5 (USA), etc. are used for TC research and operational forecasting purpose. Apart from these, the Eulerian-mass-based dynamical core of Weather Research and Forecasting (WRF) model, designed as the successor to MM5 is also used to predict TCs. The variants of WRF regional model are Advanced Research WRF or ARW and non-hydrostatic mesoscale model or WRF-NMM. Though these numerical models are quite capable for real-time predictions in regional scale, they need appropriate initial and boundary conditions from global models. For example, a recent study carried out by Kumar [43] discusses about the impact of European Centre for Medium-Range Weather Forecasts (ECMWF), National Centers for Environmental Prediction (NCEP) and National Centre for Medium Range Weather Forecasting (NCMRWF) global model analysis on the WRF model forecast for TC prediction over Indian region. This study indicates some of the inherent limitations of such global analyses data sets including the consideration of few fundamental aspects like that of the middle tropospheric humidity profiles those are important for TC genesis. Another limitation of such data sets is their horizontal resolution though recent advancements have made availability of some of the usable global analyses for the desired purpose with higher spatial resolutions up to 0.25°.

Since NWP models are equipped with real-time prediction capability, they are being used increasingly for the TC prediction over NIO region as well. Some of the numerical models and their skills are discussed here. For instance, QLM regional model was adopted by Prasad [44] for cyclone track prediction over NIO region and found the performance to be reasonable. The recurvature of the cyclones were also well predicted. However, the model performance for TC intensity prediction was not satisfactory. Another notable study by Mohanty et al. [45] used MM5 to simulate Orissa (Odisha) super cyclone (1999) for predicting track, intensity, mean sea level pressure and associated precipitation. Though such types of studies were able to improve the prediction of several relevant parameters including TC tracks, they were not so successful in predicting the intensity accurately like the studies performed using QLM. Similarly, some recent studies used three variants of the next-generation mesoscale WRF model (i.e. ARW, WRF-NMM, and Hurricane Weather Research and Forecasting Model or HWRF) for TC research and operational purpose as well [51, 53, 56, 57, 66]. It may be noted that ARW uses Arakawa C-grid staggering while WRF-NMM and HWRF use Arakawa E-grid. All of the WRF model variants use terrain following co-ordinate system and specific physical parameterizations. Since several modelling features in WRF are quite advanced (e.g. moving nest feature in HWRF) as compared to MM5, it is expected that at least one or more variants of it would show better performance for TC prediction over NIO region. Extensive research in this direction using ARW suggests some significant improvements in predicting the tropical cyclogenesis and cyclone tracks [10, 46–54]. However, it is noticed that improvement in prediction of TC intensity is found to be slower than that of track [51, 55].

A comparison study among MM5, WRF-ARW and WRF-NMM for very severe cyclone Mala (2006) developed over BOB found that ARW could simulate the TC intensity in terms of minimum central pressure and maximum sustainable wind with better accuracy [56]. However, MM5 simulated a more rapidly intensified storm and delayed landfall and WRF-NMM failed to simulate the intensity of the storm properly. On the other hand, WRF-NMM predicted TC track more accurately as compared to ARW and MM5. The TC Mala when simulated using HWRF with different initial conditions, the track error was found to be ∼200 km and the intensity prediction was reasonably good for some considered initial conditions though the amount and spatial distribution of rainfall was well simulated by the model [57]. In order to improve the predictability, appropriate nesting technique, horizontal and vertical resolutions as well as physical parameterizations are considered [59, 68] besides data assimilation [60]. In view of these aspects, the HWRF system is now implemented at IMD along with the already operational ARW model for forecasting of TCs over NIO basin. As part of the Forecast Demonstration Project (FDP) conducted by IMD, it is analysed that the performance of ARW without data assimilation is reasonable over BOB [61]. Its performance improves when available observations are assimilated. Similar is the case with WRF-NMM. On the other hand, HWRF is capable of simulating rapid intensification of TCs over NIO region due to its improved vortex relocation and initialization procedures [49].

The high-resolution mesoscale modelling systems provide better guidance for TC forecast up to 72 h over NIO region [61]. They require high-resolution global analyses data sets for appropriate initial and boundary conditions in order to bring in large-scale boundary forcing [62]. In order to reduce model errors, the initial and boundary conditions can be improved by adopting appropriate data assimilation techniques by incorporating the conventional, radar and satellite observations before running the model [61]. Thus, these aspects need special attention as far as predictability of TCs over NIO region is concerned.

#### **4.2. Role of physical parameterizations**

Since NWP models are equipped with real-time prediction capability, they are being used increasingly for the TC prediction over NIO region as well. Some of the numerical models and their skills are discussed here. For instance, QLM regional model was adopted by Prasad [44] for cyclone track prediction over NIO region and found the performance to be reasonable. The recurvature of the cyclones were also well predicted. However, the model performance for TC intensity prediction was not satisfactory. Another notable study by Mohanty et al. [45] used MM5 to simulate Orissa (Odisha) super cyclone (1999) for predicting track, intensity, mean sea level pressure and associated precipitation. Though such types of studies were able to improve the prediction of several relevant parameters including TC tracks, they were not so successful in predicting the intensity accurately like the studies performed using QLM. Similarly, some recent studies used three variants of the next-generation mesoscale WRF model (i.e. ARW, WRF-NMM, and Hurricane Weather Research and Forecasting Model or HWRF) for TC research and operational purpose as well [51, 53, 56, 57, 66]. It may be noted that ARW uses Arakawa C-grid staggering while WRF-NMM and HWRF use Arakawa E-grid. All of the WRF model variants use terrain following co-ordinate system and specific physical parameterizations. Since several modelling features in WRF are quite advanced (e.g. moving nest feature in HWRF) as compared to MM5, it is expected that at least one or more variants of it would show better performance for TC prediction over NIO region. Extensive research in this direction using ARW suggests some significant improvements in predicting the tropical cyclogenesis and cyclone tracks [10, 46–54]. However, it is noticed that improvement in

202 Recent Developments in Tropical Cyclone Dynamics, Prediction, and Detection

prediction of TC intensity is found to be slower than that of track [51, 55].

improved vortex relocation and initialization procedures [49].

A comparison study among MM5, WRF-ARW and WRF-NMM for very severe cyclone Mala (2006) developed over BOB found that ARW could simulate the TC intensity in terms of minimum central pressure and maximum sustainable wind with better accuracy [56]. However, MM5 simulated a more rapidly intensified storm and delayed landfall and WRF-NMM failed to simulate the intensity of the storm properly. On the other hand, WRF-NMM predicted TC track more accurately as compared to ARW and MM5. The TC Mala when simulated using HWRF with different initial conditions, the track error was found to be ∼200 km and the intensity prediction was reasonably good for some considered initial conditions though the amount and spatial distribution of rainfall was well simulated by the model [57]. In order to improve the predictability, appropriate nesting technique, horizontal and vertical resolutions as well as physical parameterizations are considered [59, 68] besides data assimilation [60]. In view of these aspects, the HWRF system is now implemented at IMD along with the already operational ARW model for forecasting of TCs over NIO basin. As part of the Forecast Demonstration Project (FDP) conducted by IMD, it is analysed that the performance of ARW without data assimilation is reasonable over BOB [61]. Its performance improves when available observations are assimilated. Similar is the case with WRF-NMM. On the other hand, HWRF is capable of simulating rapid intensification of TCs over NIO region due to its

The high-resolution mesoscale modelling systems provide better guidance for TC forecast up to 72 h over NIO region [61]. They require high-resolution global analyses data sets for appropriate initial and boundary conditions in order to bring in large-scale boundary forcing The physical parameterizations which include cumulus convection, surface fluxes of heat, moisture, momentum and vertical mixing in the planetary boundary layer play an important role in determining structural development, intensification and movement of TCs [10, 46, 48, 50, 53, 58, 63–65]. A number of studies emphasized upon these aspects during the past three decades. For the simulation purpose, they use the previously mentioned models (see Section 5.1). Most of these studies conduct simulations over a particular ocean basin. For instance, Osuri et al. [50] conducted a systematic study on customization of ARW model considering several physical parameterization schemes for the simulation of five TCs over NIO region. The study found that the combination of Yonsei University (YSU) planetary boundary layer (PBL) parameterization with KF convection scheme provided a better prediction for structural characteristics, intensity, track and rainfall. Similar results were also achieved by several studies including that of [10, 46, 48]. Thus, most of the studies (including [65]) found the performance of KF scheme to be better for the prediction of TCs over NIO region. However, recent studies by Kanase and Salvekar [53] obtained that the Betts‐Miller‐Janjic (BMJ) convec‐ tion scheme performs better as compared to other parameterizations in the group although the study also favoured using YSU PBL physics. On the other hand, it found that WRF single‐ moment (WSM)‐6 microphysics better represents mid‐tropospheric heating as compared to WSM‐3 favouring better intensity simulation.

Though HWRF has not been extensively used for sensitivity studies with respect to physical parameterizations for simulation of TCs over NIO region, its primitive variant WRF‐NMM was used in recent past by some of the researchers. For example, studies by Pattanayak et al. [66] found that the combination of Simplified Arakawa‐Schubert (SAS) convection, YSU PBL, Ferrier microphysics and NMM land‐surface parameterization schemes in WRF‐NMM performs better in predicting track and intensity of TC Nargis (2008) over BOB. Therefore, an extensive evaluation of HWRF is needed in order to determine the combination of physical parameterizations that performs better for TC prediction over NIO region before it is adopted for the operational forecasting purpose.

#### **4.3. Significance of grid resolution**

The grid resolution of a model also impacts the TC prediction [51, 58, 59, 67]. However, there are very few studies available relating to the impact of grid resolution on TC prediction over NIO region. One of the notable studies by Rao [68] evaluated the impact of horizontal resolu‐ tion and the advantages of the nested domain approach in the prediction of Orissa (Odisha) super cyclone intensification and movement by using MM5 model. Results from this study indicate that the enhancement of resolution produces higher intensity but does not influence the track of the storm. The nested experiments produced cyclone track closely agreeing with the observations, while the single domain based simulations show the deviation of the track towards north. A more recent study by Osuri et al. [51] found that the use of high resolutions in operational ARW model improves the prediction of recurving TC tracks and their intensity. In a climatological framework, Community Atmospheric Model or CAM showed sensitiveness to the prediction of more number of intensified tropical cyclones over most of the global basins including NIO. Further, it also found that the duration of tropical storms would be much larger in high resolutions simulations. Thus, it is realized that the model horizontal grid resolution impacts significantly the TC track, intensity and duration besides other relevant meteorological parameters.

#### **4.4. Significance of data assimilation**

Most of the times, the use of data assimilation techniques in TC simulations helps in improving the model predictability. For this purpose, satellite-based observations, aircraft measurements and radar data are used besides the conventional data sets. The widely used data assimilation techniques are primarily based on either ensemble Kalman filter (EnKF) or variational techniques (3DVAR or 4DVAR). Most of the studies related to TC simulation were done using variational data assimilation techniques for improving the TC prediction over NIO region. For example, the studies such as [52, 69–71] used 3DVAR techniques for assimilating satellite, radar and conventional measurements for improving the initial and boundary conditions of MM5 and ARW mesoscale models in order to better predict TC structure, track, intensity and associated relevant meteorological variables including rainfall. In some situations, the improvement was not significantly noticed. For instance, the studies by Singh et al. [70] found that assimilation of SSM/I wind speed data resulted in simulating weak intensity and failed to make an impact on track prediction.

Although there are no significant studies related to the use of 4DVAR and EnKF techniques for simulating NIO TCs, there are literatures, which demonstrate the usage of four dimensional data assimilation (FDDA) nudging technique in order to improve the ARW model predictability. For example, [71–73] used FDDA nudging technique in order to improve ARW initial and boundary conditions for the simulation of several TCs over NIO region those occurred during 2007–2010. These studies primarily emphasized upon TC track and intensity forecasts. While some of them reported remarkable improvements in track prediction and landfall position with either 12- or 18-h of nudging yielding maximum impact [72, 74], some others noticed relatively less impact of FDDA observational nudging on intensity prediction [73].
