Rapid intensification of Arabian Sea cyclonic storm and numerical simulation of ESCS Tauktae using the high-resolution ARW model

1 Introduction

Tropical cyclones (TCs) are destructive natural disasters that can cause significant loss of life and property (Chen et al., 2015; Nystrom et al., 2020). The North Indian Ocean (NIO), which comprises the Arabian Sea (AS) and the Bay of Bengal (BoB), is one of the most active regions in the world for tropical cyclone (TC) development (Balaji et al., 2018). The Arabian Sea basin appears to have more potential for producing strong TCs than the Bay of Bengal (BoB) basin, as energy parameters such as accumulated cyclone energy (ACE) and the Power Dissipation Index (PDI) have increased (Singh and Roxy, 2022; Thomas and Lekshmy, 2022; Das et al., 2023). For instance, on 6 June 2023, the AS was severely impacted by Biparjoy, an extremely severe cyclonic storm (ESCS), which has the highest life span of approximately 13 days (RSMC Report, 2023). In 2021, the AS experienced the ESCS Tauktae, whereas the BoB experienced the very severe cyclonic storm (VSCS) Yaas (RSMC Report, 2021). Noticeably, in 2019, there was a rare phenomenon: a back-to-back occurrence of cyclones over the AS, super cyclonic storm (SuCS) Kyarr and ESCS Maha. Furthermore, in the same year, it was noted that the AS experienced four intense cyclones, whereas the BoB experienced only two intense cyclones (RSMC Report, 2019). A study reported a 52% increase in the frequency of AS TCs and an 8% decrease in the BoB TCs from 2001 to 2019 (Deshpande et al., 2021). Rising sea surface temperatures (SSTs) associated with the El Niño/Southern Oscillation (ENSO) and Indian Ocean Dipole (IOD) contribute to the intensification of tropical cyclones in the Arabian Sea (Pathirana et al., 2022). In general, these studies have demonstrated that the frequency, strength, and lifespan of the AS TCs are increasing, which leads to an increase in the vulnerability of the west coast of India. Therefore, an accurate TC forecast will help prevent losses to livelihoods, physical property, ecosystems, and the environment caused by storm activity.

Advances in numerical weather prediction (NWP) have improved TC forecasting, but the accurate prediction of intensity, particularly rapid intensification (RI), remains challenging because it depends on multiple large-scale environmental conditions, such as vertical wind shear and upper-tropospheric conditions (Wang and Wang, 2014; Zhang et al., 2017; Chang et al., 2020). The availability of high-performance computing systems has facilitated numerous simulation studies employing high-resolution mesoscale models to examine tropical cyclones over the NIO (Satoh et al., 2017; Judt and Rios-Berrios, 2021; Singh et al., 2021a, 2021b). High-resolution models can explicitly focus on convection processes, resulting in significant improvements in tropical cyclone simulations over the past three decades, including more accurate representations of wind speed, intensity, and structure (Davis, 2018; Gao et al., 2023). The Advanced Research version of the Weather Research and Forecasting (ARW-WRF) model is commonly used to simulate various weather phenomena, such as severe rainfall (Routray et al., 2010; Dodla and Ratna, 2010), both in the Indian monsoon region and around the world. It is widely used to simulate and forecast TCs (Osuri et al., 2012; Davis et al., 2008).

Several studies have demonstrated that planetary boundary layer (PBL) schemes have a huge impact on tropical cyclone movement, intensity, and structure (Gopalakrishnan et al., 2013; Vijaya Kumari et al., 2019; Njuki et al., 2022). A study suggested that the PBL scheme regulates surface fluxes of momentum, latent heat, and sensible heat, which are vital for the genesis and development of tropical cyclones and play a crucial role in cyclone intensification (Pattanaik and Rama Rao, 2009). Moreover, PBL schemes regulate energy flow at the air–sea interface and throughout the atmosphere, thereby playing a crucial role in the forecast of tropical cyclones (TCs). The track, landfall, and intensity forecasts were enhanced with the use of the Yonsei University (YSU) PBL scheme in conjunction with the enthalpy coefficient for the heat and air–sea flux parameterization (FLUX-1) at a horizontal resolution of 3 km in the WRF model (Singh et al., 2023). A sensitivity study conducted by Ashok and Singh (2025) suggested that employing the Goddard microphysics scheme improved the track, intensity, and structure of the AS tropical cyclone using the WRF model with a high horizontal resolution of 3 km. Similarly, Singh et al. (2020) reported that cyclone forecasts were improved by employing the Goddard microphysics scheme in combination with the PBL scheme YSU in the WRF model.

This study is structured into five sections: Section 1 presents the introduction and relevant literature review; Section 2 provides a brief description of the ESCS and case study. Section 3 describes the experimental design, methodology, data used, and PBL schemes. Section 4 presents the results and discussion, including subsections “Predicted cyclone track and intensity” and “Performance of the model on structure of the storm.” Section 5 concludes the study.

2 Brief description of ESCS over the Arabian Sea

During the period 1990–2022, the Arabian Sea experienced a total of 11 extremely severe cyclonic storms (ESCSs). A notable increase is observed during 2014–2021, in which the frequency and duration of maximum wind speed also increased, as indicated in IMD best-fit track data. This suggests that, in recent years, there has been an increase in the intensity of AS tropical cyclones. The occurrence of rapid intensification (RI) is dangerous during landfalling, and this can lead to enormous damage (wind speed changes of at least 30 knots) in a changing climate (Singh et al., 2022). As shown in Figure 1, it was found that, during 2012–2022, the RI and rapid dissipation (RD, wind speed decreased by at least 24 knots in a 24-h duration) increased over the AS. The green line approximately represents the 300-km offshore boundary from the Indian coastline, which provides a useful reference for checking whether these RI changes specifically occur over open ocean or closer to the coastal region. The results showed a clear spatial clustering of RI events, although their locations and frequency vary considerably between the three periods. During 1990–2000, RI events (black circles) appear relatively concentrated over the central and eastern Arabian Sea. In contrast, RD events are distributed more widely, including the northern Arabian Sea and regions farther east and north. During 2001–2011, RI events exhibit a more scattered distribution, with occurrences extending from the southern Arabian Sea toward the central Arabian Sea. Several blue RD events are observed over the Arabian Sea and the Indian subcontinent, indicating substantial spatial variability in rapid weakening. The 2012–2022 period shows a noticeably larger number of both RI and RD locations compared with the earlier periods. The results show an important feature in which several RI locations occur within or near the 300-km offshore zone, emphasizing the potential importance of coastal-oceanic environmental conditions. In contrast, RD locations are frequently observed toward the northern portions of the basin and near land/coastal regions. Overall, Figure 1 demonstrates an evolution of RI and RD locations over the Arabian Sea and suggests an increase in the number of locations during 2012–2022, together with the concentration of RI locations over the Arabian Sea.

2.1 Case study: ESCS Tauktae

ESCS Tauktae developed over the Arabian Sea during 2021 and was identified as a low-pressure area on 13 May 2021. It intensified into a depression and then into a cyclonic storm on 14 May. It moved northward and intensified into a severe cyclonic storm on 15 May. It moved north-northwestward and became an ESCS on 17 May 2021. After landfall, the storm had a maximum sustained wind speed of 90 knots, gusting to 100 knots, and moved north-northeastward before weakening into a severe cyclonic storm on 18 May. It continued to move north-northeastward, causing depressions over Gujarat and South Rajasthan. With these insights, this study uses the ARW model to evaluate the effects of the PBL scheme on the forecast of ESCS Tauktae at a high horizontal resolution (3 km).

3 Methodology

3.1 Experimental design

The ARW-WRF model (Skamarock et al., 2019; Version 4.0) was used to conduct the study. The model configuration was based on the studies by Singh et al. (2021a, 2021b) and Ashok and Singh (2025), with a two-way nested domain with an outer domain of 15 km (451 × 377 grid points) and an inner domain of 3 km (221 × 221 grid points) (Figure 2). The model domain comprises 51 vertical levels, with the model top at 10 hPa. The model employed the Goddard microphysics scheme (Tao et al., 2016), Dudhia’s short-wave (SW) radiation scheme (Dudhia, 1989), and a long-wave radiation scheme based on the rapid radiative transfer model (RRTM) (Mlawer et al., 1997). The Kain–Fritsch scheme (Kain, 2004) was exclusively used for cumulus convection in the outer domain at 15 km resolution, while for the inner domain (3 km resolution), the cumulus convection was switched off. Table 1 provides further information on the model setup used in the study. Six experiments were designed using six PBL parameterization schemes, including the YSU scheme, the Mellor–Yamada–Janjić (MYJ) scheme, the QNSE scheme, the Mellor–Yamada–Nakanishi–Niino Level 2.5 (MYNN2) scheme, the Mellor–Yamada–Nakanishi–Niino Level 3 (MYNN3) scheme, and the ACM2 scheme.

Table 1. Model configuration used in the study for the simulations.

3.2 Data used

• The initial/boundary conditions were derived from the National Centers for Environmental Prediction Final Analysis (NCEP-FNL). India Meteorological Department (IMD) best-fit track data were used for the validation of the track and intensity. Integrated Multi-satellite Retrievals for Global Precipitation Measurement (GPM) Mission (IMERG) data, retrieved from https://giovanni.gsfc.nasa.gov/giovanni/, were used for rainfall validation.

The following metrics were used to measure the forecast skill of the model:

Track Error ( in km ) = � l o n 2 + � l a t 2 (1)

Absolute Error = | P i − O i | (2)

Root Mean Square Error ( RMSE ) = ∑ i = 1 N ( P i − O i ) 2 N (3)

Index of Agreement ( IOA ) = 1 − ∑ i = 1 N ( O i − P i ) 2 ∑ i = 1 N ( | P i − O ¯ | + | O i − O ¯ | ) 2 (4)

Standard Deviation , ( σ ) = 1 / N ∑ i = 1 N ( M i − M ¯ ) 2 (5)

Normalized Mean Square Error ( NMSE ) = 1 / N ∑ i = 1 N ( P i − O i ) 2 P O ¯ (6)

Here, Oi represents the observed data, and P_i represents the predicted data. O ¯ represents the observed mean, and Δlat (Δlon) represents the differences between the observed and predicted latitude (longitude). The values of IOA vary from 0 to 1, where 0 indicates poor performance and 1 represents a perfect match. The model operates well in both space and time, as seen by the low NMSE. On the other hand, a high NMSE number does not always indicate that the model is entirely incorrect. For a better understanding of rainfall, based on intensities moderate (7.6–35.5 mm/day), rather heavy rainfall (35.6–64.4 mm/day), heavy rainfall (64.5–124.4 mm/day), and very heavy rainfall (124.4–244.5 mm/day), a statistical analysis was performed in terms of probability of detection (POD or hit rate), false alarm ratio (FAR), post-agreement (PAG), and frequency bias index (FBI) (Pavani et al., 2023; Thankachan and Singh, 2024)¹ where,

P O D = A A + C (7)

F A R = B A + B (8)

P A G = A A + B (9)

F B I = A + B A + C (10)

A better (poor) forecast is indicated if the values of POD and PAG are near 1 (0), and if the values of FAR and FA are near 0 (1); such a forecast is considered better (poor). If the values of FBI are greater than 1, then the event is over-predicted; otherwise, it is under-predicted. The categorical scores were calculated using a 2 × 2 contingency table.

Table 2.

Where A indicates the number of times a “yes” forecast is followed by a “yes” occurrence, B indicates the number of times a “yes” forecast is followed by a “no” occurrence, C indicates the number of times a “no” forecast is followed by a “yes” occurrence, and D indicates the number of times a “no” forecast is followed by a “no” occurrence.

3.3 Description of PBL parameterization schemes

Planetary boundary layer (PBL) schemes play a crucial role in NWP by parametrizing the turbulent exchange of momentum, heat, and moisture between the Earth’s surface and the atmosphere. These schemes differ in their representation of turbulent mixing (Cohen et al., 2015). They are commonly classified as local or non-local closure schemes, as well as first-order, 1.5-order, and higher-order closure schemes (Ye et al., 2024). Local closure schemes assume that the turbulent fluxes are proportional to local gradients, whereas non-local schemes account for the effect of large eddies and convective transport. This study evaluates six PBL schemes: YSU (Hong and Lian, 2026), MYJ (Janjić, 1994), QNSE (Sukoriansky et al., 2005), MYNN2 (Nakanishi and Niino, 2006), MYNN3 (Nakanishi and Niino, 2009), and ACM2 (Pleim, 2007). YSU is a first-order non-local closure scheme that represents vertical mixing using a counter-gradient flux term. MYJ is a one-dimensional prognostic turbulent kinetic energy scheme with local vertical mixing. QNSE is a turbulent kinetic energy (TKE) prediction option that uses a new theory for stably stratified regions. MYNN2 is a TKE-based local closure scheme with an improved mixing-length formulation and a prognostic TKE equation, and MYNN3 is a more advanced version of MYNN2 with higher-order closure, designed to better capture non-local turbulent transport. ACM2 is a hybrid local–non-local closure scheme, which combines local TKE-based mixing in stable conditions with non-local transport in convective conditions. Owing to their differing treatments of turbulence and vertical mixing, these schemes can significantly influence tropical cyclone track, intensity, boundary layer structure, air–sea fluxes, and precipitation forecasts.

4 Results and discussion

4.1 Predicted cyclone track and intensity

The simulated tracks of cyclone Tauktae using six different PBL schemes: YSU, MYJ, QNSE, MYNN2, MYNN3, and ACM2, along with the IMD best-fit track, are represented in Figure 3. The IMD track shows that the cyclone moved northward along the Arabian Sea and made landfall on the Gujarat coast. All model-simulated tracks capture the general northward movement of Tauktae reasonably well. QNSE and ACM2 closely follow the IMD track throughout the cyclone’s life cycle, including the landfall position. During the beginning stage, QNSE shows a realistic path, but it moves quite westward before landfalling. After landfall, YSU and MYJ move eastward, which makes their errors higher than those of the IMD. There is a noticeable eastward bias observed in MYNN2 and MYNN3, which indicates that they do not represent the steering flow and boundary layer dynamics well. Figure 4 represents the time series of model-simulated TC (a) track errors (Equation 1) and (b) daily track errors and mean track errors. Figure 4a shows that all the schemes produce fewer errors (<100 km) in the first 12–24 h, indicating that there is no substantial difference between them. After 24 h, the errors gradually increase in all the experiments. Notably, over 48–60 h, the track error increases gradually for all schemes except for QNSE and ACM2, which preserve the lowest track errors (<250 km). It was found that both YSU and MYJ show moderate track errors (~ 300–400 km), and the errors of both MYNN schemes’ errors reach a maximum of 600 km. At 90 h, QNSE exhibits the lowest track error, and MYNN3 shows the highest deviation when compared with the other schemes. The rising error trend shows that all the schemes adequately capture cyclone movement in the beginning. Since QNSE continuously maintains the lowest track error, it may provide a better representation of cyclone steering currents and vertical mixing (Ye et al., 2024). Figure 4b represents the daily and mean track errors of cyclone Tauktae from different PBL schemes compared with the IMD best-fit track. The mean track error is approximately 168, 185, 104, 216, and 134 km for YSU, MYJ, QNSE, MYNN2, MYNN3, and ACM2, respectively. From Day 1 to Day 2, all the experiments show a lower track error of approximately 100 km. Gradually, the error increases from Day 3 onward, while QNSE remains consistently lower in error. Both MYNN2 and MYNN3 show larger errors, exceeding approximately 450 km. On the other hand, YSU and MYJ are off by approximately 450 km. The increase in error after Day 3 illustrates that boundary-layer processes have a progressively larger effect on the cyclone’s movement, especially at landfall. As shown in Table 2, the smallest landfall position error is seen in QNSE (115 km), followed by ACM2 (147 km). The largest errors are produced in MYNN2 (295 km) and MYNN3 (291 km). YSU and MYJ also produce large errors (212–261 km). QNSE has the smallest timing error (+2 h), which indicates that it is more accurate. ACM2 also predicts the landfall well (+6 h). However, YSU (+11 h) and MYJ (+15.5 h) predict landfall significantly later. MYNN2 and MYNN3 indicate that landfalls will happen far later, with delays of more than 29 h. In other words, QNSE shows the best PBL scheme to predict cyclone Tauktae landfall (115 km, +2 h).

Table 2. Landfall position error and time obtained from all experiments.

Figure 5 demonstrates the time variations of simulated cyclone intensity in terms of maximum surface wind (MSW, in m/s), minimum central pressure (MCP, in hPa), 24-h wind changes (in m/s), and translation speed (in m/s) obtained from different PBL schemes (YSU, MYJ, QNSE, MYNN2, MYNN3, and ACM2) compared with IMD observations. IMD observations show a maximum wind of ~50 m/s for approximately 48 h, followed by weakening after ~60 h (Figure 5a). All model experiments overestimate MSW, with peaks between 55 and 65 m/s. All the models except QNSE and ACM2 fail to reproduce the rapid weakening trend observed by IMD after landfall, maintaining higher winds for longer. As shown in Figure 5b, it is observed from IMD that there is a steady drop in pressure to ~930 hPa near peak intensity (48 h), followed by a sharp increase after landfall. Both MYNN2 and MYNN3 align with IMD up to 48 h; after that, they deviate. Both ACM2 and QNSE show better alignment with IMD after 48 h. Models underestimate the post-landfall rise, indicating delayed weakening. Figure 5c represents 24-h wind speed changes; it is found that IMD shows a strong increase in winds during intensification (30 m/s). Simulations capture the early intensification phase reasonably well. However, the majority of models underestimate the weakening phase, showing a slower reduction compared to IMD. As shown in Figure 5d, IMD shows variable translation speeds, peaking near 7 m/s just before landfall. Model experiments reproduce the general variability, with speeds ranging from 3 to 6 m/s. ACM2 and QNSE better capture the peak translation near landfall. MYNN2 and MYNN3 tend to underestimate translation speed in the later stages, consistent with their delayed landfall predictions.

Table 3 illustrates the statistical performance indices for MSW (in m/s) and MCP (in hPa) from different PBL experiments. The lowest RMSE of MSW (MCP) (Equations 2, (3)) is observed in ACM2, approximately 13.59 m/s (30.59 hPa), followed by QNSE, which is approximately 18.04 m/s (32.1 hPa). The standard deviation (SD) (Equation 5) ranges from 7.7 (ACM2) to 11.2 (MYNN3) for MSW and from 16.49 (YSU) to 20.73 (MYJ) for MCP. The best IOA (Equation 4) observed for MSW (MCP) is found in ACM2, approximately 0.68 (0.57), followed by QNSE at 0.60 (0.55), with the lowest values for MYJ. The NMSE (Equation 6) is smallest in ACM2, approximately 0.000302, and in QNSE, it is 0.000362 (0.002229) for MSW (MCP), while MYJ has the largest NMSE, approximately 0.000490 (0.003881). In general, ACM2 performs consistently well, with the lowest RMSE and NMSE and the highest IOA, indicating its strong ability to reproduce intensities. QNSE also perform well with low RMSE and high IOA. The improved performance of ACM2 and QNSE is consistent with previous studies (Singh and Bhaskaran, 2017; Verma et al., 2023; Verma et al., 2021), which highlighted the importance of PBL parameterizations in Tc track and intensity forecasts through their influence on boundary layer processes.

Table 3. Statistical performance indices in terms of RMSE, SD, IOA, and NMSE of MSW (in m/s) and MCP (in hPa) of eight cyclones obtained from different experiments.

Bold values indicate the lowest error values.

4.2 Performance of the model on the structure of the storm

4.2.1 Rainfall prediction

Figure 6 shows the cumulative rainfall (mm) from different PBL experiments and GPM (0.1°) observations² over the landfall duration (00UTC on 17 May–00UTC on 18 May) of cyclone Tauktae. The GPM dataset shows that the maximum rainfall occurred around the Gujarat coast, close to where the storm made landfall. The peak rainfall exceeds 270 mm, with heavy rainfall bands extending inland and offshore. YSU and MYJ simulate the overall rainfall pattern but with slightly displaced maxima toward inland Gujarat. Rainfall intensity is somewhat overestimated. QNSE closely matches the observed rainfall pattern, with heavy rainfall mostly near the coast and realistic bands of rainfall offshore. Simulations of MYNN2 and MYNN3 show rainfall patterns that extend to the sea, which causes the maximum rainfall to move away from the observed location. Compared to GPM, ACM2 has a better rainfall distribution, with peaks near the coast but a little more spread out. Schemes that showed better track and landfall accuracy (QNSE and ACM2) also produced more realistic rainfall distributions. Conversely, the MYNN schemes with delayed landfall predictions also misrepresent rainfall placement and intensity. In general, during cyclone Tauktae’s landfall, QNSE provides the most accurate rainfall distribution compared to GPM observations, while ACM2 also performs reasonably well. YSU and MYJ capture the rainfall pattern but misplace maxima inland, whereas MYNN2 and MYNN3 exhibit elongated offshore rainfall biases. This confirms that accurate representation of boundary layer processes strongly influences cyclone rainfall forecasts.

The performance of different PBL schemes in predicting rainfall intensity was evaluated for the duration from 00UTC on 17 May to 00UTC on 18 May during cyclone Tauktae using four categorical verification indices: POD, FAR, PAG, and FBI (Equations 7–10) across different rainfall categories (moderate, rather heavy, heavy, and very heavy). Table 4 illustrates the statistical performance indices for the rainfall forecast verification for Tauktae. QNSE provides a better representation of moderate rainfall (7.6–35.5 mm/day) because POD and PAG values are the highest, approximately 0.39 and 0.49, respectively. FAR remains moderately lower for QNSE at 0.49, confirming its relative superiority in detecting moderate events. The FBI is 0.80, which is less than 1, showing that the rainfall is underpredicted by this scheme. In general, QNSE provides better performance in capturing moderate rainfall. All schemes fail to capture rather heavy rainfall (35.6–64.4 mm/day), but ACM2 shows better prediction with comparatively high POD and PAG values, with FBI less than 1 suggesting underprediction of the event. In the heavy rainfall (64.5–124.4 mm/day) category, it is found that all the indices provide high POD and PAG values, with FAR values remaining low for the PBL schemes, ACM2, and QNSE; this shows relatively better agreement. With maximum PAG and minimum FAR, QNSE shows better detection of very heavy rainfall (124.5–244.4 mm/day). However, PAG provided the maximum range in ACM2. There is a strong overprediction bias in ACM2 since FBI is greater than 1, and for QNSE, rainfall is underpredicted. QNSE shows better performance for very heavy rainfall detection, though it tends to underpredict. In moderate, rather heavy, and very heavy rainfall categories, both MYNN2 and MYNN3 show high false alarms, indicating poor discrimination capability. Overall, it is evident that ACM2 outperforms QNSE in predicting rainfall.

Table 4. Statistical performance indices for the rainfall forecast verification for Tauktae.

Bold values indicate the lowest error values.

4.2.2 Horizontal wind

Figure 7 represents the east–west cross-section of horizontal wind speed (in m/s) through the center of the storm at 12 UTC on 17 May. MYJ, QNSE, and ACM2 all show stronger and deeper wind cores (~85 m/s) that extend from the surface to approximately 400 hPa, similar to the very severe cyclonic storm stage. YSU and MYNN2 fail to illustrate wind speeds accurately; they show weaker cores (less than 50 m/s) and a smaller vertical extent. MYNN3 provides a core that is slightly stronger (up to 68 m/s), but it is inconsistent and has less mid-level stability, which causes structural imbalance. QNSE and ACM2 show more organized and vertically aligned cores, with a symmetrical eyewall structure, which provides an actual representation of a cyclone. Additionally, MYJ has a strong core, but the top levels are tilted slightly to the west, which causes a slight displacement error. YSU and MYNN schemes show weaker and tilted structures, which correspond to their poorer track and intensity performance. Schemes with stronger, deeper, and more symmetric wind cores (QNSE and ACM2) also perform better in track accuracy, translation speed, and intensity (MSW and MCP). By contrast, the weaker and misaligned cores in the YSU and MYNN schemes explain their underestimation of intensity and higher track errors. Combined with the previous studies by Vijaya Kumari et al. (2019) and Chen and Marks (2024), the boundary layer parameterizations strongly influence low-level inflow, vortex organization, and cyclone intensity through their control of moisture transport. This is consistent with the stronger, deeper, and more vertically coherent wind cores simulated by ACM2 and QNSE, which contribute to their improved track and intensity forecasts.

4.2.3 Vertical distribution of relative humidity

The vertical profiles of relative humidity (RH) at 12 UTC on 17 May (Figure 8) indicate that different PBL schemes have significantly varying RH values. This shows the importance of moisture distribution throughout the cyclone core. In the lower troposphere (1000–700 hPa), ACM2 has the highest near-saturation RH (approximately 80–90%), which is consistent with a well-developed moist boundary layer that allows for strong convection and realistic cyclone development. MYNN2 and YSU have substantially drier conditions (approximately 40–60%), which may make convection weaker and cyclones less potent. MYJ, QNSE, and MYNN3 give intermediate RH values (approximately 60–75%), which means these schemes are good at capturing low-level moisture. In the mid- troposphere (700–400 hPa), MYJ, QNSE, and ACM2 show increased RH (>60%) through the mid-levels. These findings are consistent with continuous deep convection and vertical moisture transfer. YSU and MYNN2 show considerable loss of moisture in this layer (<40–50%). Compared to MYNN2, MYNN3 holds more moisture; it is still drier than ACM2. At the top of the troposphere (400–100 hPa), the majority of models show a sudden decrease in RH over 300 hPa. QNSE and ACM2 keep the relative humidity (RH) higher up to approximately 200 hPa, resulting in better-organized deep convection and consistent with their stronger and more symmetric wind cores. The increased RH throughout in ACM2 and QNSE matches their better simulation of vertical wind structure and stronger intensity (MSW or MCP). The drier biases in YSU and MYNN2 are associated with poorer prediction of storm strength and vertical wind fields. MYJ and MYNN3 maintain a balance, although they fail to perform better than ACM2 and QNSE. Similar results have been reported by Gopalakrishnan et al. (2013) and Vijaya Kumari et al. (2019), showing that the PBL schemes preserve higher mid-level humidity, favor stronger convection, enhance cyclone structure, and provide more realistic intensity forecasts. This explains the better performance of ACM2 and QNSE in reproducing the moisture distribution and the intensity of cyclone Tauktae.

5 Conclusion

This study evaluated the impact of six planetary boundary layer (PBL) schemes on the simulation of ESCS Tauktae that developed over the Arabian Sea in 2021. The following conclusions can be drawn from this study. The results demonstrated that the PBL schemes control the cyclone intensity and track.

• The climatology of rapid intensification (RI) and rapid dissipation (RD) locations from 1990 to 2022 suggests that RI and RD locations over the Arabian Sea have increased from 2012 to 2022. The results suggest that the concentration of several RI locations is situated within or near the offshore 300-km area, suggesting that environmental conditions in this coastal-oceanic zone may be relevant to cyclone intensification.
• The analysis of daily mean track errors confirms that QNSE and ACM2 provide the most accurate track predictions for cyclone Tauktae, while MYNN2 and MYNN3 exhibit substantial deviations, especially after Day 3. The daily mean track errors are approximately 167 km, 184 km, 104 km, 203 km, 216 km, and 133 km for YSU, MYJ, QNSE, MYNN2, MYNN3, and ACM2, respectively. QNSE is the most accurate scheme for landfall prediction (115 km, +2 h), followed by ACM2. Even though YSU and MYJ show large errors and delayed landfall, both MYNN2 and MYNN3 suffer from the largest delays of more than 29 h, with position errors of more than 290 km. These results underscore the importance of selecting appropriate PBL schemes to improve cyclone track forecasts and landfall prediction over the Arabian Sea.
• The intensity is overestimated in all schemes after 48 h. However, the QNSE and ACM2 schemes show better alignment with the IMD pattern. Moreover, the statistical analysis of MSW and MCP shows that the ACM2 scheme has the lowest RMSE, SD, and NMSE. The majority of the models underestimate the weakening phase, showing a slower reduction compared to IMD. ACM2 and QNSE better capture the peak translation near landfall. MYNN2 and MYNN3 tend to underestimate translation speed in the later stages.
• The vertical wind structure analysis confirms that QNSE and ACM2 provide realistic simulations of Tauktae’s inner-core dynamics, reproducing both the magnitude and depth of winds. MYJ shows moderate skill, while YSU and MYNN schemes substantially underestimate cyclone strength and structural organization. The RH profiles reveal that moisture availability is a key discriminator among PBL schemes. ACM2 and QNSE capture a deep, moist core favorable for cyclone intensification, whereas YSU and MYNN2 simulate excessively dry mid- and lower-tropospheric conditions, which may weaken convection and intensity.
• The rainfall verification results indicate that the PBL schemes show varying skill across different rainfall categories. QNSE shows the best performance in detecting moderate and very heavy rainfall events, with relatively higher POD and PAG values and lower FAR. ACM2 demonstrates comparatively better skill in predicting rather heavy rainfall events and shows consistent performance across all rainfall categories. In contrast, MYNN2 and MYNN3 exhibit high false alarm rates, indicating reduced reliability in rainfall prediction. Overall, ACM2 shows the most balanced PBL scheme for forecasting rainfall intensity during cyclone Tauktae, followed by QNSE.

Overall, the results suggested that employing appropriate physics parameterization schemes is required to understand the intensity, track, and inner-core structure of Arabian Sea TCs using a high-resolution (1–5 km) modeling system.

Generative AI statement

The author(s) declared that Generative AI was not used in the creation of this manuscript.

Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.

Publisher’s note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

Footnotes

Edited by: Qing Bao, Chinese Academy of Sciences (CAS), China

Reviewed by: Pradeep Kumar, Manipal University Jaipur, India

Uma Ganesan, Indian Institute of Technology Madras, India

Footnotes

http://www.cawcr.gov.au/projects/verification/

https://giovanni.gsfc.nasa.gov/giovanni/