A New Approach to Generate Dynamical Cone of Uncertainty for Cyclone Track Forecast over North Indian Ocean

doi:10.21203/rs.3.rs-2885247/v1

Download PDF

Research Article

A New Approach to Generate Dynamical Cone of Uncertainty for Cyclone Track Forecast over North Indian Ocean

https://doi.org/10.21203/rs.3.rs-2885247/v1

This work is licensed under a CC BY 4.0 License

Journal Publication

published 07 Dec, 2023

Read the published version in Natural Hazards →

You are reading this latest preprint version

The North Indian Ocean (NIO) tropical cyclones (TCs) are devastating multi-hazard disasters with associated gale wind, torrential rainfall and storm-surge which pose severe threat to humankind and responsible for huge economic loss and livelihood. Predictions of track, intensity and structure along with the track uncertainty are very important to produce forecast and warning for the populace and disaster managers. India Meteorological Department (IMD) issues operational track forecasts and associated uncertainty as a Cone of Uncertainty (CoU) based on the climatological track errors computed from past operational forecasts. Due to large variation between different TCs, instead of static, dynamic CoU (DYN-CoU) based on dynamical Ensemble Prediction Systems (EPS) is more sophisticated way to provide track uncertainty.

In this study, the utility of different CoU including (i) Climatological CoU (CLM-CoU) (ii) CoU from dynamical EPS using different methods and (iii) Hybrid CoU (utilizing CLM-CoU & DYN-CoU) are analyzed with respect to various characteristics of 13 TCs formed during 2019–2020. The result suggests that the CLM-CoU is more skillful during the first 36 hours of forecast whereas DYN-CoU based on weights assigned to different members of EPS performs better during longer lead time. Basin-wise, CLM-CoU performs better over the Arabian Sea for all the forecast hours whereas in case of Bay of Bengal during first 18 hours and DYN-CoU shows better skill with lead time longer than 18 hours. For recurving TCs, CLM-CoU performs better up to 96 hour forecast lead. However, skill of DYN-CoU in case of recurving TCs is dependent on the operational and EPS track forecast performance.

Ensemble Prediction System

Tropical cyclone and Dynamical Cone of Uncertainty

Tropical cyclones (TC) destruction can be minimised by providing improved storm’s position (location) and track forecast to general public and disaster managers. Owing to improvement in the forecasting techniques and dynamic model evolution forecast track of tropical cyclones have considerable improved in the past decades (Heming and Goerss, 2010; Mohapatra et al., 2013; Mohapatra and Sharma, 2019; Heming et al. 2019). These improvements have helped disaster managers and general public to mitigate the losses due to Cyclonic Disturbances (TCs). However, still there are unavoidable uncertainties with respect to track forecast issued by different forecasting agencies of the world. These uncertainties can be attributed to the inherent predictability bound that seems to be unbeatable by future improvements in numerical models and in forecasting techniques (Fraedrich et al. 1989; Plu 2011; Landsea & Cangialosi, 2018). In this scenario, estimation of cyclone track forecast uncertainty becomes very important for disaster risk managers and public authorities to fine tune their actions. At present India Meteorological Department (IMD) issues the cyclone forecast track uncertainty estimation for North Indian Ocean (NIO) in graphical format to be used by general public, media and public agencies (Mohapatra et al. 2015; Mohapatra et al.2017). The graphical format is a widely used format for dissemination of tropical cyclone track uncertainty information due to its telegenic properties (Lundgren and McMakin 1998; Cox et al. 2013). The CoU based on past forecast performance do not appropriately represent the uncertainty in forecast for specific TC cases. Each TC is unique with different characteristic and thus challenges in forecasting the tracks also varies case by case. Therefore, to overcome the limitations of static CoU based on climatological errors, in this paper different methods to generate Dynamical Cone of Uncertainty (CoU) are explored utilising Ensemble Prediction Systems (EPS) over NIO. This study aims to figure out the better methodology which can be used in real-time operations to provide the uncertainty information of track in the form of cone of uncertainty. A comparative study among CoUs obtained using different methods from EPS and CoU utilising climatological errors is done to propose better candidate to represent tropical cyclone track forecast uncertainty over NIO, over individual basins of NIO (Arabian Sea and Bay of Bengal) and for straight moving and recurving tropical cyclones.

The cone of uncertainty is known by different names throughout the globe such as “cone of error”, “cone of probability” and “cone of death” and is used to represent the uncertainty and expected errors in the forecast track of the tropical cyclones issued by different World meteorological Organisation (WMO) designated Regional Specialised Meteorological Centre (RSMCs) and Tropical Cyclone Warning Centers (TCWCs) for TCs (Broad et al. 2007; Ruginzki et al. 2016, Heming et al. 2019). Most of these centre’s issue CoU around their official track forecasts using predictive skills of past years. For each forecast lead time Miami, Florida builds CoU taking radius as 67% of past several seasons direct position errors (DPE) whereas Tokyo, Japan uses 70% of the climatological DPE as the radius of CoU (Dupont et al. 2011; Heming et al. 2019). The Joint Typhoon Warning Centre (JTWC) in Hawaii produces CoU for TCs over Pacific Ocean and Indian Ocean, whose radii are the sum of the climatological average DPE and the predicted 34-knot wind radius (Dupont et al. 2011; Heming et al. 2019). Bureau of Meteorology (BOM), Australia uses CoU such that there is approximately 80% likelihood of cyclone track remaining within it, during the following 72 hours (Heming et al. 2019; Tropical Cyclone Service Level Specification 2021–2022). IMD also considers 5 years average DPE as the radius of CoU, thus accommodating 67% of total cases lying within the CoU. In 2011, RSMC La Réunion (Météo-France) introduced CoU based on spread information provided in the ensemble forecasts from the ECMWF (Dupont et al. 2011; Heming et al. 2019). As per National Hurricane centre information of 2022, the tropical cyclones can be expected to be within the cone of uncertainty roughly 60–70% of the time based on previous 5 years average DPE (website www.nhc.noaa.gov).

India Meteorological Department (IMD), which is the WMO designated RSMC for TCs over NIO, is the nodal agency to monitor and predict TCs over NIO. IMD provides the cyclone track forecast based on consensus from different numerical weather prediction (NWP) models and different remotely sensed and conventional observations (Mohapatra et al. 2012 & 2017). IMD introduced the CoU in TC track in 2009, prior to which it was providing the probable area of the landfall between two coastal stations (Mohapatra et al. 2012). The CoU forecasted by IMD is static in nature, similar to most of the different tropical cyclone forecasting centers across the world. In 2009, tropical cyclone forecast was issued by IMD for 72 hours (before 2009 IMD was issuing 24 hours forecast) with CoU based on five-year average errors of official track forecasts of IMD for 12 and 24 h forecasts and for 36–72 h forecasts, five-year average track forecast error of Quasi Lagrangian Model (QLM) run in IMD was used as earlier forecasts were not available (Mohapatra et al. 2017). The forecast and corresponding CoU were issued for 120 hours from 2013 onwards. CoU for 72 hours was based on the past 5 year’s climatological errors whereas for 84 hours onwards up to 120 hours CoU was calculated by extrapolating the CoU errors (Mohapatra et al. 2017). For constructing CoU the uncertainty in the track forecast is presented with circles drawn on the forecasted position of the storm. The radii of circles are dependent on the climatological performance of operational forecasts of tropical cyclone tracks (average of past 5 years operational direct position errors is used as CoU radii). The CoU is then drawn by joining the outer boundaries of these circles by a tangential line (Mohapatra et al. 2017). The vertex of the cone represents the present location of the center of the storm (Mohapatra et al. 2017). The study of CoU based on 2014 as used by IMD was found to contain the tropical cyclone within the cone 70–80% of the time (Mohapatra et al. 2017). These CoUs convey very much essential information for the benefit of disaster managers and help them in decision making particularly regarding evacuation. Now the CoU is revised periodically, generally once in five years, to reflect the track forecast accuracy (Mohapatra et al. 2017). Accordingly, the COU radii have been revised in 2014 and 2019 (IMD, 2021).

The CoU based on past forecast performance i.e. by giving uncertainty information from past errors do not appropriately represent the uncertainty in forecast for specific TC cases. Each tropical cyclone has different characteristic and thus offers challenges in terms of forecasting the tracks. Some TCs may be following the well-defined case studies and some may be completely deviating by showing curvatures, fast movements, stagnant movements etc. Therefore, COU based on past forecasting errors remain same for all cyclones for a basin under consideration for a year thereby, which is far from realistic picture of forecast uncertainty on case-by-case basis (Mohapatra et al. 2017). Therefore, it is required to provide case specific uncertainty information to the disaster manager for effective management of resources and evacuation decisions. The static CoU fails miserably in case of the recurving cyclones (Mohapatra et al. 2015). It is needless to mention that about 20–30% of the tracks over Bay of Bengal/Arabian Sea are recurving in nature (Mohapatra et al. 2013).

In this scenario, dynamic CoU becomes necessity for better disaster management and preparedness. To generate dynamic CoU, dynamic ensemble forecasting can play an important role in apprehending the situation dependent uncertainty in the forecasts of tropical cyclones ( Dupont et al., 2011; Yamaguchi et al., 2012; Du et al., 2016; Leonardo and Colle, 2017; Zhang and Yu, 2017). Dynamical track forecast based on EPS is case dependent and their track as well as ensemble spread changes for a tropical cyclone for each initial condition for which model is run (Majumdar & Finocchio 2010, Mohapatra et al. 2017). Therefore, CoU constructed from EPS changes each time initial condition of the model is changed. Here an attempt is made to study dynamical cone of uncertainty for NIO based on two EPS based forecast uncertainty consensus. An attempt has been made to formulate different methods to construct dynamic CoU for use in real-time. The comparative evaluation is made for the different candidates of dynamical CoU and the CoU issued by IMD based on climatological error statistics.

The climatological CoU (CLM-CoU) considered in the study is taken from the IMD’s operationally used CoU (Fig. 1) since 2019 as shown in Table 1(IMD, 2021). The improvement in forecasting skill has led to reduction in CLM-CoU regularly from its inception and based on past studies it is found that 70%-80% times the observed track is found to be within the CLM-CoU for 120-hour forecasts (Mohapatra et al. 2017).

Table 1

Radius of circle used to construct CoU based on climatological error statistics
Forecast Period (Hours)	Radius of the circle to construct cone of uncertainty (nm) w.e.f. 2019
0	10
6	20
12	30
18	40
24	45
36	55
48	70
60	85
72	95
84	115
96	130
108	145
120	160

The studies have revealed TC forecast from multicenter global ensembles have more skills than the best single global model in most of the cases (Yamaguchi et.al. 2015, 2017). Therefore, in the present study the Dynamical CoU is constructed based on the ensemble spread of two real time EPS systems available in real-time for operational use in IMD viz. Global Ensemble Forecasting System (IMD-GEFS) and NCMRWF Global Ensemble Prediction System (NEPS-G). The IMD-GEFS model based on National Centers for Environmental Prediction’s Global Ensemble Forecast System (NCEP GEFS) model (Zhou et al., 2017) is run at a horizontal resolution ~ 12 km with 21 ensemble members (Deshpande et al. 2020, 2021). The NEPS-G model is adapted from Global and Regional Prediction Systems (MOGREPS) of Met Office (Bowler et al., 2008) and is run with a horizontal resolution of 12 km with 22 + 1 members lagged ensemble at National Centre for Medium Range Weather Forecasting (NCMRWF), India (Sarkar et al. 2016). While constructing the CoU it is assumed that the observed position of TCs has the same probability to be seen anywhere inside CoU. Therefore, to make the dynamical CoU realistic in size for real-time operational usage with aim to match the current performance of TCs track lying within CoU for 60–70% times, 67% (approx. 2/3rd ) of the spread in dynamical CoU is considered as radii for constructing dynamical CoU.

Different methods as given below are used to extract the uncertainty information from the two EPS under consideration.

In the first experiment DYN-CoU-1, the latitude/longitude positions of extreme members (left extreme and right extreme) of IMD-GEFS and NEPS-G model are identified for each forecast lead time. The mean Latitude/Longitude position of left extreme and right extreme members of IMD-GEFS and NEPS-G members are calculated for each forecast lead time. The 2/3rd value of the distance between the mean left extreme and mean right extreme is considered as the radius for constructing dynamic CoU for each forecast lead time.

In the second experiment DYN-CoU-2, the latitude/longitude positions of extreme members (left extreme and right extreme) of IMD-GEFS and NEPS-G model are identified for each forecast lead time. The distance between extreme members of IMD-GEFS and NEPS-G models respectively is calculated for each forecast hour. The mean distance between extreme members of two EPS under consideration is calculated for each forecast hour. The 2/3rd value of this mean distance is considered as the radius for constructing dynamic CoU for each forecast lead time.

In the third experiment DYN-CoU-3, the latitude/longitude positions of extreme members (left extreme and right extreme) and mean position of IMD-GEFS and NEPS-G model are calculated for each forecast lead time. The distance between left and right extreme members of IMD-GEFS and NEPS-G models respectively is calculated from mean position for each forecast hour. The number of ensemble members lying on the left and right side of mean positions are identified for the two EPS for each forecast hour. The weighted mean between the left and right extreme members is calculated using weights based on the proportion of the number of ensemble members on the left or right side of the mean position for the two EPS for each forecast hour. The average value from the weighted mean derived from the two EPS is calculated and its 2/3rd value is considered as the radius for constructing dynamic CoU for each forecast lead time.

The three more experiments are designed using the hybrid radius from climatological CoU and dynamical CoU. For this purpose experiment, HYD-COU-1 is the simple mean of the radius of CoU from CLM-COU & DYN-COU-1, HYD-COU-2 is the simple mean of the radius of CoU from CLM-COU & DYN-COU-2 and HYD-COU-3 is the simple mean of CLM-COU and DYN-COU-3 respectively.

For verifying the CoU predictions, 13 cyclone cases are used in this study from 2019-to 2020 as shown in Fig. 2. The 13 cyclones considered in this study are BULBUL, FANI, PABUK, HIKAA, MAHA, KYARR, PAWAN, VAYU in 2019, and GATI, NISARGA, AMPHAN, NIVAR, and BUREVI from 2020. Out of which 8 Cyclones are from 2019 and 5 from 2020. In total 128 forecasts are examined out of which 53 forecasts are from the Bay of Bengal (BoB) basin and the remaining 75 forecasts are from the Arabian Sea (AS) basin. The following steps are performed for each cyclone forecast examined with respect to CoU.

To verify the usefulness of CoU in conveying the uncertainty information in real time, the direct position error for the operational forecasts is verified to lie within the CoU bounds for most of the cases. Also, the CoU so aimed should not be un-realistically too big to be un-actionable. For verification purposes the DPE is calculated for 120-hour operational forecasts for all the time on which EPS forecasts were available during 2019–2020 cyclone periods. DPE is measured by distance between the IMD’s operational forecast position and the best track position issued by IMD for the corresponding lead time. The percentage of the observed best track position lying within the CoU for different experiments is calculated. The percentage difference of the DPE from the different CoU derived from different experiments as described above are calculated using the formula for different forecast hours:

$$\left[\left\{\frac{\left(CoU-DPE\right)}{DPE}\right\}\times 100\right]$$

The mean and standard deviation of percentage difference for different forecast hours is calculated for NIO as a whole and for respective BoB and AS basins. The percentage CoU difference is also analysed for the TC exhibiting re-curvatures during their lifetime.

Figure 3 shows the accuracy in the COU derived from different experiments based on forecasts for TCs during 2019 & 2020. The DYN-COU-1 shows that observed track lies within this CoU for more than 80% case up to 72 hours forecast and for more than 70% cases for forecast from 84 hours to 120 hours lead time. The climatological CoU and HYB-CoU-2 shows accuracy of more than 60% up to 84 hours & 96 hours forecast lead time respectively. These accuracy figures suggest that the DYN-CoU-1 is a better candidate for prediction of CoU. However, it is also to be seen that the CoU should not be unnecessarily and un-realistically too big to become un-actionable. Therefore, CoU should represent the uncertainty of the forecast, but should not cover the entire spectrum of probable forecast values.

The accuracy of different COU forecasts has been analysed considering TCs over NIO as a whole, basin of formation (BOB or AS), with re-curving tracks. The results are presented and discussed in the following subsections. The mean percentage difference (MPD) of CoU with respect to DPE and its standard deviation for NIO as a whole and AS and BoB basin are shown for forecast hours of 00 hr, 06 hr, 12 hr, 18 hr, 24 hr, 36 hr, 48 hr, 60 hr, 72 hr, 84 hr, 96 hr, 108 hr and 120 hrs in Fig. 4 (a) to Fig. 4(c).

The MPD of the DPE and different CoU over NIO as a whole show that the CLM-CoU difference is the least positive with the lowest standard deviation for the first 36 hours of the forecast. Whereas, from 48 hours forecast to 120 hours forecast (except 84-hour forecast) DYN-COU-3 experiment giving weights to the number of ensemble members shows the least positive variation with the lowest standard deviation.

If the Arabian Sea basin is considered individually, then the CLM-CoU consistently shows the least positive percentage difference along with standard deviation with DPE for all the forecast hours.

In the case of BoB basin individually, for 06 hours forecast the CLM-CoU shows the least positive difference with DPE and standard deviation; however, for 12 hours forecast CLM-CoU and DYN-CoU-3 shows an almost similar amount of positive difference with DPE, but the standard deviation is considerably less for CLM-CoU. For 18 hours forecast, the least positive difference is with respect to DYN-CoU-3 but its standard deviation is very high. Whereas the HYB-CoU-3 also shows a lesser positive difference but its standard deviation is high. For CLM-CoU the positive difference is less as well as it has the lowest standard deviation for 18th -hour forecast. From 24 hour forecast to 72 hour forecast, DYN-CoU-3 clearly shows the least amount of positive percentage difference with DPE with the least amount of standard deviation. HYB-CoU-3 shows the percentage difference − 1.0%, + 3.22 which are very close to 0 for 84and 96 hours percentage with the lowest standard deviation. For 108-hour forecasts, there are mixed results, where the least positive percentage difference is shown by HYB-CoU-2 but with a large standard deviation, Whereas, the DYN-CoU-3 is showing − 32% difference and least standard deviation. HYB-CoU-3 is showing − 7.5% difference with the second lowest standard deviation. For 120-hour forecasts, the least positive percentage difference is seen in CLM-CoU but the standard deviation is very high. HYB-CoU-1 shows second-lowest positive percentage with moderately high standard deviation whereas DYN-CoU-1 shows moderate positive percentage above DPE of 16% with the least standard deviation.

It is also seen that at initial 6–12 hour forecast lead time the MPD is maximum for all experiments, this is because the direct position error is very small/minimum for initial 6–12 hours operational forecasts, which makes its difference from CoU look very big in terms of percentage during initial forecast hours. Also, the percentage difference is maximum for DYN-CoU1 at all forecast lead times owing to the methodology which gives weights to the latitude/longitude positions for uncertainty in order to account for spread of ensemble.

The NIO basin also witnesses a number of TCs showing re-curvatures, where re-curvature is defined as a well-defined change of direction of movement from westerly to easterly and in rare cases from easterly to westerly. These re-curvatures are largely due to steering flows, their interactions with different airmasses, wind shear etc. For the period 2019–2020 under consideration, a total number of 7 TCs namely Fani, Vayu, Kyarr, Maha, Pawan, Nisarga, and Nivar witnessed re-curvature. The percentage difference between DPE and different CoUs under consideration for these cyclones for forecast hours of 00 hr, 06hr, 12 hr, 18 hr, 24 hr, 36 hr, 48 hr, 60 hr, 72 hr, 84 hr, 96 hr, 108 hr and 120 hrs are shown in Fig. 5.

The results with respect to tropical cyclones showing re-curvature are in general pointing to the superiority of CLM-CoU for predicting track forecast uncertainty up to 96 hours of forecast lead time. CLM-CoU consistently displays the lowest positive percentage as well as the least amount of standard deviation. For 108 hours forecast lead time, the CLM-CoU and DYN-CoU-3 are showing comparable candidature, with both showing similar positive differences and standard deviation. For 120 hours forecast, DYN-CoU-3 is displaying the least positive percentage difference with DPE and is also giving the least standard deviation. However, the skill of DYN-CoU depends on the performance of operational track forecast and EPS based track guidance. It is also seen that the operational track forecast errors are in general more than the straight moving tropical cyclones over NIO (Mohapatra et al. 2013; Osuri et al. 2013; Nadimpalli et al. 2020). The dynamic CoU generated based on EPS is centered around EPS track but when transformed to operational forecasted track in recurving cyclones leads to lower skills. Therefore, as per the current study CLM-CoU is a better candidate for disseminating track forecast uncertainty up to 96 hours and DYN-CoU-3 is better for forecast periods more than 96 hours up to 120 hours for recurving TCs.

The difference in the CoU statistics discussed above with respect to two basins namely Bay of Bengal (BoB) and Arabian Sea (AS) is largely related to difference in forecast skill of NWP models over these basins. The NWP models are in general having higher accuracy and forecast skill over BoB as compared to AS (Mohapatra et al. 2013; Osuri et al. 2013). This is largely due to the fact that the open sea area in the AS is largely a data sparse region as compared to the BoB. Dense network of observatories and cyclone detection radars are absent along the west coast of India and along the coasts of other country sharing the Arabian sea coastline. On contrary, BoB coast is surrounded by dense network of observatories and entire Indian east coast is covered by cyclone detection radars (Mohapatra et al. 2013). The availability of observations and radar detections lead to accurate determination of location and intensity of TCs over the BoB as compared to AS and thus helps in improving the initial conditions for TC formed over the BoB in NWP models. It is also found that the higher initial errors in location and intensity estimation of TCs in Arabian Sea leads to higher track forecast errors (Mohanty et al. 2010; Osuri et al. 2012) and thereby increases the uncertainty in the track forecasts. It is also found in studies that the NWP models understanding of the physical processes which are related to the recurvature of the TCs is still not developed. This in turn leads to lower skills of models in predicting the tracks of recurving TCs. One more reason of the more uncertainty over the AS than the BoB is due to the fact that the percentage of recurving TCs over AS (63%) are higher than that of TCS over the BoB (45%) (Mohapatra et al. 2013). The NWP models also face challenges in terms of understanding the multiple interactions among the TCs wind field with mid-latitude flows and steering anticyclones and this leads to increase in the uncertainty (Mishra et al. 2021).

In the following section forecasted CoU for two super cyclonic storms from NIO is verified. During the period under consideration, NIO witnessed two super cyclonic storm namely AMPHAN over BoB and KYARR over Arabian Sea. Kyarr had a recurving track where as Amphan was more or less a straight moving cyclone as shown in Fig. 6 respectively. The cone of uncertainty for different dynamic methods as explained above and climatological statistics-based methods were analysed. The Fig. 7(a) and & (b) shows the accuracy of forecast cone of uncertainty in percentage terms along with the average radii of cone of uncertainty taken from all the initial conditions of EPS models available from different dynamic methods explained above for super cyclonic storms Kyarr and Amphan respectively.

It can be seen from the accuracy of the forecasted cone of uncertainty that the DYN-CoU-1 is giving better results than all other CoUs. However, the radii values are too big to be realistic and are almost 4 times the radii as calculated from CLM-CoU. This makes it very much un-actionable and mitigation measures based on it would be also very much in efficient in terms of resource utilization. For initial forecast hours the CLM-CoU is showing good accuracy (80%) with least radii of CoU for KYARR recurving cyclone followed by DYN-CoU-2 showing better accuracy however, DYN-CoU-3 shows radii with comparatively better accuracy then CLM-CoU for long forecast hours. If we see the specific case of straight moving cyclone AMPHAN then it is seen that the forecast radii from the CLM-CoU is consistently more than the DYN-CoU-3. Radii from CLM-CoU are approximately 2–3 times of radii of DYN-CoU-3 for 96 hours to 120 hours forecast with both showing the 100% accuracy making DYN-CoU-3 more efficient in forecasting the uncertainty for large forecast hours. DYN-CoU-2 has shown better accuracy for most of the forecast hours for this particular case and its radii were also comparable to CLM-CoU up to 72 hours of forecast but it was more than that of DYN-CoU-3 radii.

Therefore, considering the size of cone of uncertainty and accuracy of forecast cone of uncertainty for cyclones showing recurvature, the CLM-CoU radii seem to be better up-to 72 hours of forecast lead time followed by DYN-CoU-2 and DYN-CoU-3. For re-curving cyclones this can also be understood as different members will be pointing towards different direction because of re-curvature signals giving more uncertainty which is reflected in the higher values of radii of CoU forecasted from EPS. For the straight moving cyclones it seems that there is better consensus among EPS models and ensemble members converge thereby giving small radii of cone of uncertainty for all the forecast lead times as compared to the CLM-CoU based CoU. Therefore, for straight moving cyclones the DYN-CoU-3 should be able to given better uncertainty picture and actionable inputs for the disaster managers as compared to CLM-CoU radii-based uncertainty.

In this study, different methods to construct Uncertainty circles around the official TC forecast track of IMD utilizing IMD-GEFS and NEPS-G ensemble prediction systems are described and assessed. The constructed COU forecast of TCs is analyzed with respect to the basin of formation of TCs (BoB, AS, and NIO as a whole) and the type of track of TCs (recurving type). The validation and comparison of dynamic CoU with climatological error statistics-based CoU is also performed. It is also found that the greater number of observed tracks position lying within CoU necessarily doesn’t make CoU better candidate for representing uncertainty, as it may be unrealistically amplifying the errors forecast and thus the cone of uncertainty. The conclusions of this study are

From the results, it can be inferred that for NIO as a whole the climatological CoU is a better candidate for the cone of uncertainty up to 36 hours of the forecast. However, from 48 hours to 120 hours (except the 84th hour) forecast the Dynamic COU based on weights given to a number of members (DYN-CoU-3) is better in representing the uncertainty.

For the Arabian Sea basin, the cone of uncertainty based on climatological error statistics (CLM-CoU) consistently outperforms the other derived CoU based on EPS and hybrid methods. This is largely due to the data sparse Arabian sea basin, lack of observatory network and lack of cyclone detection network along the Arabian sea coasts leads to inefficient initial condition which creates lot of uncertainty in the NWP model forecast. There is need to improve the model representation of the interaction between mid-latitude westerlies with the TCs and steering anticyclone by improving the initial and boundary conditions to reduce uncertainty in NWP model forecasts with respect to TCs tracks (Mishra et al. 2020).

For Bay of Bengal basin, the cone of uncertainty based on climatological error statistics (CLM-CoU) gives better representation of uncertainty for first 18 hours of forecast. For 24 hours to 72 hours forecast Dynamic COU based on weights given to a number of members (DYN-CoU-3) is better in representing the uncertainty. For 84 hours to 108 hours forecast lead time the mean of Climatological error statistics based CoU and dynamic COU based on weights given to a number of members represented as experiment HYB-CoU-3 in this study is better in representing the track forecast uncertainty. For 120 hours forecast, there were mixed results with DYN-CoU-1 showing some merit over others. The DYN-CoU-1 experiment uses the mean Latitude/Longitude of extreme EPS members to calculate the cone of uncertainty radii. The EPS based dynamic CoU performs better over Bay of Bengal than Arabian Sea due to the more percentage of straight moving TCs than Arabian Sea. The data availability is also more in Bay of Bengal basin and along the coasts leading to improved initial conditions for models which gives better TC track forecast.

For TCs that witnessed re-curvature during their lifetime, climatological error statistics based CoU (CLM-CoU) displayed better representation of forecast track uncertainty up to 96 hours, whereas, DYN-CoU-3 is better for forecast periods more than 96 hours up to 120 hours for recurving TCs. It is to be mentioned that in general operational track forecast errors are high for recurving tropical cyclones when compared to straight moving tropical cyclones over NIO (Mohapatra et al. 2013; Osuri et al. 2013; Nadimpalli et al. 2020). The dynamic CoU used in the present study are generated utilizing the track guidance of EPS. However, when this dynamic CoU is associated with the operational forecasted track in recurving cyclones leads to lower skills.

The following are the implications of the above study: -

The choice of CoU methods to be used in real-time will depend upon the forecast track. Based on the straight or recurving TC forecast track the dynamic CoU method as described above may be used. This forecast track will be majorly dependent on the output of NWP model. Therefore, improvement in NWP model forecast would further refine the CoU.

The choice of method of dynamic CoU would be dependent on NWP model track forecast therefore, it’s efficient implementation would require improvement in the physical processes and multiple interactions between different fields in the model to generate better track forecast.

The current study is based on two EPS whose outputs are available in IMD in near real-time for operational usage. More EPS members (TIGGE; Bougeault et al. 2010).) data spread and clustering approach may be used for the construction of dynamic CoU based on their real-time availability for operational usage and research purpose. This study aims to propose an initial strategy for constructing dynamical CoU which may be implemented for operational forecasting of TCs over NIO by IMD based on their merits in presenting the uncertainties. Further similar kind of study has to be extended for the structural uncertainties of quadrant wind radii.

Author’s Contribution

Akhil Srivastava: Data curation, Formal Analysis, Investigation, Software, Validation, Visualization, Writing original Draft. M. Mohapatra: Conceptualization, Resources, Supervision, Writing-review and editing. Ananda K Das: Methodology, computing resources, Writing-review and editing. Raghu Nadimpalli: Formal analysis, Software, Validation, Writing- review and editing. Suneet Dwivedi: Writing-review and editing.

Acknowledgements

The authors gratefully acknowledge the support of the Indian Institute of Tropical Meteorology (IITM) and the National Center for Medium-Range Weather Forecasting (NCMRWF) for providing the IMD-GEFS and NEPS-G ensemble prediction system data for tropical cyclones. We would also like to thank the Cyclone Warning Division (CWD) of India Meteorological Department for maintaining and providing the records and statistics with respect to the operational track forecasts issued by IMD over NIO.

Data Availability

The ensemble model outputs from GEFS and NEPS will be shared upon reasonable request from the third party. IMD climatological dataset including best track are freely available under www.rsmcnewdelhi.imd.gov.in.

Conflict of Interest

On behalf of all the authors, the corresponding author declare that there is no conflict of interest.

ORCID OF AUTHORS:-

Akhil Srivastava:- 0000-0002-6862-531X

M. Mohapatra:- 0000-0002-8009-198X

Ananda K. Das:- 0000-0002-8919-8070

Raghu Nadimpalli:- 0000-0001-7769-3165

Suneet Dwivedi:- 0000-0001-9377-7713

Bougeault P, Toth Z, Bishop C, Brown B, Burridge D, Chen DH, Ebert B, Fuentes M, Hamill TM, Mylne K, Nicolau J (2010) The THORPEX interactive grand global ensemble. Bull Am Meteorol Soc 91(8):1059–1072
Bowler NE, Arribas A, Mylne KR, Robertson KB, Beare SE (2008) The MOGREPS short-range ensemble prediction system. Q J Royal Meteorological Society: J atmospheric Sci Appl Meteorol Phys Oceanogr 134(632):703–722
Broad K, Leiserowitz A, Weinkle J, Steketee M (2007) Misinterpretations of the “cone of uncertainty” in Florida during the 2004 hurricane season. Bull Am Meteorol Soc 88(5):651–668
Cox J, House D, Lindell M (2013) Visualizing uncertainty in predicted hurricane tracks.International Journal for Uncertainty Quantification, 3(2)
Deshpande M, Kanase R, Krishna RPM, Tirkey S, Mukhopadhyay P, Prasad VS, Johny CJ, Durai VR, Devi S, Mohapatra M (2021) Global Ensemble Forecast System (GEFS T1534) evaluation for tropical cyclone prediction over the North Indian Ocean. MAUSAM 72(1):119–128
Deshpande M, Kanase R, Krishna RPM, Tirkey S, Mukhopadhyay P, Prasad VS, Johny CJ, Durai VR, Devi S, Mohapatra M (2021) Global Ensemble Forecast System (GEFS T1534) evaluation for tropical cyclone prediction over the North Indian Ocean. MAUSAM 72(1):119–128
Du Y, Qi L, Cao X (2016) Selective ensemble-mean technique for tropical cyclone track forecast by using time‐lagged ensemble and multi‐centre ensemble in the western North Pacific. Q J R Meteorol Soc 142(699):2452–2462
Dupont T, Plu M, Caroff P, Faure G (2011) Verification of ensemble-based uncertainty circles around tropical cyclone track forecasts. Weather Forecast 26(5):664–676
Fraedrich K, Leslie LM (1989) Estimates of cyclone track predictability. I: Tropical cyclones in the Australian region. Q J R Meteorol Soc 115(485):79–92
Heming J, Goerss J (2010) Track and structure forecasts of tropical cyclones. In Global Perspectives on Tropical Cyclones: From Science to Mitigation (pp. 287–323)
Heming JT, Prates F, Bender MA, Bowyer R, Cangialosi J, Caroff P, Coleman T, Doyle JD, Dube A, Faure G, Fraser J (2019) Review of recent progress in tropical cyclone track forecasting and expression of uncertainties. Trop Cyclone Res Rev 8(4):181–218
IMD (2021) Cyclone Warning in India: Standard Operation Procedure
Landsea CW, Cangialosi JP (2018) Have we reached the limits of predictability for tropical cyclone track forecasting? Bull Am Meteorol Soc 99(11):2237–2243
Leonardo NM, Colle BA (2017) Verification of multimodel ensemble forecasts of North Atlantic tropical cyclones. Weather Forecast 32(6):2083–2101
Majumdar SJ, Finocchio PM (2010) On the ability of global ensemble prediction systems to predict tropical cyclone track probabilities. Weather Forecast 25(2):659–680
McMakin AH, Lundgren RE (2018) Risk communication: A handbook for communicating environmental, safety, and health risks. John Wiley & Sons
Mishra K, Sharma M, Mohapatra M (2021) Performance of numerical weather prediction models in predicting track of recurving cyclone Vayu over Arabian Sea during June 2019. J Earth Syst Sci 130:25. https://doi.org/10.1007/s12040-020-01533-7
Mohanty UC, Osuri Krishna K, Routray A, Mohapatra M, Pattanayak Sujata 2010 Simulation of Bay of Bengal Tropical Cyclones with WRF model: Impact of initial and boundary condition;Marine Geodesy33294–314
Mohapatra M, Nayak DP, Bandyopadhyay BK (2012) Evaluation of cone of uncertainty in tropical cyclone track forecast over north Indian ocean issued by India Meteorological Department. Trop Cyclone Res Rev 1(3):331–339
Mohapatra M, Nayak DP, Sharma RP et al (2013) Evaluation of official tropical cyclone track forecast over north Indian Ocean issued by India Meteorological Department. J Earth Syst Sci 122:589–601. https://doi.org/10.1007/s12040-013-0291-1
Mohapatra M, Geetha B, Sharma M (2017) Reduction in uncertainty in tropical cyclone track forecasts over the North Indian Ocean.Current Science, pp.1826–1830
Nadimpalli R, Osuri KK, Mohanty UC et al (2020) Forecasting tropical cyclones in the Bay of Bengal using quasi-operational WRF and HWRF modeling systems: an assessment study. Meteorol Atmos Phys 132:1–17. https://doi.org/10.1007/s00703-019-00669-6
Osuri Krishna K, Mohanty UC, Routray A, Mohapatra M (2012) The impact of satellite-derived wind data assimilation on track, intensity and structure of tropical cyclones over the North Indian Ocean. Int J Rem Sens 33:1627–1165
Osuri Krishna K et al (2013) Real-Time Track Prediction of Tropical Cyclones over the North Indian Ocean Using the ARW Model. J Appl Meteorol Climatology 52(11):2476–2492
Plu M (2011) A new assessment of the predictability of tropical cyclone tracks. Mon Weather Rev 139(11):3600–3608
Ruginski IT, Boone AP, Padilla LM, Liu L, Heydari N, Kramer HS, Hegarty M, Thompson WB, House DH, Creem-Regehr SH (2016) Non-expert interpretations of hurricane forecast uncertainty visualizations. Spat Cognition Comput 16(2):154–172
Sarkar A, Chakraborty P, John PG, Rajagopal EN (2016) Implementation of unified model based ensemble prediction system at NCMRWF (NEPS). NMRF/TR/02
Sarkar A, Kumar S, Dube A, Prasad SK, Mamgain A, Chakraborty P, Ashrit R, Mitra AK (2021) Forecasting of tropical cyclone using global and regional ensemble prediction systems of NCMRWF: A review. MAUSAM 72(1):77–86
Tropical Cyclone Service Level Specification, 2021–2022. Season 2021–2022 Severe Weather Environmental Prediction Services. Retrieved Jun 2 (2022) from http://www.bom.gov.au/cyclone/tropical-cyclone-knowledge-centre/warnings/Tropical_Cyclone_Service_Level_Specification.pdf
Yamaguchi M, Nakazawa T, Hoshino S (2012) On the relative benefits of a multi-centre grand ensemble for tropical cyclone track prediction in the western North Pacific. Q J R Meteorol Soc 138(669):2019–2029
Yamaguchi M, Vitart F, Lang ST, Magnusson L, Elsberry RL, Elliott G, Kyouda M, Nakazawa T (2015) Global distribution of the skill of tropical cyclone activity forecasts on short-to medium-range time scales. Weather Forecast 30(6):1695–1709
Yamaguchi M, Ishida J, Sato H, Nakagawa M (2017) WGNE intercomparison of tropical cyclone forecasts by operational NWP models: A quarter century and beyond. Bull Am Meteorol Soc 98(11):2337–2349
Zhang X, Yu H (2017) A probabilistic tropical cyclone track forecast scheme based on the selective consensus of ensemble prediction systems. Weather Forecast 32(6):2143–2157
Zhou X, Zhu Y, Hou D, Luo Y, Peng J, Wobus D (2017) The NCEP global ensemble forecast system with the EnKF initialization. Weather Forecast 32(5):1989–2004

Download PDF

Journal Publication

published 07 Dec, 2023

Read the published version in Natural Hazards →

Reviewers agreed at journal
13 May, 2023
Reviewers invited by journal
13 May, 2023
Editor invited by journal
06 May, 2023
Editor assigned by journal
02 May, 2023
First submitted to journal
02 May, 2023

You are reading this latest preprint version

A New Approach to Generate Dynamical Cone of Uncertainty for Cyclone Track Forecast over North Indian Ocean

Status:

Journal Publication

Version 1

Abstract

Figures

1. Introduction

2. Need for Dynamic Cone of Uncertainty

3. Data and Methodology

4. Results and Discussion

5. Conclusion

Declarations

References

Status:

Journal Publication

Version 1