Dvorak Technique Research Articles

Bridging satellite missions: deep transfer learning for enhanced tropical cyclone intensity estimation

ABSTRACT Geostationary satellites are valuable tools for monitoring the entire lifetime of tropical cyclones (TCs). Although the most widely used method for TC intensity estimation is manual, several automatic methods, particularly artificial intelligence (AI)-based algorithms, have been proposed and have achieved significant performance. However, AI-based techniques often require large amounts of input data, making it challenging to adopt newly introduced data such as those from recently launched satellites. This study proposed a transfer-learning-based TC intensity estimation method to combine different source data. The pre-trained model was built using the Swin Transformer (Swin-T) model, utilizing data from the Communication Ocean and Meteorological Satellite Meteorological Imager sensor, which has been in operation for an extensive period (2011–2021) and provides a large dataset. Subsequently, a transfer learning model was developed by fine-tuning the pre-trained model using the GEO-KOMPSAT-2A Advanced Meteorological Imager, which has been operational since 2019. The transfer learning approach was tested in three different ways depending on the fine-tuning ratio, with the optimal performance achieved when all layers were fine-tuned. The pre-trained model employed TC observations from 2011 to 2017 for training and 2018 for testing, whereas the transfer learning model utilized data from 2019 and 2020 for training and 2021 for testing to evaluate the model performance. The best pre-trained and transfer learning models achieved mean absolute error of 6.46 kts and 6.48 kts, respectively. Our proposed model showed a 7–52% improvement compared to the control models without transfer learning. This implies that the transfer learning approach for TC intensity estimation using different satellite observations is significant. Moreover, by employing a deep learning model visualization approach known as Eigen-class activation map, the spatial characteristics of the developed model were validated according to the intensity levels. This analysis revealed features corresponding to the Dvorak technique, demonstrating the interpretability of the Swin-T-based TC intensity estimation algorithm. This study successfully demonstrated the effectiveness of transfer learning in developing a deep learning-based TC intensity estimation model for newly acquired data.

Detecting Tropical Cyclone from the basic overview of life cycle of Extremely Severe Cyclonic Storm, Tauktae

The Extremely Severe Cyclonic Storm (ESCS) Tauktae, which made landfall on the Gujarat coast on May 17, 2021, is discussed in the current study. The analysis is based on INSAT-3D and passive microwave (PMW) images, fo-cusing on the cyclone’s eye characteristics and intensity. The satellite imag-es and products are utilized to determine the cyclone's intensity and the spe-cific characteristics of its eye. The study showcases the variations in intensi-ty and eye features of the cyclone throughout its life span, providing insights into the process of tropical cyclone intensification. The paper is structured to cover the methods for analysis, an overview of ESCS Tauktae's life cycle, the regulation of intensity based on Dvorak's technique, intensity estimation us-ing ADT9.0 and comparison with SATCON method and Indian Meteorologi-cal Department (IMD) provided best track data. Furthermore, the formation of Tauktae's eye, timeframes for eye scenes, comparison with sea surface temperature data, and concluding thoughts are also discussed.

Estimating tropical cyclone intensity using dynamic balance convolutional neural network from satellite imagery

Accurate estimation of tropical cyclone (TC) intensity helps to understand the evolution of TCs throughout their life cycle and plays an essential role in mitigating TC impact. Although TC intensity estimation methods based on deep learning have made significant progress, the existing techniques do not apply good methods to overcome the intensity overestimation and underestimation problems caused by the unbalanced distribution of TC data. Therefore, we propose a dynamic balance convolutional neural network to overcome these issues. The model consists of two branches, one branch is the learning of the raw data, and the other is the learning of strong (weak) TCs that account for a few data samples. Finally, the model is dynamically adjusted by adaptive trade-off parameters, gradually from the learning of the raw data to the learning of strong (weak) TCs, thus reducing errors in underestimation (overestimation) of strong (weak) TCs. Furthermore, an attention mechanism is employed to obtain the correlation between channels to improve the accuracy of TC intensity estimation further. We used globally 1285 TC cases from 2003–2016 to train the model and globally 94 TC cases from 2017 as independent test data. The results showed that the root-mean-square error of TC intensity estimation was 8.32 kt, 35% lower than that of the advanced Dvorak technique and 26% lower than that of the deep learning method visual geometry group (VGG). For a subset of 482 samples (from East Pacific and Atlantic) analyzed with reconnaissance observations, a root-mean-square intensity difference of 7.95 kt is achieved. Finally, we explored the model’s feature learning process and the contribution of each component of the satellite image to the TC intensity estimation through model visualization.

A Lightweight Multitask Learning Model With Adaptive Loss Balance for Tropical Cyclone Intensity and Size Estimation

Accurate tropical cyclone (TC) intensity and size estimation are key in disaster management and prevention. While great breakthroughs have been made in TC intensity estimation research, there is currently a lack of research on TC size reflecting TC influence radius. Therefore, we propose a lightweight multi-task learning model (TC-MTLNet) with adaptive loss balance to simultaneously estimate TC intensity and size. Adaptive loss balance is utilized to solve the problem of inconsistent convergence speed of TC intensity and size estimation tasks. The model based on four 2-D convolutions, four 3-D convolutions and three fully connected layers takes up less computational and storage space and improves the accuracy of TC intensity and size estimation by sharing knowledge among multiple tasks. In addition, due to the imbalanced distribution of TC samples, with significantly few low-intensity and high-intensity TC satellite data, this phenomenon poses a great challenge to TC intensity and size estimation. So, we utilize the influence of nearby samples to calibrate the sample density to weight the loss function to enable the model to be generalized to all samples. The result shows that the root-mean-square error (RMSE) of TC intensity estimation is <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$\text{8.40}\,\text{kts}$</tex-math></inline-formula> , which is 33.5% lower than that of the Advanced Dvorak Technique (ADT) and 11.4% lower than that of the deep learning method (3DAttentionTCNet). The mean absolute error (MAE) of the TC size estimation is <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$\text{20.89}\,\text{nmi}$</tex-math></inline-formula> , which is a 16% reduction compared to the Multi-Platform Tropical Cyclone Surface Winds Analysis (MTCSWA).

Application of Empirical Orthogonal Function Analysis to 1 km Ensemble Simulations and Himawari–8 Observation in the Intensification Phase of Typhoon Hagibis (2019)

An empirical orthogonal function (EOF) analysis was performed for the inner core of Typhoon Hagibis (2019) in the intensification phase. The Himawari–8 geostationary infrared (IR) brightness temperature (BT) collocated at the Hagibis’s center was combined with the IR BT simulated by a radiative transfer model, with 1 km ensemble simulations conducted by an atmosphere model and the coupled atmosphere–wave–ocean model. The ensemble simulations were conducted under one control atmospheric initial condition and the 26 perturbed ones with two different oceanic initial conditions. The first four EOF modes showed symmetric and asymmetric patterns such as a curved band, cloud dense overcast, and eye pattern used in the classification of the Dvorak technique. The influence of ocean coupling on the modes appeared only in the early intensification phase but was relatively small compared to the difference from the Himawari–8 observations. While ocean coupling and different oceanic initial condition quantitatively affected the IR BT, the normalized amplitude for the first EOF mode did not become close to that of the Himawari–8 observation in the late intensification phase. The intensification rate in the late intensification phase was inconsistent between the simulation results and the estimate from the Himawari–8 normalized amplitude by multiple linear regression analysis.

An Evaluation of the Advanced Dvorak Technique (9.0) for the tropical cyclones over the North Indian Ocean

The Advanced Dvorak Technique (ADT) is used by tropical cyclone prediction centres around the world to accurately evaluate the intensity of tropical cyclones (TCs) from meteorological operational satellites. The algorithm development team has introduced new improvements to the objective ADT to further extend its capabilities and accuracy. A study has therefore undergone to evaluate the new edition of ADT (9.0) based on all the North Indian Ocean Tropical cyclones during 2018, 2019 and 2020 (Total 15 No.). It is found that ADT (9.0) performed well with the conformity of IMD’s best track T. No estimates. ADT is reasonably good in estimating the intensity for T ≥ 4.0 (VSCS to SuCS) and overestimate the intensity for T ≤ 3.5(CS/SCS).

Investigation of Machine Learning Using Satellite-Based Advanced Dvorak Technique Analysis Parameters to Estimate Tropical Cyclone Intensity

AbstractSeveral simple and computationally inexpensive machine learning models are explored that can use advanced Dvorak technique (ADT)-retrieved features of tropical cyclones (TCs) from satellite imagery to provide improved maximum sustained surface wind speed (MSW) estimates. ADT (version 9.0) TC analysis parameters and operational TC forecast center best track datasets from 2005 to 2016 are used to train and validate the various models over all TC basins globally and select the best among them. Two independent test sets of TC cases from 2017 to 2018 are used to evaluate the intensity estimates produced by the final selected model called the “artificial intelligence (AI)” enhanced advanced Dvorak technique (AiDT). The 2017–18 MSW results demonstrate a global RMSE of 7.7 and 8.2 kt (1 kt ≈ 0.51 m s−1), respectively. Basin-specific MSW RMSEs of 8.4, 6.8, 7.3, 8.0, and 7.5 kt were obtained with the 2017 dataset in the North Atlantic, east/central Pacific, northwest Pacific, South Pacific/south Indian, and north Indian Ocean basins, respectively, with MSW RMSE values of 8.9, 6.7, 7.1, 10.4, and 7.7 obtained with the 2018 dataset. These represent a 30% and 23% improvement over the corresponding ADT RMSE for the 2017–18 datasets, respectively, with the AiDT error reduction significant to 99% in both sets. The AiDT model represents a notable improvement over the ADT performance and also compares favorably to more computationally expensive and complex machine learning models that interrogate satellite images directly while still preserving the operational familiarity of the ADT.

Analog Ensemble Methods for Improving Satellite-Based Intensity Estimates of Tropical Cyclones

Accurate, reliable estimates of tropical cyclone (TC) intensity are a crucial element in the warning and forecast process worldwide, and for the better part of 50 years, estimates made from geostationary satellite observations have been indispensable to forecasters for this purpose. One such method, the Advanced Dvorak Technique (ADT), was used to develop analog ensemble (AnEn) techniques that provide more precise estimates of TC intensity with instant access to information on the reliability of the estimate. The resulting methods, ADT-AnEn and ADT-based Error Analog Ensemble (ADTE-AnEn), were trained and tested using seventeen years of historical ADT intensity estimates using k-fold cross-validation with 10 folds. Using only two predictors, ADT-estimated current intensity (maximum wind speed) and TC center latitude, both AnEn techniques produced significant reductions in mean absolute error and bias for all TC intensity classes in the North Atlantic and for most intensity classes in the Eastern Pacific. The ADTE-AnEn performed better for extreme intensities in both basins (significantly so in the Eastern Pacific) and will be incorporated in the University of Wisconsin’s Cooperative Institute for Meteorological Satellite Studies (UW-CIMSS) workflow for further testing during operations in 2021.

Physics-augmented Deep Learning to Improve Tropical Cyclone Intensity and Size Estimation from Satellite Imagery

AbstractA deep learning-based method augmented by prior knowledge of tropical cyclones (TCs), called DeepTCNet, is introduced to estimate TC intensity and wind radii from infrared (IR) imagery over the North Atlantic Ocean. While standard deep learning practices have many advantages over conventional analysis approaches and can produce reliable estimates of TCs, the data-driven models informed by machine-readable physical knowledge of TCs could achieve higher performance. To this end, two approaches are explored to develop the physics-augmented DeepTCNet: (i) infusing the auxiliary physical information of TCs into models for single-task learning; (ii) learning auxiliary physical tasks for multi-task learning. More specifically, augmented by auxiliary information of TC fullness (a measure of the radial decay of the TC wind field), the DeepTCNet yielded a 12% improvement in estimating TC intensity over the non-augmented one. By learning TC wind radii and auxiliary TC intensity task simultaneously, the model’s wind radii estimation skill is improved by 6% over only learning four wind radii tasks, and by 9% over separately learning a single wind radii task. The evaluation results showed that the DeepTCNet is in-line with the Satellite Consensus technique (SATCON) but systematically outperforms the Advanced Dvorak Technique (ADT) at all intensity scales with an averaged 39% enhancement in TC intensity estimation. The DeepTCNet also surpasses the Multi-platform Tropical Cyclone Surface Wind Analysis technique (MTCSWA) with an average improvement of 32% in wind radii estimation.

Revisions to the Australian tropical cyclone best track database

The Australian tropical cyclone (TC) best track database (BT) maintained by the Bureau of Meteorology has records since 1909 of varying quality and completeness. Since 2005 a series of efforts to improve the database have included: removing internal inconsistencies, adding fixes, and identifying errors using comparisons with other datasets; upgrading intensity information since 1973 including adding maximum winds (Vm) prior to 1984–85, rederiving Dvorak Current Intensity numbers from archived material and accounting for different wind–pressure relationships used; a partial reanalysis of satellite imagery including microwave imagery using the HURSAT dataset since 1987; and considering an objective intensity dataset. The BT homogeneity is reviewed in the context of improvements in satellite technology, observational coverage, scientific developments, BT procedures and the subjective variation between analysts across time and offices. The scale of these variances is greatest in the early stages prior to 1981 in the absence of geostationary satellite imagery until 1978, satellite calibration issues from 1978–80 and prior to the introduction of the enhanced infra-red Dvorak technique in 1981. The current era since 2003 is considered to be the most accurate, comprehensive and homogeneous corresponding to the expansion of the TC database to include the current suite of fields; the application of microwave and scatterometry imagery; greater standardisation of BT practices and slight changes in the application of the Dvorak technique. These improvements have generated a more consistent dataset that could be used for weather and climate research and other TC-related work.

PREDICTION OF CYCLONE USING KALMAN SPATIO TEMPORAL AND TWO DIMENSIONAL DEEP LEARNING MODEL

Cyclone Classification and Prediction models rely on large intensity based on the maximum speed of the wind, along with the classification of intensity. The computational constraints blended with the formation of those intensities, cyclone classification and prediction firmly depreciate the full range of optimal features required for classification and hence accurate representation is less possible. Keeping this point, our study inspects the potential of Spatio-temporal features using a machine learning algorithm as an alternative to the current study of cyclones. This method is called, Spatio-Temporal Kalman Momentum and Two Dimensional Deep Learning (SKM-2DDP) for cyclone classification and prediction. To start with, pre-processing is performed by applying the Kalman Momentum Conservation Filter mechanism based on the design of the Dvorak technique to obtain optimal estimates of state variables and reduce the computational burden involved to remove noise from input cyclone images. With the resultant denoised input cyclone images, Spatio Temporal Feature Extraction is performed. Features obtained from the inherent intrinsic properties of pre-processed cyclone images of several weather conditions result in successful classification. Followed by pre-processing, in this work, features constituting both pixel-wise intensities over time and the directional length are being considered. The dependency of autocorrelation with each pixel’s intensities over time and two temporal features are helped for coarse classification of weather conditions according to their visual effects. Finally, with the inherent intrinsic features extracted, a Two Dimensional Deep Learning model is utilized to foretell the cyclone intensity. Experimental evaluation of the proposed SKM-2DDP method is accomplished using images of cyclone dataset with many factors just as peak signal to noise ratio, prediction accuracy, prediction time and false-positive rate. Our own selves have considered with several cyclone images acquired from two different cyclone image datasets namely OCKHI_DEC2017 and VARDAH_DEC2016.

An Improved Conversion Relationship between Tropical Cyclone Intensity Index and Maximum Wind Speed for the Advanced Dvorak Technique in the Northwestern Pacific Ocean Using SMAP Data

The Advanced Dvorak Technique (ADT) uses geostationary satellite data to estimate tropical cyclone (TC) intensity owing to the difficulty in directly observing a TC’s internal structure. This study presents a new relationship (Hong and Ryu scale) between the current intensity (CI) number and estimated maximum wind speed (MWS) of TCs over the northwestern Pacific region; the CI number is the TC intensity index retrieved from the ADT. The Soil Moisture Active Passive (SMAP) with the L-band (1.4 GHz) microwave radiometer, is used to calibrate and produce the new Hong and Ryu scale for the ADT algorithm. Japan Meteorological Agency (JMA) best track MWS data, SMAP sea surface wind speed estimates, and ADT’s TC intensity data between 2015–2018 are spatiotemporally collocated for the calibration process. The CI number is derived from the Korea Meteorological Administration (KMA) operational ADT which uses the Koba scale to convert to the MWS for validation against the MWS of the best track. The conversion relationships between CI number and SMAP MWS, and between SMAP MWS and MWS of the best track a derived, and the MWS of two ADTs with the Koba and Hong and Ryu scales are then estimated using the same CI numbers with TC intensity data between 2015–2018. Finally, the MWS of the ADT with the Koba scale and the new ADT with the proposed Hong and Ryu scale are independently validated on best track data from 2013–2014. The MWS root mean square error (RMSE) is 4.39 m/s for the new ADT using the Hong and Ryu scale, which is lower than 4.77 m/s RMSE of the ADT using the Koba scale. Hence, the ADT using the Hong and Ryu scale can modestly improve the accuracy of TC intensity analysis in the northwestern Pacific region.

Constraints in Dvorak Wind Speed Estimates: How Quickly Can Hurricanes Intensify?

AbstractThe Dvorak technique is used operationally by meteorological agencies throughout the world for estimating tropical cyclone intensity and position. The technique consists of constraints that put a maximum threshold for which the final T-number, relating directly to intensity, can change during a certain time interval (6, 12, 18, and 24 h). There are cases when these constraints could be broken, especially during rapid intensification. This research tests whether the constraints used for intensity change are warranted or need to be changed. A database of cases with the largest intensity changes for 2000–17 Atlantic tropical cyclones was compiled. A reconnaissance or scatterometer “fix” is required within 3 h of both the beginning and ending of the period for each case to inform the best track and to be included for analysis. Dvorak classifications from the Tropical Analysis and Forecast Branch are noted for each case, which includes the initial and final T-numbers, current intensity numbers, and data T-numbers. Statistical parameters, including correlations, intensity errors, absolute intensity errors, root-mean-square errors, and significance tests are calculated and analyzed for each period. Results suggest that the T-number constraints for the 18- and 24-h periods could be increased to a 2.5 and a 3.0, respectively. However, results also suggest that the constraints for the 6- and 12-h time intervals should remain the same.

Towards an objective historical tropical cyclone dataset for the Australian region

The appropriate design of infrastructure in tropical cyclone (TC) prone regions requires an understanding of the hazard risk profile underpinned by an accurate, homogenous long-term TC dataset. The existing Australian region TC archive, or ‘best track’ (BT), suffers from inhomogeneities and an incomplete long-term record of key TC parameters. This study assesses mostly satellite-based objective techniques for 1981–2016, the period of a geostationary satellite imagery dataset corrected for navigation and calibration issues. The satellite-based estimates of Australian-region TCs suffer from a general degradation in the 1981–1988 period owing to lower quality and availability of satellite imagery.The quality of the objective techniques for both intensity and structure is compared to the reference BT 2003–2016 estimates. For intensity the Advanced Dvorak Technique algorithm corresponds well with the BT 2003–2016, when the algorithm can use passive microwave data (PMW) as an input. For the period prior to 2003 when PMW data is unavailable, the intensity algorithm has a low bias. Systematic corrections were made to the non-PMW objective estimates to produce an extended (1989–2016) homogeneous dataset of maximum wind that has sufficient accuracy to be considered for use where a larger homogeneous sample size is valued over a shorter more accurate period of record. An associated record of central pressure using the Courtney-Knaff-Zehr wind pressure relationship was created.For size estimates, three techniques were investigated: the Deviation Angle Variance and the ‘Knaff’ techniques (IR-based), while the ‘Lok’ technique used model information (ECMWF reanalysis dataset and TC vortex specification from ACCESS-TC). However, results lacked sufficient skill to enable extension of the reliable period of record. The availability of scatterometer data makes the BT 2003–2016 dataset the most reliable and accurate. Recommendations regarding the best data source for each parameter for different periods of the record are summarised.

A Study of the Interaction between Typhoon Francisco (2013) and a Cold-Core Eddy. Part I: Rapid Weakening

AbstractTyphoon Francisco (2013) experienced unusually rapid weakening (RW) with its maximum surface wind decreasing by 45 kt (1 kt ≈ 0.51 m s−1) over 24 h as measured from the satellite-derived advanced Dvorak technique (ADT) dataset, which is more than twice the weakening rate defined as RW by DeMaria. The mechanisms leading to the extreme RW event of Francisco are examined based on observational analysis and simulations by coupling the Weather Research and Forecasting (WRF) Model, version 3.7, with the Stony Brook Parallel Ocean Model (sbPOM). The RW of Francisco took place in a relatively favorable atmospheric environment while passing over detrimental oceanic conditions, dominated by the presence of a cold-core eddy. The passages of two prior typhoons apparently intensified the cold-core eddy, contributing to a major role of eddy feedback on RW for Francisco. The structural changes in Francisco accompanying eddy interaction are characterized by a substantially enlarged eye size, which evolved from ~20 to ~100 km in diameter, as indicated from satellite images. Numerical simulations suggest that the eddy is prominent in weakening the intensity of Francisco during the storm–eddy interaction, with its role less significant but still comparable to that of the cold wake. Both the cooler water and stronger upward motion in the eddy lead to a larger sea surface temperature decrease induced by Francisco, which results in a nearly 50% decrease of surface enthalpy flux, suppressed convective bursts, and a 50% reduction in latent heat release. These results underscore the potential importance of open-ocean, cold-core eddies in contributing to the RW of tropical cyclones.

The Advanced Dvorak Technique (ADT) for Estimating Tropical Cyclone Intensity: Update and New Capabilities

Abstract The advanced Dvorak technique (ADT) is used operationally by tropical cyclone forecast centers worldwide to help estimate the intensity of tropical cyclones (TCs) from operational geostationary meteorological satellites. New enhancements to the objective ADT have been implemented by the algorithm development team to further expand its capabilities and precision. The advancements include the following: 1) finer tuning to aircraft-based TC intensity estimates in an expanded development sample, 2) the incorporation of satellite-based microwave information into the intensity estimation scheme, 3) more sophisticated automated TC center-fixing routines, 4) adjustments to the intensity estimates for subtropical systems and TCs undergoing extratropical transition, and 5) addition of a surface wind radii estimation routine. The goals of these upgrades and others are to provide TC analysts/forecasters with an expanded objective guidance tool to more accurately estimate the intensity of TCs and those storms forming from, or converting into, hybrid/nontropical systems. The 2018 TC season is used to illustrate the performance characteristics of the upgraded ADT.

ANALISIS SIKLON TROPIS NOCK-TEN BERBASIS DATA SATELIT HIMAWARI

Siklon tropis Nock-Ten yang melintasi Filipina pada 25-26 Desember 2016 termasuk dalam kategori super taifun. Dalam skala meteorologi, siklon tropis termasuk dalam fenomena berskala sinoptik. Dampak yang ditimbulkan dapat memengaruhi kondisi atmosfer di wilayah Indonesia, terutama siklon tropis yang terbentuk di perairan Samudera Pasifik Barat Daya dan Samudera Hindia sebelah utara Australia. Oleh karena itu, perlu dilakukan pemantauan aktivitasnya karena berpengaruh pada dinamika atmosfer yang memicu angin kencang dan tumbuhnya awan hujan. Penelitian ini bertujuan untuk menganalisis pertumbuhan, pergerakan, danintensitas Taifun Nock-Ten dengan menggunakan data satelit Himawari 8 kanal IR1. Metode yang digunakan berupa analisis indeks konvektif dengan menggunakan data TBB (Temperature of black body) sebagai suhu puncak awan dan metode pengamatan visual berbasis teknik Dvorak untuk memantau pertumbuhan dan pergerakan siklon, dan untuk mendapatkan T-Number yang dapat digunakan untuk memprakirakan intensitasnya berupa kecepatan angin maksimum dan tekanan minimum pusat siklon. Berdasarkan analisis suhu puncak awan dari siklon tropis Nock-Ten didapatkan nilai indeks konvektif yang tinggi yaitu berkisar dari 49.9 hingga 74.4. Siklon tropis Nock-Ten memiliki masa hidup 8 hari, terbentuk di Samudera Pasifik sebelah utara Papua berupa MCS (Mesoscale Convective System) pada 20 Desember 2016 yang berpropagasi ke barat dan punah pada 28 Desember 2016. Intensitas maksimum terjadi pada 25 Desember 2016 pukul 00.00 UTC dengan T-Number mencapai 7.0 dengan nilai prakiraan kecepatan angin maksimum mencapai 140 knots dan tekanan udara minimum mencapai 898 hPa, setara dengan siklon tropis kategori 5 skala Saffir-Simpson.

Estimating Tropical Cyclone Intensity by Satellite Imagery Utilizing Convolutional Neural Networks

Abstract Accurately estimating tropical cyclone (TC) intensity is one of the most critical steps in TC forecasting and disaster warning/management. For over 40 years, the Dvorak technique (and several improved versions) has been applied for estimating TC intensity by forecasters worldwide. However, the operational Dvorak techniques primarily used in various agencies have several deficiencies, such as inherent subjectivity leading to inconsistent intensity estimates within various basins. This collaborative study between meteorologists and data scientists has developed a deep-learning model using satellite imagery to estimate TC intensity. The conventional convolutional neural network (CNN), which is a mature technology for object classification, requires several modifications when being used for directly estimating TC intensity (a regression task). Compared to the Dvorak technique, the CNN model proposed here is objective and consistent among various basins; it has been trained with satellite infrared brightness temperature and microwave rain-rate data from 1097 global TCs during 2003–14 and optimized with data from 188 TCs during 2015–16. This paper also introduces an upgraded version that further improves the accuracy by using additional TC information (i.e., basin, day of year, local time, longitude, and latitude) and applying a postsmoothing procedure. An independent testing dataset of 94 global TCs during 2017 has been used to evaluate the model performance. A root-mean-square intensity difference of 8.39 kt (1 kt ≈ 0.51 m s−1) is achieved relative to the best track intensities. For a subset of 482 samples analyzed with reconnaissance observations, a root-mean-square intensity difference of 8.79 kt is achieved.

The First International Workshop on Satellite Analysis of Tropical Cyclones: Summary of Current Operational Methods to Estimate Intensity

ABSTRACT This paper summarizes the results and findings from the first International Workshop on Satellite Analyses of Tropical Cyclones (IWSATC). Estimating tropical cyclone intensity in the absence of penetrating reconnaissance aircraft is mainly left to satellite-based remote sensing methods. The primary method used globally since the latter part of the last century is the Dvorak Technique. While this method has withstood the test of time, it is not without limitations, and has been subject to regional modifications by various national tropical cyclone analysis centers. Differences between agencies in both the application of the Dvorak Technique and in general operational procedures to derive final intensity estimates can lead to real-time warning conflicts, as well as issues with final best track values for climatological analyses. As coastal populations increase, the WMO recognizes these issues as potentially serious problems, and organized a workshop in 2011 to bring together tropical cyclone experts to 1) discuss current operational analysis practices, 2) learn about developing satellite techniques, and 3) come up with a set of recommendations to start down a path towards a global congruence on intensity estimation procedures.

An Objective Tropical Cyclone Intensity Estimation Model Based on Digital IR Satellite Images

ABSTRACT An objective tropical cyclone (TC) intensity estimation model is proposed based on the statistical relationship between TC intensity and its inner-core convection, plus the persistence of TC intensity. In the model, the inner-core convection is described by several parameters retrieved from digital infrared (IR) satellite images, including the number of convective cores (Num), their distance to the TC center, and their blackbody temperature (TBB), among others. The persistence of TC intensity is embodied by the TC intensity six hours previous (V6h). The model was set up by the stepwise regression technique using a five-year dataset (2006-2010) and was tested using an independent dataset covering 2011-2012, with V6h from the best-track dataset. Selected factors of the model included V6h, Num, Lat (TC center latitude), Lon (TC center longitude), DISmin (minimum distance between convective cores and TC center), and TBBdif (difference between the maximum and minimum TBB value of convective cores). Results showed that, for independent samples during 2011-2012, the MAE (mean absolute error) and RMSE (root-mean-square error) of Vmax estimation were 1.8 m s− 1 and 2.4 m s− 1, respectively. In order to make the model totally independent from the best-track dataset, the model estimation from six hours previous was used as the V6h for a second independent test covering 2011-2012. The results showed that the model had an MAE and RMSE of 5.4 m s− 1 and 7.3 m s− 1, respectively. Large errors were found for strong TCs (Severe Typhoon or Super Typhoon). The error statistics of the proposed model are comparable to published statistics on the widely used Dvorak technique or its objective versions, implying its potential to be used as an alternative tool for TC intensity estimation in either real-time operation or post-season best-track analyses.