Browse
Deep Learning for Typhoon Intensity Classification Using Satellite Cloud ImagesAbstract
A tropical cyclone, also known as a typhoon, is one of the most destructive weather phenomena. Its intense cyclonic eddy circulations often cause serious damage to coastal areas. Accurate classification or prediction for typhoon intensity is crucial to disaster warning and mitigation management. But typhoon intensity-related feature extraction is a challenging task as it requires significant preprocessing and human intervention for analysis, and its recognition rate is poor due to various physical factors such as tropical disturbance. In this study, we built a Typhoon-CNNs framework, an automatic classifier for typhoon intensity based on a convolutional neural network (CNN). The Typhoon-CNNs framework utilized a cyclical convolution strategy supplemented with dropout zero-set, which extracted sensitive features of existing spiral cloud bands (SCBs) more effectively and reduces the overfitting phenomenon. To further optimize the performance of Typhoon-CNNs, we also proposed the improved activation function (T-ReLU) and the loss function (CE-FMCE). The improved Typhoon-CNNs was trained and validated using more than 10 000 multiple sensor satellite cloud images from the National Institute of Informatics. The classification accuracy reached to 88.74%. Compared with other deep learning methods, the accuracy of our improved Typhoon-CNNs was 7.43% higher than ResNet50, 10.27% higher than InceptionV3, and 14.71% higher than VGG16. Finally, by visualizing hierarchic feature maps derived from Typhoon-CNNs, we can easily identify the sensitive characteristics such as typhoon eyes, dense-shadowing cloud areas, and SCBs, which facilitate classifying and forecasting typhoon intensity.
Abstract
A tropical cyclone, also known as a typhoon, is one of the most destructive weather phenomena. Its intense cyclonic eddy circulations often cause serious damage to coastal areas. Accurate classification or prediction for typhoon intensity is crucial to disaster warning and mitigation management. But typhoon intensity-related feature extraction is a challenging task as it requires significant preprocessing and human intervention for analysis, and its recognition rate is poor due to various physical factors such as tropical disturbance. In this study, we built a Typhoon-CNNs framework, an automatic classifier for typhoon intensity based on a convolutional neural network (CNN). The Typhoon-CNNs framework utilized a cyclical convolution strategy supplemented with dropout zero-set, which extracted sensitive features of existing spiral cloud bands (SCBs) more effectively and reduces the overfitting phenomenon. To further optimize the performance of Typhoon-CNNs, we also proposed the improved activation function (T-ReLU) and the loss function (CE-FMCE). The improved Typhoon-CNNs was trained and validated using more than 10 000 multiple sensor satellite cloud images from the National Institute of Informatics. The classification accuracy reached to 88.74%. Compared with other deep learning methods, the accuracy of our improved Typhoon-CNNs was 7.43% higher than ResNet50, 10.27% higher than InceptionV3, and 14.71% higher than VGG16. Finally, by visualizing hierarchic feature maps derived from Typhoon-CNNs, we can easily identify the sensitive characteristics such as typhoon eyes, dense-shadowing cloud areas, and SCBs, which facilitate classifying and forecasting typhoon intensity.
Hurricane Imaging Radiometer (HIRAD) Wind Speed Retrievals and Validation Using DropsondesAbstract
Surface wind speed retrievals have been generated and evaluated using Hurricane Imaging Radiometer (HIRAD) measurements from flights over Hurricane Joaquin, Hurricane Patricia, Hurricane Marty, and the remnants of Tropical Storm Erika—all in 2015. Procedures are described here for producing maps of brightness temperature, which are subsequently used for retrievals of surface wind speed and rain rate across a ~50-km-wide swath for each flight leg. An iterative retrieval approach has been developed to take advantage of HIRAD’s measurement characteristics. Validation of the wind speed retrievals has been conducted, using 636 dropsondes released from the same WB-57 high-altitude aircraft carrying HIRAD during the Tropical Cyclone Intensity (TCI) experiment. The HIRAD wind speed retrievals exhibit very small bias relative to the dropsondes, for winds of tropical storm strength (17.5 m s−1) or greater. HIRAD has reduced sensitivity to winds weaker than tropical storm strength and a small positive bias (~2 m s−1). Two flights with predominantly weak winds according to the dropsondes have abnormally large errors from HIRAD and large positive biases. From the other flights, the root-mean-square differences between HIRAD and the dropsonde winds are 4.1 m s−1 (33%) for winds below tropical storm strength, 5.6 m s−1 (25%) for tropical storm–strength winds, and 6.3 m s−1 (16%) for hurricane-strength winds. The mean absolute differences for those three categories are 3.2 m s−1 (25%), 4.3 m s−1 (19%), and 4.8 m s−1 (12%), respectively, with a bias near zero for winds of tropical storm and hurricane strength.
Abstract
Surface wind speed retrievals have been generated and evaluated using Hurricane Imaging Radiometer (HIRAD) measurements from flights over Hurricane Joaquin, Hurricane Patricia, Hurricane Marty, and the remnants of Tropical Storm Erika—all in 2015. Procedures are described here for producing maps of brightness temperature, which are subsequently used for retrievals of surface wind speed and rain rate across a ~50-km-wide swath for each flight leg. An iterative retrieval approach has been developed to take advantage of HIRAD’s measurement characteristics. Validation of the wind speed retrievals has been conducted, using 636 dropsondes released from the same WB-57 high-altitude aircraft carrying HIRAD during the Tropical Cyclone Intensity (TCI) experiment. The HIRAD wind speed retrievals exhibit very small bias relative to the dropsondes, for winds of tropical storm strength (17.5 m s−1) or greater. HIRAD has reduced sensitivity to winds weaker than tropical storm strength and a small positive bias (~2 m s−1). Two flights with predominantly weak winds according to the dropsondes have abnormally large errors from HIRAD and large positive biases. From the other flights, the root-mean-square differences between HIRAD and the dropsonde winds are 4.1 m s−1 (33%) for winds below tropical storm strength, 5.6 m s−1 (25%) for tropical storm–strength winds, and 6.3 m s−1 (16%) for hurricane-strength winds. The mean absolute differences for those three categories are 3.2 m s−1 (25%), 4.3 m s−1 (19%), and 4.8 m s−1 (12%), respectively, with a bias near zero for winds of tropical storm and hurricane strength.
Abstract
The High-Definition Sounding System (HDSS) is an automated system deploying the expendable digital dropsonde (XDD) designed to measure wind and pressure–temperature–humidity (PTH) profiles, and skin sea surface temperature (SST) within and around tropical cyclones (TCs) and other high-impact weather events needing high sampling density. Three experiments were conducted to validate the XDD.
On two successive days off the California coast, 10 XDDs and 14 Vaisala RD-94s were deployed from the navy’s Center for Interdisciplinary Remotely-Piloted Aircraft Studies (CIRPAS) Twin Otter aircraft over offshore buoys. The Twin Otter made spiral descents from 4 km to 60 m at the same descent rate as the sondes. Differences between successive XDD and RD-94 profiles due to true meteorological variability were on the same order as the profile differences between the spirals, XDDs, and RD-94s. XDD SST measured via infrared microradiometer, referred to as infrared skin SST (SSTir), and surface wind measurements were within 0.5°C and 1.5 m s−1, respectively, of buoy and Twin Otter values.
A NASA DC-8 flight launched six XDDs from 12 km between ex-TC Cosme and the Baja California coast. Repeatability was shown with good agreement between features in successive profiles. XDD SSTir measurements from 18° to 28°C and surface winds agreed well with drifting buoy- and satellite-derived estimates.
Excellent agreement was found between PTH and wind profiles measured by XDDs deployed from a NASA WB-57 at 18-km altitude offshore from the Texas coast and NWS radiosonde profiles from Brownsville and Corpus Christi, Texas. Successful XDD profiles were obtained in the clear and within precipitation over an offshore squall line.
Abstract
The High-Definition Sounding System (HDSS) is an automated system deploying the expendable digital dropsonde (XDD) designed to measure wind and pressure–temperature–humidity (PTH) profiles, and skin sea surface temperature (SST) within and around tropical cyclones (TCs) and other high-impact weather events needing high sampling density. Three experiments were conducted to validate the XDD.
On two successive days off the California coast, 10 XDDs and 14 Vaisala RD-94s were deployed from the navy’s Center for Interdisciplinary Remotely-Piloted Aircraft Studies (CIRPAS) Twin Otter aircraft over offshore buoys. The Twin Otter made spiral descents from 4 km to 60 m at the same descent rate as the sondes. Differences between successive XDD and RD-94 profiles due to true meteorological variability were on the same order as the profile differences between the spirals, XDDs, and RD-94s. XDD SST measured via infrared microradiometer, referred to as infrared skin SST (SSTir), and surface wind measurements were within 0.5°C and 1.5 m s−1, respectively, of buoy and Twin Otter values.
A NASA DC-8 flight launched six XDDs from 12 km between ex-TC Cosme and the Baja California coast. Repeatability was shown with good agreement between features in successive profiles. XDD SSTir measurements from 18° to 28°C and surface winds agreed well with drifting buoy- and satellite-derived estimates.
Excellent agreement was found between PTH and wind profiles measured by XDDs deployed from a NASA WB-57 at 18-km altitude offshore from the Texas coast and NWS radiosonde profiles from Brownsville and Corpus Christi, Texas. Successful XDD profiles were obtained in the clear and within precipitation over an offshore squall line.
