Presentation Information
[AAS08-P19]Monitoring the movement of volcanic ash after the Sakurajima eruption using Himawari satellite and X-band radar images
*Shinto Yokoyama1, Masafumi Ohashi1 (1.Kagoshima University)
Keywords:
volcanic ash,X-band radar,Himawari 8/9 image
Sakurajima, an active volcano, frequently erupts, causing ash fall damage to surrounding areas. Therefore, measuring instruments such as X-band radar have been installed to track the distribution of volcanic ash after eruptions. These observations allow us to understand the distribution of ash fall around Sakurajima. Still, it is difficult to determine how volcanic ash from eruptions reaching heights of 2,500 meters moves beyond Sakurajima and across the horizon. The Himawari-8/9 meteorological satellites cover the western Pacific region, including Japan, and continuously observe the same locations at 10-minute intervals, making them expected to provide valuable insights into volcanic ash behavior. This satellite observes images across 16 wavelength bands: 3 bands in the visible spectrum, 3 bands in the near-infrared spectrum, and 10 bands in the infrared spectrum. Satellite imagery shows volcanic ash and clouds, and to understand their distribution, it is necessary to develop an analysis algorithm that distinguishes between them. Previously, we attempted to detect volcanic ash using the Random Forest algorithm on 16-band images and difference images from Himawari-8/9 satellites. However, it proved difficult to improve the accuracy of the volcanic ash detection filter when using manually labeled data in which human eyes identified volcanic ash in Himawari-8/9 images. Therefore, we prepared data that included not only the 16-band images and difference images from Himawari-8/9, but also ground temperature distributions estimated from Japan Meteorological Agency data using volcanic ash distribution identified by X-band radar observations installed in Tarumi as training data. The goal was to improve the accuracy of volcanic ash distribution detection by determining classification rules for ground/sea surface, clouds, and volcanic ash using machine learning with Logistic regression algorithm.
The training data for volcanic ash distribution derived from X-band radar data were created as follows. First, we downloaded raw X-band radar data from the Ministry of Land, Infrastructure, Transport and Tourism database for eruptions exceeding 2,500 meters in height, occurring 30 minutes before and 2 hours after the eruptions at Sakurajima since July 2015. Subsequently, the radar equation proposed by Melnikov et al. was applied to obtain uncorrected pure backscatter data. 3D labeling was then applied to this data to remove noise from volcanic ash dispersing after the eruption. We created volcanic ash distribution images using X-band radar, synchronized with the timing of Himawari 8/9 imagery.
By utilizing ground temperature distributions, the Himawari-8/9 infrared imagery enabled more accurate identification of cloud-free ground and sea surfaces, volcanic ash, and clouds. Using Himawari-8/9 multi-wavelength imagery and differential characteristics across wavelengths, X-band radar, surface air temperature distributions, and data from various sources, we developed rules for identifying volcanic ash distribution in Himawari-8/9 imagery using machine learning with Logistic regression algorithm. In particular, the difference images between 6.95x10-6m and 7.35 x10-6m, and between 10.45 x10-6m and 11.20 x10-6m, strongly indicated the distribution and behavior of volcanic ash.
Using rules derived from machine learning with Logistic regression algorithm, we first distinguished between bare ground/sea surface and volcanic ash or clouds based on the infrared and surface air temperature distributions from Himawari-8/9. Subsequently, we differentiated between volcanic ash and clouds, thereby labeling approximately 2 km2 pixels in the Himawari-8/9 images as ground/sea surface, clouds, or volcanic ash. Additionally, by arranging the distribution of volcanic ash after the eruption in chronological order, we tracked its movement.
The training data for volcanic ash distribution derived from X-band radar data were created as follows. First, we downloaded raw X-band radar data from the Ministry of Land, Infrastructure, Transport and Tourism database for eruptions exceeding 2,500 meters in height, occurring 30 minutes before and 2 hours after the eruptions at Sakurajima since July 2015. Subsequently, the radar equation proposed by Melnikov et al. was applied to obtain uncorrected pure backscatter data. 3D labeling was then applied to this data to remove noise from volcanic ash dispersing after the eruption. We created volcanic ash distribution images using X-band radar, synchronized with the timing of Himawari 8/9 imagery.
By utilizing ground temperature distributions, the Himawari-8/9 infrared imagery enabled more accurate identification of cloud-free ground and sea surfaces, volcanic ash, and clouds. Using Himawari-8/9 multi-wavelength imagery and differential characteristics across wavelengths, X-band radar, surface air temperature distributions, and data from various sources, we developed rules for identifying volcanic ash distribution in Himawari-8/9 imagery using machine learning with Logistic regression algorithm. In particular, the difference images between 6.95x10-6m and 7.35 x10-6m, and between 10.45 x10-6m and 11.20 x10-6m, strongly indicated the distribution and behavior of volcanic ash.
Using rules derived from machine learning with Logistic regression algorithm, we first distinguished between bare ground/sea surface and volcanic ash or clouds based on the infrared and surface air temperature distributions from Himawari-8/9. Subsequently, we differentiated between volcanic ash and clouds, thereby labeling approximately 2 km2 pixels in the Himawari-8/9 images as ground/sea surface, clouds, or volcanic ash. Additionally, by arranging the distribution of volcanic ash after the eruption in chronological order, we tracked its movement.