Domain knowledge integration into deep learning for typhoon intensity classification

Abstract

In this report, we propose a deep learning technique for high-accuracy estimation of the intensity class of a typhoon from a single satellite image, by incorporating meteorological domain knowledge. By using the Visual Geometric Group’s model, VGG-16, with images preprocessed with fisheye distortion, which enhances a typhoon’s eye, eyewall, and cloud distribution, we achieved much higher classification accuracy than that of a previous study, even with sequential-split validation. Through comparison of t-distributed stochastic neighbor embedding (t-SNE) plots for the feature maps of VGG with the original satellite images, we also verified that the fisheye preprocessing facilitated cluster formation, suggesting that our model could successfully extract image features related to the typhoon intensity class. Moreover, gradient-weighted class activation mapping (Grad-CAM) was applied to highlight the eye and the cloud distributions surrounding the eye, which are important regions for intensity classification; the results suggest that our model qualitatively gained a viewpoint similar to that of domain experts. A series of analyses revealed that the data-driven approach using only deep learning has limitations, and the integration of domain knowledge could bring new breakthroughs.

Thermodynamic and kinematic structure of tropical cyclones in the western North Pacific based on ACARS/AMDAR

“…The root mean square difference (RMSD) between this simple statistical model and the best track was 7.33 hPa, which is comparable to, or smaller than the uncertainties in the best track and other techniques (Hoshino and Nakazawa 2007;Nakazawa and Hoshino 2009;Shimada et al, 2016;Ito et al, 2018;Higa et al, 2021). Although δθ 1000 is difficult to calculate for individual case because aircrafts does not densely observe the whole region between rings of 800 km and 1,200 km in most cases, we can calculate the potential temperature anomaly with respect to the corresponding region based on the ERA5 reanalysis by European Centre for Medium-Range Weather Forecasts (Hersbach et al, 2020).…”

USFP: An unbalanced severe typhoon formation prediction framework based on transfer learning

“…Based on WindSat satellite ocean surface wind and precipitation data, Park et al (2016) from the Busan Institute of Ocean Science and Technology in South Korea used decision trees to analyze the intensity of tropical cyclones. Higa et al (2021) successfully estimated typhoon intensity with high accuracy by using the VGG-16 model to process a single satellite image and combining the knowledge of the meteorological domain. The machine learning-based methods as data-driven methods can ignore the imprecise physical mechanisms of typhoon formation and have significant advantages in capturing the nonlinear relationship between forecast factors and forecast targets (Reichstein et al, 2019).…”

Spatiotemporal fusion convolutional neural network: tropical cyclone intensity estimation from multisource remote sensing images

“…Second, existing intensity estimation model designs have not fully considered the temporal features of TCs, leading to significant estimation errors. Additionally, in deep learning-based intensity estimation research, many researchers have applied smoothing techniques to the MSW label data to enhance the accuracy of intensity estimation 22 , 23 . However, smoothing techniques require prior knowledge of the overall distribution of input data, which is not feasible for real-time TC intensity estimation tasks where the overall data distribution of TCs is unknown in advance, thus hindering practical applications.…”