# Typhoon Intensity Forecasting Based on LSTM Using the Rolling Forecast Method

^{1}

^{2}

^{*}

^{†}

## Abstract

**:**

## 1. Introduction

## 2. Study Area and Data

## 3. Methodologies

#### 3.1. Model Settings

#### 3.2. Forecasting Method

#### 3.3. Data Processing and Evaluation Indicators

## 4. LSTM Models

## 5. Experimental Results

#### 5.1. Determining the Optimal Prediction Factors

#### 5.2. Validation on the Test Set

#### 5.3. Validation on Typhoon Cases

## 6. Conclusions

## Author Contributions

## Funding

## Institutional Review Board Statement

## Informed Consent Statement

## Data Availability Statement

## Conflicts of Interest

## References

- Huo, Z.; Duan, W.; Zhou, F. Ensemble Forecastings of Tropical Cyclone Track with Orthogonal Conditional Nonlinear Optimal Perturbations. Adv. Atmos. Sci.
**2019**, 36, 231–247. [Google Scholar] [CrossRef] - Jiang, G.; Xu, J.; Wei, J. A Deep Learning Algorithm of Neural Network for the Parameterization of Typhoon-Ocean Feedback in Typhoon Forecast Models. Geophys. Res. Lett.
**2018**, 45. [Google Scholar] [CrossRef][Green Version] - Japan Meteorological Agency. Available online: http://www.jma.go.jp/jma/en/NMHS/indexe_nmhs.html (accessed on 14 January 2021).
- National Hurricane Center. Available online: https://www.nhc.noaa.gov/modelsummary.shtml (accessed on 14 January 2021).
- Sobrevilla, K.L.M.D.; Reyes, E.O.; Hendrickx, C.A.; Yao, S.S. Typhoon Forecasting in the Philippines Using an Optimal Multilayer Feedforward Artificial Neural Network Model Trained in Resilient Propagation Algorithm. In Proceedings of the 2016 IEEE Region 10 Conference (TENCON), Singapore, 22–25 November 2016. [Google Scholar] [CrossRef]
- Wang, R.; Wang, T.; Zhang, X.; Fang, Q.; Wu, C.; Zhang, B. An Artificial Neural Network Model for Predicting Typhoon Intensity and Its Application. In LSMS/ICSEE 2017, Part III, CCIS 763; Li, K., Xue, Y., Cui, S., Niu, Q., Yang, Z., Luk, P., Eds.; Springer Nature Singapore Pte Ltd.: Singapore, 2017; pp. 762–770. [Google Scholar] [CrossRef]
- Kim, S.; Matsumi, Y.; Pan, S.; Mase, H. A real-time forecast model using artificial neural network for after-runner storm surges on the Tottori coast, Japan. Ocean. Eng.
**2016**, 122, 44–53. [Google Scholar] [CrossRef] - Huang, Y.; Jin, L.; Zhao, H.; Huang, X. Fuzzy neural network and LLE Algorithm for forecasting precipitation in tropical cyclones: Comparisons with interpolation method by ECMWF and stepwise regression method. Nat. Hazards
**2018**, 91, 201–220. [Google Scholar] [CrossRef] - Shao, L.; Fu, G.; Chao, X.; Zhou, J. Application of BP neural network to forecasting typhoon tracks. J. Nat. Disasters
**2009**, 18. (In Chinese) [Google Scholar] [CrossRef] - Wei, C.-C.; Peng, P.-C.; Tsai, C.-H.; Huang, C.-L. Regional Forecasting of Wind Speeds during Typhoon Landfall in Taiwan: A Case Study of Westward-Moving Typhoons. Atmosphere
**2018**, 9, 141. [Google Scholar] [CrossRef][Green Version] - Lu, D.; Wang, X.; He, X. Hybrid population particle algorithm and multi-quantile robust extreme learning machine based short-term wind speed forecasting. Power Syst. Prot. Control.
**2019**, 47. (In Chinese) [Google Scholar] [CrossRef] - Zhang, Y.; Yang, S.; Guo, Z.; Guo, Y.; Zhao, J. Wind speed forecasting based on wavelet decomposition and wavelet neural networks optimized by the Cuckoo search algorithm. Atmos. Ocean. Sci. Lett.
**2019**, 12, 107–115. [Google Scholar] [CrossRef][Green Version] - Khelil, K.; Berrezzek, F.; Bouadjila, T. GA-based design of optimal discrete wavelet filters for efficient wind speed forecasting. Neural Comput. Appl.
**2020**, 1–14. [Google Scholar] [CrossRef] - Li, Y.; Yang, R.; Yang, C.; Yu, M.; Hu, F.; Jiang, Y. Leveraging LSTM for rapid intensifications prediction of tropical cyclones. In Proceedings of the 2nd International Symposium on Spatiotemporal Computing, Cambridge, MA, USA, 7–9 August 2017. [Google Scholar] [CrossRef][Green Version]
- Xu, G.; Liu, Y. Application of LSTM in Typhoon Path Prediction. Jisuanji Yu Xiandaihua
**2019**, 5. (In Chinese) [Google Scholar] [CrossRef] - Gao, S.; Zhao, P.; Pan, B.; Li, Y.; Zhou, M.; Xu, J.; Zhong, S. A nowcasting model for the prediction of typhoon tracks based on a long short term memory neural network. Acta Oceanol. Sin.
**2018**, 37, 8–12. [Google Scholar] [CrossRef] - Wei, Y.; Xu, X. Ultra-short-term wind speed prediction model using LSTM networks. J. Electron. Meas. Instrum.
**2019**, 33. (In Chinese) [Google Scholar] [CrossRef] - Yin, H.; Huang, S.Q.; Liu, Z.; Meng, A.B.; Yang, L. Short-term wind speed prediction based on fuzzy information granulation and LSTM. Electr. Meas. Instrum.
**2019**, 56. (In Chinese) [Google Scholar] [CrossRef] - Memarzadeh, G.; Keynia, F. A new short-term wind speed forecasting method based on fine-tuned LSTM neural network and optimal input sets. Energy Convers. Manag.
**2020**, 213. [Google Scholar] [CrossRef] - Liao, X.; Liu, Z.; Deng, W. Short-term wind speed multistep combined forecasting model based on two-stage decomposition and LSTM. Wind Energy
**2021**, 1–22. [Google Scholar] [CrossRef] - Chen, R.; Wang, X.; Zhang, W.; Zhu, X.; Li, A.; Yang, C. A hybrid CNN-LSTM model for typhoon formation forecasting. Geoinformatica
**2019**, 23, 375–396. [Google Scholar] [CrossRef] - Ying, M.; Zhang, W.; Yu, H.; Lu, X.; Feng, J.; Fan, Y.; Zhu, Y.; Chen, D. An overview of the china meteorological administration tropical cyclone database. J. Atmos. Oceanic Technol.
**2014**, 31, 287–301. [Google Scholar] [CrossRef][Green Version] - Yuan, S.; Luo, X.; Mu, B.; Li, J.; Dai, G. Prediction of North Atlantic Oscillation Index with Convolutional LSTM Based on Ensemble Empirical Mode Decomposition. Atmosphere
**2019**, 10, 252. [Google Scholar] [CrossRef][Green Version] - Mu, B.; Peng, C.; Yuan, S.; Chen, L. ENSO Forecasting over Multiple Time Horizons Using ConvLSTM Network and Rolling Mechanism. In Proceedings of the IJCNN, Budapest, Hungary, 14–19 July 2019. [Google Scholar]
- Mu, B.; Li, J.; Yuan, S.; Luo, X.; Dai, G. NAO Index Prediction using LSTM and ConvLSTM Networks Coupled with Discrete Wavelet Transform. In Proceedings of the IJCNN, Budapest, Hungary, 14–19 July 2019. [Google Scholar]
- Horinik, K.; Stinchcombe, M.; White, H. Multilayer Feedforward Networks are Universal Approximators. Neural Netw.
**1989**, 2, 359–366. [Google Scholar] [CrossRef] - Nakisa, B.; Rastgoo, M.N.; Rakotonirainy, A.; Maire, F.; Chandran, V. Long Short Term Memory Hyperparameter Optimization for a Neural Network Based Emotion Recognition Framework. IEEE Access
**2018**, 6, 49325–49338. [Google Scholar] [CrossRef] - Neshat, M.; Abbasnejad, E.; Shi, Q.; Alexander, B.; Wagner, M. Adaptive Neuro-Surrogate-Based Optimisation Method for Wave Energy Converters Placement Optimisation. In Proceedings of the International Conference on Neural Information Processing (ICONIP), Sydney, Australia, 12–15 December 2019. [Google Scholar] [CrossRef][Green Version]
- Abbasimehr, H.; Shabani, M.; Yoousefi, M. An optimized model using LSTM network for demand forecasting. Comput. Ind. Eng.
**2020**, 143. [Google Scholar] [CrossRef]

**Figure 1.**The number of typhoons recorded from 2000 to 2019. TY, typhoon; STY, strong typhoon; super TY, super typhoon.

**Figure 3.**The typhoon intensity forecasting models based on LSTM and the feed-forward neural network (FNN).

**Figure 6.**The prediction curves of the models based on LSTM and the FNN on test sets of 6, 12, 24 and 48 h ahead.

**Figure 7.**The prediction curves of the models based on LSTM and the FNN on test sets of 72, 96 and 120 h ahead.

**Figure 8.**The evaluation indicators of the model based on an FNN and LSTM using the typhoon Chan-hom (2015).

**Figure 9.**The pressure prediction curves of the models based on an FNN and LSTM on typhoon Chan-hom (2015).

**Figure 10.**The wind speed prediction curves of the models based on an FNN and LSTM on typhoon Chan-hom (2015).

Date (YYYYMMDDHH) | Intensity Level | Latitude | Longitude | Minimum Pressure (h Pa) | Maximum Wind Speed (m/s) |
---|---|---|---|---|---|

2015063000 | 1 | 95 | 1607 | 1000 | 15 |

2015063006 | 1 | 98 | 1601 | 1000 | 15 |

2015063012 | 2 | 100 | 1595 | 998 | 18 |

2015063018 | 2 | 101 | 1587 | 995 | 20 |

2015070100 | 2 | 105 | 1575 | 995 | 20 |

2015070106 | 2 | 110 | 1566 | 995 | 20 |

2015070112 | 2 | 113 | 1556 | 992 | 23 |

2015070118 | 3 | 113 | 1543 | 990 | 25 |

2015070200 | 3 | 111 | 1528 | 990 | 25 |

**Table 2.**Descriptive statistics of the variables based on the processed typhoon data for training and testing. P, minimum pressure of the typhoon’s center; WS, maximum wind speed of the typhoon’s center; Lat, latitude of the of the typhoon’s center; Lon, longitude of the typhoon’s center; MS, moving speed of the typhoon.

Statistics | P (h Pa) | WS (m/s) | Lat | Lon | MS (km/h) | |||||
---|---|---|---|---|---|---|---|---|---|---|

Train | Test | Train | Test | Train | Test | Train | Test | Train | Test | |

No. | 5394 | 2308 | 5394 | 2308 | 5394 | 2308 | 5394 | 2308 | 5258 | 2251 |

Max | 1010 | 1012 | 65 | 78 | 519 | 537 | 2029 | 2260 | 81.54 | 80.64 |

Min | 910 | 888 | 10 | 8 | 48 | 28 | 1008 | 1023 | 0 | 0 |

Mean | 977.77 | 975.91 | 29.31 | 30.26 | 212.1 | 206.24 | 1356.11 | 1345.11 | 12.28 | 12.39 |

**Table 3.**The variables and predictors of the typhoon data set, where t is the current time, t-1 is the previous 6 h, t-2 is the previous 12 h, t-3 is the previous 18 h, and t-4 is the previous 24 h.

Variables | Time | ||||
---|---|---|---|---|---|

t | t-1 | t-2 | t-3 | t-4 | |

P | P (t) | P (t-1) | P (t-2) | P (t-3) | P (t-4) |

WS | WS (t) | WS (t-1) | WS (t-2) | WS (t-3) | WS (t-4) |

Lat | Lat (t) | Lat (t-1) | Lat (t-2) | Lat (t-3) | Lat (t-4) |

Lon | Lon (t) | Lon (t-1) | Lon (t-2) | Lon (t-3) | Lon (t-4) |

MS | MS (t) | MS (t-1) | MS (t-2) | MS (t-3) | MS (t-4) |

**Table 4.**The hyperparameters of the typhoon intensity forecasting models. MSE, mean squared error; LSTM, long short-term memory; FNN, feed-forward neural network.

Hyperparameters | LSTM Models | FNN Model |
---|---|---|

Number of hidden layers | 1 | 1 |

Number of prediction factors | n | n |

Number of hidden layer neurons | 4*n | 4*n |

Activation function of the hidden layer | Tanh | Sigmoid |

Loss function | MSE | MSE |

Maximum iteration step | 300 | 5000 |

Optimizer | Adam | Default |

Learning rate | 0.01 | 0.001 |

Goal learning rate | Default | 0.0001 |

**Table 5.**The typhoon intensity forecasting models and the function expression of it using the different prediction factors’ combination: minimum pressure of the typhoon center (P), maximum wind speed of the typhoon center (WS), latitude (Lat) and longitude (Lon) of the typhoon center, and moving speed of the typhoon (MS).

Model | Function Expression |
---|---|

LSTM-1 | P(i+1), WS(t+i) = f[P(t+i-1), WS(t+i-1)] |

LSTM-2 | P(t+i), WS(t+i) = f[P(t+i-1), WS(t+i-1), P(t+i-2), WS(t+i-2)] |

LSTM-3 | P(t+i), WS(t+i), Lat(t+i) = f[P(t+i-1), WS(t+i-1), Lat(t+i-1)] |

LSTM-4 | P(t+i), WS(t+i), Lat(t+i) = f[P(t+i-1), WS(t+i-1), Lat(t+i-1), P(t+i-2), WS(t+i-2), LAT(t+i-2)] |

LSTM-5 | P(t+i), WS(t+i), Lon(t+i) = f[P(t+i-1), WS(t+i-1), Lon(t+i-1)] |

LSTM-6 | P(t+i), WS(t+i), Lon(t+i) = f[P(t+i-1), WS(t+i-1), Lon(t+i-1), P(t+i-2), WS(t+i-2), Lon(t+i-2)] |

LSTM-7 | P(t+i), WS(t+i), Lat(t+i), Lon(t+i) = f[P(t+i-1), WS(t+i-1), Lat(t+i-1), Lon(t+i-1)] |

LSTM-8 | P(t+i), WS(t+i), Lat(t+i), Lon(t+i) = f[P(t+i-1), WS(t+i-1), Lat(t+i-1), Lon(t+i-1), |

P(t+i-2), WS(t+i-2), Lat(t+i-2), Lon(t+i-2)] | |

LSTM-9 | P(t+i), WS(t+i), Lat(t+i), Lon(t+i), MS(t+i) = f[P(t+i-1), WS(t+i-1), Lat(t+i-1), |

Lon(t+i-1), MS(t+i-1)] | |

LSTM-10 | P(t+i), WS(t+i), Lat(t+i), Lon(t+i), MS(t+i) = f[P(t+i-1), WS(t+i-1), Lat(t+i-1), |

Lon(t+i-1), MS(t+i-1), P(t+i-2), WS(t+i-2), Lat(t+i-2), Lon(t+i-2), MS(t+i-2)] |

Models | Pressure (Lead Time, h) | Wind Speed (Lead Time, h) | ||||||||||||
---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|

6 | 12 | 24 | 48 | 72 | 96 | 120 | 6 | 12 | 24 | 48 | 72 | 96 | 120 | |

LSTM-1 | 4.02 | 7.14 | 12.24 | 19.65 | 24.18 | 26.23 | 26.04 | 2.26 | 4.11 | 7.17 | 11.71 | 14.43 | 15.53 | 15.3 |

LSTM-2 | 3.44 | 6.07 | 11 | 18.15 | 22.44 | 24.64 | 25.09 | 1.95 | 3.43 | 6.31 | 10.62 | 13.1 | 14.3 | 14.52 |

LSTM-3 | 3.85 | 6.46 | 10.59 | 16.45 | 20.4 | 23.02 | 24.17 | 2.16 | 3.61 | 6.01 | 9.44 | 11.65 | 13.03 | 13.79 |

LSTM-4 | 3.42 | 5.74 | 9.87 | 15.49 | 18.9 | 21.42 | 23.6 | 1.89 | 3.15 | 5.45 | 8.81 | 10.82 | 12.34 | 13.7 |

LSTM-5 | 4.02 | 7.04 | 11.75 | 18.02 | 21.49 | 22.78 | 22.41 | 2.28 | 4.06 | 6.92 | 10.78 | 12.76 | 13.39 | 13.12 |

LSTM-6 | 3.35 | 5.73 | 10.09 | 16.1 | 19.6 | 21.23 | 21.38 | 1.9 | 3.3 | 5.9 | 9.56 | 11.5 | 12.31 | 12.38 |

LSTM-7 | 3.72 | 6.11 | 9.84 | 14.76 | 17.68 | 19.57 | 22.54 | 2.06 | 3.37 | 5.52 | 8.36 | 9.95 | 11.13 | 12.42 |

LSTM-8 | 3.33 | 5.38 | 8.67 | 12.46 | 14.78 | 17.35 | 20 | 1.84 | 2.98 | 4.87 | 7.12 | 8.5 | 9.99 | 11.57 |

LSTM-9 | 3.79 | 6.27 | 10.1 | 15.19 | 17.94 | 19.28 | 23.74 | 2.09 | 3.46 | 5.67 | 8.71 | 10.34 | 11.27 | 12.49 |

LSTM-10 | 3.42 | 5.55 | 8.93 | 12.71 | 15.19 | 18.08 | 20.85 | 1.9 | 3.09 | 5.03 | 7.32 | 8.78 | 10.41 | 11.91 |

**Table 7.**The LSTM models using the prediction factors at the current time and in the previous 6, 12, and 18 h.

Model | Function Expression |
---|---|

LSTM-11 | P(t+i), WS(t+i), Lat(t+i), Lon(t+i)=f[P(t+i-1), WS(t+i-1), Lat(t+i-1), Lon(t+i-1), P(t+i-2), WS(t+i-2), Lat(t+i-2), Lon(t+i-2), |

P(t+i-3), WS(t+i-3), Lat(t+i-3), Lon(t+i-3)] | |

LSTM-12 | P(t+i), WS(t+i), Lat(t+i), Lon(t+i)=f[P(t+i-1), WS(t+i-1), Lat(t+i-1), Lon(t+i-1), P(t+i-2), WS(t+i-2), Lat(t+i-2), Lon(t+i-2), |

P(t+i-3), WS(t+i-3), Lat(t+i-3), Lon(t+i-3), P(t+i-4), WS(t+i-4), Lat(t+i-4), Lon(t+i-4)] |

Models | Pressure (Lead Time, h) | Wind Speed (Lead Time, h) | ||||||||||||
---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|

6 | 12 | 24 | 48 | 72 | 96 | 120 | 6 | 12 | 24 | 48 | 72 | 96 | 120 | |

LSTM-8 | 3.33 | 5.38 | 8.67 | 12.46 | 14.78 | 17.35 | 20 | 1.84 | 2.98 | 4.87 | 7.12 | 8.5 | 9.99 | 11.57 |

LSTM-11 | 3.4 | 5.52 | 9.14 | 14 | 16.63 | 18.48 | 19.64 | 1.96 | 3.16 | 5.32 | 8.14 | 9.53 | 10.47 | 11.14 |

LSTM-12 | 3.41 | 5.46 | 8.93 | 14.07 | 16.88 | 18.16 | 18.83 | 1.95 | 3.13 | 5.29 | 8.41 | 9.94 | 10.58 | 10.89 |

Evaluation Indicators | Models | Pressure (Lead Time, h) | Wind Speed (Lead Time, h) | ||||||||||||
---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|

6 | 12 | 24 | 48 | 72 | 96 | 120 | 6 | 12 | 24 | 48 | 72 | 96 | 120 | ||

MAE | FNN | 3.45 | 5.67 | 9.17 | 13.35 | 15.41 | 16.61 | 18.29 | 1.96 | 3.19 | 5.17 | 7.69 | 8.94 | 9.79 | 10.85 |

LSTM-8 | 3.33 | 5.38 | 8.67 | 12.46 | 14.78 | 17.35 | 20 | 1.84 | 2.98 | 4.87 | 7.12 | 8.5 | 9.99 | 11.57 | |

RMSE | FNN | 5.21 | 6.99 | 13.17 | 24.71 | 32.21 | 34.81 | 35.87 | 2.79 | 4.02 | 7.74 | 14.65 | 19.36 | 21.48 | 22.12 |

LSTM-8 | 4.84 | 7.86 | 12.32 | 17.62 | 21.57 | 25.65 | 29.47 | 2.5 | 4.15 | 6.65 | 9.76 | 12.1 | 14.45 | 16.62 |

Typhoon Cases | Models | Pressure (Lead Time, h) | Wind Speed (Lead Time, h) | ||||||||||||
---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|

6 | 12 | 24 | 48 | 72 | 96 | 120 | 6 | 12 | 24 | 48 | 72 | 96 | 120 | ||

Chan-hom | FNN | 7.17 | 9.99 | 14.29 | 20.83 | 18.98 | 17.18 | 19.31 | 4.25 | 6.24 | 8.05 | 11.37 | 8.89 | 8.11 | 8.44 |

LSTM-8 | 4.04 | 8.21 | 13.02 | 13.73 | 9.99 | 7.54 | 15.47 | 2.59 | 4.97 | 7.86 | 8.46 | 5.86 | 3.79 | 8.94 | |

Soudelor | FNN | 8.37 | 12.17 | 17.69 | 34.14 | 46.35 | 44.45 | 39.55 | 5.09 | 7.87 | 13.92 | 19.15 | 33 | 45.47 | 43.77 |

LSTM-8 | 3.85 | 6.79 | 11.07 | 19.67 | 27.58 | 30.79 | 30.69 | 2.08 | 3.48 | 5.13 | 9.48 | 14.19 | 16.69 | 16.89 |

Typhoon Cases | Models | Pressure (Lead Time, h) | Wind Speed (Lead Time, h) | ||||||||||||
---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|

6 | 12 | 24 | 48 | 72 | 96 | 120 | 6 | 12 | 24 | 48 | 72 | 96 | 120 | ||

Chan-hom | FNN | 7.24 | 11.54 | 17.03 | 21.11 | 19.6 | 16.28 | 15.31 | 5.11 | 7.92 | 11.79 | 14 | 13.42 | 12.63 | 9.02 |

LSTM-8 | 5.61 | 10.72 | 16.19 | 16.58 | 13.47 | 10.07 | 9.69 | 3.5 | 6.82 | 10.42 | 10.33 | 8.05 | 5.73 | 5.63 | |

Soudelor | FNN | 11.25 | 16.75 | 24.33 | 43.24 | 52.81 | 51.67 | 47.53 | 7.63 | 11.84 | 22.67 | 27.72 | 40.48 | 50.21 | 49.06 |

LSTM-8 | 5.33 | 9.33 | 15.29 | 25.59 | 33.68 | 36.28 | 33.55 | 2.84 | 4.62 | 7.13 | 12.6 | 17.58 | 19.71 | 18.89 |

Publisher’s Note: MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations. |

© 2021 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

## Share and Cite

**MDPI and ACS Style**

Yuan, S.; Wang, C.; Mu, B.; Zhou, F.; Duan, W. Typhoon Intensity Forecasting Based on LSTM Using the Rolling Forecast Method. *Algorithms* **2021**, *14*, 83.
https://doi.org/10.3390/a14030083

**AMA Style**

Yuan S, Wang C, Mu B, Zhou F, Duan W. Typhoon Intensity Forecasting Based on LSTM Using the Rolling Forecast Method. *Algorithms*. 2021; 14(3):83.
https://doi.org/10.3390/a14030083

**Chicago/Turabian Style**

Yuan, Shijin, Cheng Wang, Bin Mu, Feifan Zhou, and Wansuo Duan. 2021. "Typhoon Intensity Forecasting Based on LSTM Using the Rolling Forecast Method" *Algorithms* 14, no. 3: 83.
https://doi.org/10.3390/a14030083