# Atlantic Hurricane Activity Prediction: A Machine Learning Approach

^{1}

^{2}

^{*}

## Abstract

**:**

## 1. Introduction

- Short term predictions: These are generally made 6 h to 7 days in advance. The predictands are spatio-temporal quantities, such as track and intensity of an individual cyclone.
- Long term predictions: These are seasonal forecasts generally made at least a month in advance. The predictands are one-dimensional time series like the total number of cyclones and Accumulated Cyclone Energy (ACE) in a basin (see Section 4 for details). Generally, certain handcrafted climatological indices are used as predictors.

## 2. Related Work

## 3. Hurricane Dynamics

- A preexisting low-pressure disturbance
- A sufficient Coriolis force to provide rotational force to oncoming winds.
- Sufficiently warm sea surface temperatures.
- High humidity in the lower to middle levels of the troposphere.
- Atmospheric instability to sustain a vertical movement of air parcels.
- Low vertical wind shear.

## 4. Predictors and Predictands

#### 4.1. NINO Indices for ENSO

**monthly**average SST at location $(x,y)$ at month ${t}_{m}$.

**daily**average SST at $(x,y)$ at day ${t}_{d}$ and ${N}_{d}=30\times {N}_{m}$.

#### 4.2. Accumulated Cyclone Energy (ACE)

## 5. Problem Formulation

## 6. Proposed Solution

#### Dimensionality Reduction/Information Extraction

- Approximately 10–20 s-values hold above 5% relative significance. Most of these s-values have relatively similar significance.
- 50–100 s-values hold above 1% relative significance.
- No single s-value holds above 15% significance.

## 7. Experiments

## 8. Results

#### 8.1. Prediction Accuracy

#### 8.2. Prediction Reliability

## 9. Conclusions

## Author Contributions

## Funding

## Data Availability Statement

## Acknowledgments

## Conflicts of Interest

## References

- Bauer, P.; Thorpe, A.; Brunet, G. The quiet revolution of numerical weather prediction. Nature
**2015**, 525, 47. [Google Scholar] [CrossRef] [PubMed] - Blake, E.S.; Gray, W.M. Prediction of August Atlantic basin hurricane activity. Weather Forecast.
**2004**, 19, 1044–1060. [Google Scholar] [CrossRef][Green Version] - Klotzbach, P.; Blake, E.; Camp, J.; Caron, L.P.; Chan, J.C.; Kang, N.Y.; Kuleshov, Y.; Lee, S.M.; Murakami, H.; Saunders, M.; et al. Seasonal Tropical Cyclone Forecasting. Trop. Cyclone Res. Rev.
**2014**, 8, 134–149. [Google Scholar] [CrossRef] - Camargo, S.J.; Barnston, A.G.; Klotzbach, P.J.; Landsea, C.W. Seasonal tropical cyclone forecasts. WMO Bull.
**2007**, 56, 297–309. [Google Scholar] - Hughes, G.F. On the mean accuracy of statistical pattern recognizers. IEEE Trans. Inf. Theory
**1968**, 14, 55–63. [Google Scholar] [CrossRef][Green Version] - Roheda, S.; Krim, H.; Riggan, B.S. Robust Multi-Modal Sensor Fusion: An Adversarial Approach. IEEE Sens. J.
**2021**, 21, 1885–1896. [Google Scholar] [CrossRef] - Roheda, S.; Krim, H.; Luo, Z.; Wu, T. Decision Level Fusion: An Event Driven Approach. In Proceedings of the 2018 26th European Signal Processing Conference (EUSIPCO), Rome, Italy, 3–7 September 2018; pp. 2598–2602. [Google Scholar]
- Saunders, M.A.; Klotzbach, P.J.; Lea, A.S.R.; Schreck, C.J.; Bell, M.M. Quantifying the probability and causes of the surprisingly active 2018 North Atlantic hurricane season. Earth Space Sci.
**2020**, 7, e2019EA000852. [Google Scholar] [CrossRef][Green Version] - Davis, K.; Zeng, X.; Ritchie, E.A. A New Statistical Model for Predicting Seasonal North Atlantic Hurricane Activity. Weather. Forecast.
**2015**, 30, 730–741. [Google Scholar] [CrossRef] - Kozar, M.E.; Mann, M.E.; Camargo, S.J.; Kossin, J.P.; Evans, J.L. Stratified statistical models of North Atlantic basin-wide and regional tropical cyclone counts. J. Geophys. Res. Atmos.
**2012**, 117, D18. [Google Scholar] [CrossRef][Green Version] - Keith, E.; Xie, L. Predicting Atlantic Tropical Cyclone Seasonal Activity in April. Weather. Forecast.
**2009**, 24, 436–455. [Google Scholar] [CrossRef] - Alemany, S.; Beltran, J.; Perez, A.; Ganzfried, S. Predicting Hurricane Trajectories Using a Recurrent Neural Network. Proc. Aaai Conf. Artif. Intell.
**2019**, 33, 468–475. [Google Scholar] [CrossRef][Green Version] - Richman, M.B.; Leslie, L.M.; Ramsay, H.A.; Klotzbach, P.J. Reducing tropical cyclone prediction errors using machine learning approaches. Procedia Comput. Sci.
**2017**, 114, 314–323. [Google Scholar] [CrossRef] - LeCun, Y.; Bengio, Y.; Hinton, G. Deep learning. Nature
**2015**, 521, 436. [Google Scholar] [CrossRef] - Ledig, C.; Theis, L.; Huszár, F.; Caballero, J.; Cunningham, A.; Acosta, A.; Aitken, A.; Tejani, A.; Totz, J.; Wang, Z.; et al. Photo-realistic single image super-resolution using a generative adversarial network. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), Honolulu, HI, USA, 21–26 July 2017; IEEE: New York, NY, USA, 2017. [Google Scholar]
- Tran, K.; Panahi, A.; Adiga, A.; Sakla, W.; Krim, H. Nonlinear Multi-scale Super-resolution Using Deep Learning. ICASSP
**2019**, 2019, 3182–3186. [Google Scholar] - Yang, D.; Sun, J. Bm3d-net: A convolutional neural network for transformdomain collaborative filtering. IEEE Signal Process. Lett.
**2018**, 25, 55–59. [Google Scholar] [CrossRef] - Ham, Y.G.; Kim, J.H.; Luo, J.J. Deep learning for multi-year ENSO forecasts. Nature
**2019**, 573, 568. [Google Scholar] [CrossRef] - Merrill, R.T. Environmental influences on hurricane intensification. J. Atmos. Sci.
**1988**, 45, 1678–1687. [Google Scholar] [CrossRef][Green Version] - Shaman, J.; Esbensen, S.K.; Maloney, E.D. The dynamics of the ENSO—Atlantic hurricane teleconnection: ENSO-related changes to the North African—Asian Jet affect Atlantic basin tropical cyclogenesis. J. Clim.
**2009**, 22, 2458–2482. [Google Scholar] [CrossRef][Green Version] - Klotzbach, P.J. El Niño—Southern Oscillation’s Impact on Atlantic Basin Hurricanes and U.S. Landfalls. J. Clim.
**2011**, 24, 1252–1263. [Google Scholar] [CrossRef] - Trenberth, K.E. The Definition of El Nino. Bull. Am. Meteorol. Soc.
**1997**, 78, 2771–2777. [Google Scholar] [CrossRef][Green Version] - Trenberth, K.; David, E.; Stepaniak, P. Indices of El Niño evolution. J. Clim.
**2001**, 14, 1697–1701. [Google Scholar] [CrossRef][Green Version] - Sun, C.; Shrivastava, A.; Singh, S.; Gupta, A. Revisiting Unreasonable Effectiveness of Data in Deep Learning Era. In Proceedings of the 2017 IEEE International Conference on Computer Vision (ICCV), Venice, Italy, 22–29 October 2017; pp. 843–852. [Google Scholar] [CrossRef][Green Version]
- Gray, W.M.; Landsea, C.W.; Sea, C.W.; Mielke, P.W.; Berry, K.J. Forecast of Atlantic Seasonal Hurricane Activity for 1988; Colorado State University: Fort Collins, CO, USA, 1988; pp. 13–14. [Google Scholar]
- Kalnay, E.; Kanamitsu, M.; Kistler, R.; Collins, W.; Deaven, D.; Gandin, L.; Iredell, M.; Saha, S.; White, G.; Woollen, J.; et al. The NCEP/NCAR 40-year reanalysis project. Bull. Am. Meteorol. Soc.
**1996**, 77, 437–470. [Google Scholar] [CrossRef][Green Version] - Lindsey, R. In Watching for El Niño and La Niña, NOAA Adapts to Global Warming. Underrstanding Climate, NOAA Climate.gov. Available online: https://www.climate.gov/news-features/understanding-climate/watching-el-niño-and-la-niña-noaa-adapts-global-warming (accessed on 2 April 2021).
- Saunders, M.A.; Lea, A. Extended Range Forecast for Atlantic Hurricane Activity in 2021. 2020. Available online: Tropicalstormrisk.com (accessed on 2 April 2021).
- Livieris, I.E.; Stavroyiannis, S.; Pintelas, E.; Pintelas, P. A novel validation framework to enhance deep learning models in time-series forecasting. Neural Comput. Appl.
**2020**, 32, 17149–17167. [Google Scholar] [CrossRef] - Ljung, G.M.; Box, G.E.P. On a measure of lack of fit in time series models. Biometrika
**1978**, 65, 297–303. [Google Scholar] [CrossRef] - Brockwell, P.J.; Davis, R.A. Time Series: Theory and Methods, 2nd ed.; Springer: New York, NY, USA, 1991. [Google Scholar]

**Figure 2.**Column vectors, ${\mathbf{u}}_{SLP}^{0}$, ${\mathbf{u}}_{SLP}^{1}$, ${\mathbf{u}}_{SLP}^{2}$ and ${\mathbf{u}}_{SLP}^{212}$ for ${\mathsf{\Gamma}}_{SLP}$. x-axis is the index value and y-axis is the amplitude. The vectors are plotted as 1-D time series.The D matrix is formed by ${\widehat{\eta}}_{3.4}(t+3$ months).

**Figure 3.**Reshaped column vectors (in clockwise order starting from upper left), ${\mathbf{v}}_{SLP}^{0}$, ${\mathbf{v}}_{SLP}^{1}$, ${\mathbf{v}}_{SLP}^{2}$ and ${\mathbf{v}}_{SLP}^{212}$ for ${\mathsf{\Gamma}}_{SLP}$. The vectors were reshaped to 2-D matrices with same shape as the original SLP map ($25\times 72$). The W matrix is formed by ${\widehat{\eta}}_{3.4}(t+3$ months).

**Figure 4.**Monthly NINO3.4 predictions (January 2012 to December 2017) using SVD based dimensionality reduction compared with the original NINO3.4 index (black curve). x-axis is months starting from January, 2012.

**Figure 5.**Monthly NINO1.2 predictions (January 2012 to December 2017) using SVD based dimensionality reduction compared with the original NINO1.2 index (black curve). x-axis is months starting from January, 2012.

**Figure 6.**Annual Atlantic ACE predictions (2012 to 2018) using SVD based dimensionality reductions compared with the Poisson regression based seasonal predictions done by Colorado State University in August each year (dashed blue curve) and the original ACE curve (black curve).

**Figure 7.**Autocorrelation function plots for residuals from ACE prediction models. Dashed blue lines show intervals for 95% confidence.

Model | Role in the Model | Spatial Resolution | Original Temporal Resolution | Temporal Resolution Used in the Model | Data Source |
---|---|---|---|---|---|

NINO3.4 (${\eta}_{34}\left(t\right)$) | Predictand | NA | Monthly | Bimonthly | NOAA |

NINO1 + 2 (${\eta}_{1\phantom{\rule{3.33333pt}{0ex}}+\phantom{\rule{3.33333pt}{0ex}}2}\left(t\right)$) | Predictand | NA | Monthly | Bimonthly | NOAA |

ACE ($\varphi \left(t\right)$) | Predictand | NA | Annual | Bimonthly | HURDAT2 |

Sea Level Pressure ($SLP(t,x,y)$) | Predictor | ${5}^{\circ}\times {5}^{\circ}$ | 4 times Daily | Bimonthly | NCEP-NCAR |

Sea Surface Temperature ($SST(t,x,y)$) | Predictor | ${5}^{\circ}\times {5}^{\circ}$ | 4 times Daily | Bimonthly | NCEP-NCAR |

Zonal Wind Speed at 850 mb ($UWN{D}_{850mb}(t,x,y)$) | Predictor | ${5}^{\circ}\times {5}^{\circ}$ | 4 times Daily | Bimonthly | NCEP-NCAR |

Meridional Wind Speed at 850 mb ($VWN{D}_{850mb}(t,x,y)$) | Predictor | ${5}^{\circ}\times {5}^{\circ}$ | 4 times Daily | Bimonthly | NCEP-NCAR |

Zonal Wind Speed at 200 mb ($UWN{D}_{200mb}(t,x,y)$) | Predictor | ${5}^{\circ}\times {5}^{\circ}$ | 4 times Daily | Bimonthly | NCEP-NCAR |

Meridional Wind Speed at 200 mb ($VWN{D}_{200mb}(t,x,y)$) | Predictor | ${5}^{\circ}\times {5}^{\circ}$ | 4 times Daily | Bimonthly | NCEP-NCAR |

Relative Humidity at 700 mb ($R{H}_{700mb}(t,x,y)$) | Predictor | ${5}^{\circ}\times {5}^{\circ}$ | 4 times Daily | Bimonthly | NCEP-NCAR |

**Table 2.**ACE pediction efficiency metrics for different prediction windows compared with CSU seasonal predictions made in August.

Model | MAE | RMSE | R2 | MSSS |
---|---|---|---|---|

CSU August Prediction | 51.71 | 62.26 | 0.0719 | −9.48% |

Fused CNN with ${t}_{p}=3$ months | 41.44 | 46.05 | 0.4571 | 40.09% |

Fused CNN with ${t}_{p}=6$ months | 61.15 | 83.31 | 0.0800 | −96.04% |

Fused CNN with ${t}_{p}=9$ months | 39.11 | 56.80 | 0.3079 | 8.87% |

Fused CNN with ${t}_{p}=12$ months | 37.19 | 48.45 | 0.5779 | 33.70% |

Fused CNN with ${t}_{p}=18$ months | 61.19 | 81.00 | 0.0656 | −85.35% |

**Table 3.**Prediction reliability metrics over the Test set (2012–2018) for different prediction windows with significance level of 0.05.

Model | Ljung Box Q Score | p-Value | Null ${\mathit{H}}_{0}$ |
---|---|---|---|

CSU August Prediction | 5.45 | 0.49 | Accepted |

Fused CNN with ${t}_{p}=3$ months | 10.57 | 0.1 | Accepted |

Fused CNN with ${t}_{p}=6$ months | 4.82 | 0.57 | Accepted |

Fused CNN with ${t}_{p}=9$ months | 6.66 | 0.35 | Accepted |

Fused CNN with ${t}_{p}=12$ months | 7.02 | 0.32 | Accepted |

Fused CNN with ${t}_{p}=18$ months | 3.86 | 0.7 | Accepted |

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 (https://creativecommons.org/licenses/by/4.0/).

## Share and Cite

**MDPI and ACS Style**

Asthana, T.; Krim, H.; Sun, X.; Roheda, S.; Xie, L. Atlantic Hurricane Activity Prediction: A Machine Learning Approach. *Atmosphere* **2021**, *12*, 455.
https://doi.org/10.3390/atmos12040455

**AMA Style**

Asthana T, Krim H, Sun X, Roheda S, Xie L. Atlantic Hurricane Activity Prediction: A Machine Learning Approach. *Atmosphere*. 2021; 12(4):455.
https://doi.org/10.3390/atmos12040455

**Chicago/Turabian Style**

Asthana, Tanmay, Hamid Krim, Xia Sun, Siddharth Roheda, and Lian Xie. 2021. "Atlantic Hurricane Activity Prediction: A Machine Learning Approach" *Atmosphere* 12, no. 4: 455.
https://doi.org/10.3390/atmos12040455