Hybrid AI-Physics Models for Improving Tropical Cyclone Rapid Intensification Forecasts

Authors: Dr. Sachin S. Chinchorkar
Hybrid AI-Physics Models for Improving Tropical Cyclone Rapid Intensification Forecasts
DOI
10.25125/ijoear-oct-2026-21
DIN
IJOEAR-OCT-2026-21
Abstract

One of the major challenges in forecasting is the rapid intensification (RI) of tropical cyclones, wherein cyclone intensity can increase rapidly under dynamic and changing atmospheric and oceanic conditions. This paper explores how a hybrid artificial intelligence (AI)-physics model can be used to enhance tropical cyclone RI forecasting. The proposed method integrates the nonlinear pattern-recognition capability of AI with physically meaningful environmental predictors, such as sea-surface temperature, ocean heat content, vertical wind shear, relative humidity, central pressure, and previous intensity change. Historical tropical cyclone observations are analysed and organized in Microsoft Excel, and model performance is evaluated using accuracy, precision, recall, F1-score, mean absolute error (MAE), and root mean square error (RMSE). The comparative analysis involves a physics-informed baseline, a standalone AI model, and the proposed hybrid model. The findings show that the hybrid model gives better RI classification and reduces intensity forecasting error as compared to the alternative methods. The results reveal that incorporating a combination of physical information and AI can enhance the precision and physical applicability of cyclone forecasts. This paper concludes that hybrid AI-physics modelling offers a valuable framework for developing tropical cyclone RI prediction and can aid improved early warning, disaster preparedness, and operational forecasting.

Keywords
Tropical Cyclones Rapid Intensification Artificial Intelligence Physics-Informed Modelling Intensity Forecasting.
Download Options
Download Full PDF
Share Article
Facebook Twitter LinkedIn
Introduction

Tropical cyclones are considered to be among the most devastating weather phenomena, producing severe effects due to extreme winds, storm surge, torrential rainfall, and coastal flooding. Proper prediction of cyclone intensity is thus critical to disaster preparation, evacuation, and mitigation of risks [1]. However, rapid intensification (RI), defined as a significant increase in maximum sustained winds over a short period, is especially difficult to forecast. RI may occur when environmental and oceanic factors are very favourable, presenting immense difficulties to traditional forecasting systems [2]. Classical numerical weather prediction (NWP) models are based on physical equations and data assimilation methods that model atmospheric and oceanic processes. Although these models offer physically consistent forecasts, they have limited capability to capture the complex and nonlinear interactions linked to tropical cyclone intensification owing to model resolution, parameterization uncertainty, and poor representation of ocean–atmosphere interactions [3]. Recent breakthroughs in the field of artificial intelligence (AI) offer a different solution, since machine learning models are capable of detecting complex correlations in extensive amounts of past atmospheric, oceanic, and satellite data. Nonetheless, purely data-based models can produce physically unsound predictions and perform poorly in situations that differ from their training data.

Hybrid AI-physics approaches are a possible solution, as they combine the pattern-recognition properties of AI with a well-understood physical understanding of tropical cyclone behaviour. Such models can incorporate predictors, constraints, or outputs of NWP systems to better reflect physical consistency while achieving improved predictive accuracy. This is especially applicable to RI forecasting, where the influences of sea-surface temperature, ocean heat content, vertical wind shear, atmospheric moisture, and upper-level atmospheric conditions interact in complex processes. This study thus examines hybrid AI-physics models to enhance rapid intensification predictions of tropical cyclones. The paper examines how the combination of machine learning methods and physics-based information can enhance the detection and forecasting of RI events. The study identifies whether physics-informed hybrid predictions are more reliable, accurate, and physically plausible in predicting changes in tropical cyclone intensity by comparing hybrid predictions with conventional prediction methods and standalone AI prediction.

Conclusion

This research examined how hybrid AI-physics models could be used to enhance the rapid intensification (RI) forecasting of tropical cyclones. Rapid intensification is a significant forecasting problem since cyclone intensity may shift at a wide scale due to the interaction of atmospheric conditions and oceanic conditions. Traditional numerical weather prediction gives physically consistent predictions but may be influenced by perturbations in initial conditions, model resolution, and physical parameterization. Conversely, AI models are able to detect more complex nonlinear relations in large meteorological datasets but fail to offer physically consistent forecasts when used beyond their training scenarios. The outcomes of this research illustrate the possibility of the fruitfulness of integrating these two strategies. The hybrid AI-physics system included the cyclone intensity data along with physically significant environmental variables such as sea-surface temperature, ocean heat content, vertical wind shear, and relative humidity. The comparative analysis revealed that the hybrid approach performed better in terms of RI classification compared to either the physics-informed baseline or the isolated AI model. It also led to a smaller mean absolute error and root-mean-square error of intensity change prediction. These data indicate that adding physical information is potentially able to supplement the pattern-recognizing ability of AI and enhance the accuracy of RI predictions.

The discussion also indicated that there are several interacting factors that affect RI as opposed to one environmental variable. Other most significant predictors were previous intensity change, ocean heat content, SST, atmospheric moisture, and vertical wind shear. This advocates the use of unified forecasting frameworks that can depict the statistical associations and known physical dynamics. However, the results must be analyzed in the context of the constraints of the employed dataset and model. The methodology should be tested in more tropical cyclone basins, more extensive historical records, and extreme cases of RI in future studies. Additional more complicated deep-learning frameworks, real-time satellite imaging, and fully coupled atmosphere-ocean models can also be examined in future work. All in all, hybrid AI-physics modelling is a potential research area in the development of more accurate, reliable, and physics-informed tropical cyclone rapid intensification forecasting systems.

Agriculture Journal IJOEAR Call for Papers

References
  1. Baisong, Y., Lihao, D., Xu, N., Yaxuan, L., & Yani, C. (2025). Maximum wave height prediction based on buoy data: Application of LightGBM and TCN-BiGRU. Journal of Marine Science and Engineering, *13*(10), 2009.
  2. Barr, B. W., & Chen, S. S. (2025). Impacts of seastate-dependent sea spray heat fluxes on tropical cyclone structure and intensity in fully coupled atmosphere-wave-ocean model simulations. Journal of Advances in Modeling Earth Systems, *17*(7), 33.
  3. Barros, L. O., Sapucci, L. F., Caroline, V., Ranieri, V. A., Baños, I. H., Bastarz, C. F., Vendrasco, E. P., Lopes, T. G., Almeida, S. S. S., de Mattos João G. Z., & Aravequia, J. A. (2026). Impact of GOES atmospheric motion vector data assimilation on forecasts over South America: Akará cyclone case study. Remote Sensing, *18*(11), 1799.
  4. Carmela, V., Scarcella, G. E., Loredana, A., Deborah, B., Francesco, C., & Roberto, C. (2026). Linking meteo-marine forcing and spatial damage patterns in Calabria after Cyclone Harry (Southern Italy). Earth, *7*(4), 129.
  5. Chen, B., Yu-Te Kuo, & Huang, T. (2023). A deep learning ensemble approach for predicting tropical cyclone rapid intensification. Atmospheric Science Letters, *24*(5).
  6. Dang, H. T. B., Roantree, M., & Tran, V. Q. (2026). How ocean thermal structure and atmospheric steering govern typhoon evolution: Evidence from interpretable machine learning. Environmental Research Communications, *8*(10), 101002.
  7. Deskos, G., Wang, J., Arwade, S., Fisher, M., Hirth, B., Larsén, X., Lundquist, J., Myers, A., Pang, W., Pringle, W., Rogers, R., Sanchez-Gomez, M., Sun, C., Yamaguchi, A., & Veers, P. (2026). Grand challenges in designing resilient wind energy systems in areas prone to tropical cyclones. Wind Energy Science, *11*(8), 2749-2782.
  8. Eusebi, R., Vecchi, G. A., Lai, C., & Tong, M. (2024). Realistic tropical cyclone wind and pressure fields can be reconstructed from sparse data using deep learning. Communications Earth & Environment, *5*(1), 8.
  9. Fancheng, M., Xiran, X., & Zhao, L. (2026). A time and frequency domain based dual-attention neural network for tropical cyclone track prediction. Applied Sciences, *16*(1), 436.
  10. Feng, C., & Pu, Z. (2023). The impacts of assimilating Aeolus horizontal line-of-sight winds on numerical predictions of Hurricane Ida (2021) and a mesoscale convective system over the Atlantic Ocean. Atmospheric Measurement Techniques, *16*(10), 2691-2708.
  11. Fry, M., Harrison, D., & King, J. (2026). An Arctic field campaign using stratospheric balloons for micro-dropsonde deployment: Insights and evaluation. Meteorological Applications, *33*(5), 13.
  12. Gensini, V. A. (2026). Extreme events and the insurance industry in a changing climate. Environmental Research Letters, *21*(7), 071001.
  13. Gi-Moon, Y., Zhu, J., Sun-Kwon, Y., Jong-Suk, K., & Young-Il, M. (2025). Improving typhoon-induced rainfall forecasts based on similar typhoon tracks. Applied Sciences, *15*(21), 11597.
  14. Huang, H., Difei, D., Hu, L., Chen, Y., & Sun, N. (2025). Beyond the backbone: A quantitative review of deep-learning architectures for tropical cyclone track forecasting. Remote Sensing, *17*(15), 2675.
  15. Jingyuan, P., Bosen, J., Qiuji, D., Lei, C., Chu, Z., Yueqin, S., & Liu, Y. (2026). Forecasting a hailstorm in Western China Plateau by assimilating XPAR radar network data with WRF-FDDA-HLHN. Remote Sensing, *18*(7), 968.
  16. Jin-Young, K., Albers, S., Sen, P., Hyun-Goo, K., Kim, K., & Su-Jin, H. (2022). The impact of assimilating winds observed during a tropical cyclone on a forecasting model. Atmosphere, *13*(8), 1302.
  17. Jung, C., Xue, P., Huang, C., Pringle, W., Biswas, M., Nain, G., & Wang, J. (2026). Fully coupled, high-resolution atmosphere–ocean–wave simulations of the offshore wind energy environment during Hurricane Henri (2021). Wind Energy Science, *11*(4), 1321-1341.
  18. Kay, J., Wang, X., & Yamamoto, M. (2022). An Observing System Simulation Experiment (OSSE) to study the impact of ocean surface observation from the Micro Unmanned Robot Observation Network (MURON) on tropical cyclone forecast. Atmosphere, *13*(5), 779.
  19. Khan, M. A., & Ulrich, F. (2026). Summertime increase in the frequency of low-pressure systems in the Mediterranean Region from 1940 to 2024. Climate, *14*(5), 93.
  20. Koteshwaramma, T., Singh, K. S., & Nayak, S. (2025). The performance of a high-resolution WRF modelling system in the simulation of severe tropical cyclones over the Bay of Bengal using the IMDAA Regional Reanalysis Dataset. Climate, *13*(1), 17.
  21. Lai, S. K., He, Y., Chan, P. W., Kerns, B. W., Chen, S. S., & Su, H. (2025). Towards skillful tropical cyclone forecasting by AI-model-driven high-resolution regional coupled model. Meteorological Applications, *32*(5), 12.
  22. Lin, Z., Chu, J., & Ham, Y. (2025). Enhancing tropical cyclone track and intensity predictions with the OWZP-Transformer model. NPJ Artificial Intelligence, *1*(1), 33.
  23. Liu, K., Xu, D., Zheng, F., He, J., Li, C., Leung, J., Zhang, M., Zhao, D., He, Q., Zhang, Y., Li, Y., & Zhang, B. (2026). Improvement of the computational efficiency in SVD-3DEnVar data assimilation scheme and its preliminary application to the TRAMS 3.0 model. Geoscientific Model Development, *19*(6), 2279-2297.
  24. Lu, X., Weng, Y., Liu, B., Zhang, Z., Wang, X., Cheng, J., Liu, S., Kleist, D., & Tallapragada, V. (2025). Implementation and evaluation of storm-following 3DIAU for hurricane intensity prediction improvements in operational HAFS. Earth and Space Science, *12*(10), 13.
  25. Lyu, X., & Wang, X. (2024). The impact of high-density airborne observations and atmospheric motion vector observation assimilation on the prediction of rapid intensification of Hurricane Matthew (2016). Atmosphere, *15*(4), 395.
  26. Manju, M. S., Pateriya, H., Gupta, R. K., Deepak, S. T., Gupta, P., & Butkovic, A. (2025). ConvLSTM-based tropical cyclone intensity estimation and classification using satellite imagery over the North Indian ocean. PLoS One, *20*(12), 25.
  27. Md, W., & Alea, Y. (2026). Climate variability in the South Pacific: A systematic review of key drivers and processes. Atmosphere, *17*(2), 147.
  28. Mohammad Mahboob, H. K., Ahmed, A. B., Dulmini, A., Rawfun, N. K., Azmain, D., Muhammad, A. S., Mahin, U. Q., Rahman, M., & Rois, R. (2026). Station-specific thunderstorm forecasting using climate predictors with seasonal decomposition analysis in Bangladesh's most vulnerable regions. PLoS One, *21*(9), 29.
  29. Montoro-Mendoza, A., Calvo-Sancho, C., González-Alemán, J. J., Díaz-Fernández, J., Bolgiani, P., & Martín, M. L. (2026). Strengthening of favorable environments for North Atlantic tropical cyclogenesis in midlatitudes in a warmer climate. NPJ Climate and Atmospheric Science, *9*(1), 45.
  30. Nadimpalli, R., Nekkali, Y. S., Osuri, K. K., Mohapatra, M., & Niyogi, D. (2025). Understanding the role of antecedent land conditions on rapid intensity changes in landfalling tropical cyclones over the Bay of Bengal. Meteorological Applications, *32*(6), 18.
Article Preview
Quick Actions
Download PDF View References
Article Info
Pages
120-130