All Issue

2026 Vol.59, Issue 2 Preview Page

Research Article

Atmospheric motion vector derivation technique using generative AI–based satellite image prediction

생성형 AI 기반 위성사진 예측을 활용한 대기운동벡터 산출 기법 연구

Dongjae Nama
,
Jongtak Anb
,
Byoungjoon Nac*
남 동재a
,
안 종탁b
,
나 병준c*
aUndergraduate Student, Department of Civil Engineering, Kumoh National Institute of Technology, Gumi, Korea
bUndergraduate Student, Department of Civil Engineering, Kumoh National Institute of Technology, Gumi, Korea
cAssistant Professor, Department of Civil Engineering, Kumoh National Institute of Technology, Gumi, Korea
a국립금오공과대학교 건축토목환경공학부 토목공학과 학사과정
b국립금오공과대학교 건축토목환경공학부 토목공학과 학사과정
c국립금오공과대학교 건축토목환경공학부 토목공학과 조교수
*Corresponding Author
28 February 2026. pp. 213-224
PDF XML Citation
Abstract

In this study, a method for deriving atmospheric motion vectors (AMVs) by combining generative AI–based satellite image frame prediction with a cross-correlation technique is proposed and validated using radiosonde observations. For training the generative adversarial network (GAN) model, infrared satellite images of 68 typhoons observed by the COMS geostationary satellite were used to predict satellite images up to 1 hour ahead. The predicted frames exhibited high pixel-wise correlation coefficients with the corresponding ground truth images (approximately 0.97–0.98), confirming reliable reconstruction of cloud structures. AMVs derived from the predicted consecutive frames showed overall reductions in mean vector difference (MVD) and root mean square error (RMSE) compared to numerical weather prediction (NWP) based wind vectors when validated against radiosonde observations. However, for typhoon cases characterized by low cloud brightness contrast, it was found that a training iteration number of at least approximately 700,000 was required to maintain a certain level of prediction accuracy.

Keywords
Generative Adversarial Network (GAN)
Atmospheric motion vector (AMV)
Satellite image prediction
Radiosonde observation

본 연구에서는 생성형 AI 기반 위성영상 프레임 예측과 상호연관 기법을 결합하여 대기운동벡터(AMV)를 산출하는 기법을 제시하고 라디오존데 관측값과 검증하였다. GAN 모델 학습을 위해 천리안 1호 태풍 위성영상(68개 태풍)을 학습해 최대 1시간 이후 영상을 예측하였다. AI 모델의 평가 사례들에서 예측된 프레임은 실제 촬영된 프레임과 높은 픽셀 상관(약 0.97~0.98)을 보여 구름 구조 재현성을 확인하였다. 예측된 연속 프레임들을 이용해 산출한 AMV는 라디오존데 관측과 비교했을 때 NWP 기반 벡터보다 MVD 및 RMSE가 전반적으로 감소하였다. 다만 구름 명암 변화가 낮은 사례에서는 정확도 유지를 위해 학습 반복횟수를 약 70만 회 이상 수행할 필요가 있음을 확인하였다.

키워드
생성적 적대 신경망(GAN)
대기운동벡터
위성영상 예측
라디오존데 관측
References
1

Alemany, S., Beltran, J., Perez, A., and Ganzfried, S. (2018). “Predicting hurricane trajectories using a recurrent neural network.” arXiv:1802.02548vol. 2.

2

Berger, H., Langland, R., Velden, C.S., Reynolds, C.A., and Pauley, P.M. (2011). “Impact of enhanced satellite-derived atmospheric motion vector observations on numerical tropical track forecasts in the western North Pacific during TPARC/TCS-08.” Journal of Applied Meteorology and Climatology, Vol. 50, pp. 2309-2318. doi: 10.1175/JAMC-D-11-019.1.

10.1175/JAMC-D-11-019.1
3

Chuang, W., Chou, C., Chang, K., Chang, Y., and Chin, H. (2019). “Atmospheric motion vectors derived from an infrared window channel of a geostationary satellite using particle image velocimetry.” Journal of Applied Meteorology and Climatology, Vol. 58, pp. 199-211. doi: 10.1175/JAMC-D-18-0105.1.

10.1175/JAMC-D-18-0105.1
4

Endlich, R.M., and Wolf, D.E. (1981). “Automatic cloud tracking applied to GOES and METEOSAT observations.” Journal of Applied Meteorology and Climatology, Vol. 20, No. 3, pp. 309-319. doi: 10.1175/1520-0450(1981)020<0309:ACTATG>2.0.CO;2.

10.1175/1520-0450(1981)020<0309:ACTATG>2.0.CO;2
5

Giffard-Roisin, S., Yang, M., Charpiat, G., Kegl, B., and Monteleoni, C. (2018). “Fused deep learning for hurricane track forecast from reanalysis data.” Climate Informatics Workshop Proceedings 2018, NCAR, Boulder, CO, US, pp. 69-72.

6

Goodfellow, I.J., Pouget-Abadie, J., Mirza, M., Xu, B., Warde-Farley, D., Ozair, S., Courville, A., and Bengio, Y. (2014). “Generative adversarial nets. Advances in Neural Information Processing Systems,Vol. 27, pp. 2672-2680.

7

Holland, G.J. (1980). “An analytic model of the wind and pressure profiles in hurricanes.” Monthly Weather Review, Vol. 108, pp. 1212-1218.

10.1175/1520-0493(1980)108<1212:AAMOTW>2.0.CO;2
8

Hong, S., Kim, S., Joh, M., and Song, A. (2017). “Globenet: convolutional neural networks for typhoon eye tracking from remote sensing imagery.” arXiv:1708.03417vol. 1.

9

Kordmahalleh, M.M., Sefidmazgi, M.G., and Homaifar, A. (2016). “A sparse recurrent neural network for trajectory prediction of atlantic hurricanes.” The Genetic and Evolutionary Computation Conference 2016, Denver, CO, U.S. pp. 957-964.

10.1145/2908812.2908834
10

Langland, R.H., Velden, C., Pauley, P.M., and Berger, H. (2009). “Impact of satellite-derived rapid-scan wind observations on numerical model forecasts of Hurricane Katrina.” Monthly Weather Review, Vol. 137, pp. 1615-1622.

10.1175/2008MWR2627.1
11

Le Marshall, J., Howard, D., Xiao, Y., Daniels, J., Wanzong, S., Jung, J., Bresky, W., Bailey, A., Tingwell, C., Le, T., and Margetic, D. (2017). “Himawari-8 atmospheric motion vectors-operational generation and assimilation.” Journal of Southern Hemisphere Earth Systems Science, Vol. 67, No. 1, pp. 12-24.

10.1071/ES17003
12

Lee, R.S.T., and Liu, J.N.K. (2000). “Tropical cyclone identification and tracking system using integrated neural oscillatory elastic graph matching and hybrid RBF network track mining techniques.” IEEE Trans. Neural Network, Vol. 11, pp. 680-689.

10.1109/72.846739
13

Lim, H.J., Chang, K.A., Huang, Z.C., and Na, B. (2015). “Experimental study on plunging breaking waves in deep water.” Journal of Geophysical Research: Oceans, Vol. 120, pp. 2007-2049.

10.1002/2014JC010269
14

Mathieu, M., Couprie, C., and LeCun, Y. (2015). “Deep multi-scale video prediction beyond mean square error, 05440.” arXiv: 1511.

15

Menzel, W.P. (2001). “Cloud tracking with satellite imagery: From the pioneering work of Ted Fujita to the present.” Bulletin of the American Meteorological Society, Vol. 82, pp. 33-48.

10.1175/1520-0477(2001)082<0033:CTWSIF>2.3.CO;2
16

Na, B., and Son, S. (2021). “Prediction of atmospheric motion vectors around typhoons using generative adversarial network.” Journal of Wind Engineering and Industrial Aerodynamics, Vol. 214, 104643.

10.1016/j.jweia.2021.104643
17

Na, B., Chang, K.-A., Huang, Z.-C., and Lim, H.-J. (2016). “Turbulent flow field and air entrainment in laboratory plunging breaking waves.” Journal of Geophysical Research: Oceans, Vol. 121, pp. 2980-3009.

10.1002/2015JC011377
18

Vega-Riveros, J.F., and Jabbour, K. (1989). “Review of motion analysis techniques.” IEE Proceedings I: Communications, Speech and Vision, Vol. 136, pp. 397-404.

10.1049/ip-i-2.1989.0060
19

Velden, C., Daniels, J., Stettner, D., Santek, D., Key, J., Dunion, J., Holmlund, K., Dengel, G., Bresky, W., and Menzel, P. (2005). “Recent innovations in deriving tropospheric winds from meteorological satellites.” Bulletin of the American Meteorological Society, Vol. 86, No. 2, pp. 205-224.

10.1175/BAMS-86-2-205
20

Zhang, Y., Chandra, R., and Gao, J. (2018). “Cyclone track prediction with matrix neural networks.” 2018 International Joint Conference on Neural Networks (IJCNN), IEEE, Rio de Janeiro, Brazil, pp. 1-8.

10.1109/IJCNN.2018.8489077
Information
  • Publisher :KOREA WATER RESOURECES ASSOCIATION
  • Publisher(Ko) :한국수자원학회
  • Journal Title :Journal of Korea Water Resources Association
  • Journal Title(Ko) :한국수자원학회 논문집
  • Volume : 59
  • No :2
  • Pages :213-224
  • Received Date : 2026-01-08
  • Revised Date : 2026-01-21
  • Accepted Date : 2026-01-24
  • DOI :https://doi.org/10.3741/JKWRA.2026.59.2.213
Journal Informaiton Journal of Korea Water Resources Association Journal of Korea Water Resources Association
Online Submission
Archives
  • scopus
  • KCI-수자원
  • NRF
  • KOFST
  • KISTI Current Status
  • KISTI Cited-by
  • crosscheck
  • orcid
  • open access
  • ccl
Journal Informaiton Journal Informaiton - close