All Issue

2019 Vol.24, Issue 3

Article

Estimation of Surface fCO2 in the Southwest East Sea using Machine Learning Techniques

기계학습법을 이용한 동해 남서부해역의 표층 이산화탄소분압(fCO2) 추정

DOSHIK HAHM1*
,
SOYEONA PARK2
,
SANG-HWA CHOI2
,
DONG-JIN KANG23
,
TAEKEUN RHO4
,
TONGSUP LEE1
함 도식1*
,
박 소예나2
,
최 상화2
,
강 동진23
,
노 태근4
,
이 동섭1
1Department of Oceanography, Pusan National University
2Marine Environmental & Climate Research Division, Korea Institute of Ocean Science & Technology
3Department of Oceanography, University of Science &Technology
4Instrumental Development & Management Center, Korea Institute of Ocean Science & Technology
1부산대학교 해양학과
2한국해양과학기술원 해양환경·기후연구본부
3과학기술연합대학원 해양학전공
4한국해양과학기술원 해양기기개발·운영센터
* Corresponding Author
31 August 2019. pp. 375-388
PDF XML Citation
Abstract

지구의 탄소순환을 이해하고 미래 대기 CO2의 농도와 기후 변화를 예측하기 위해서는 해양과 대기 사이 CO2 교환율(sea-to-air CO2 flux)의 시공간 변화를 정확하게 추정하는 것이 필요하다. 연구선을 이용한 현장 관측이 갖고 있는 시공간 제약으로 인해 동해에는 매우 제한적인 표층 이산화탄소분압(fCO2) 자료만 존재한다. 이 연구에서는 위성 및 수치모형에서 얻은 수온, 염분, 엽록소, 혼합층 자료를 세 종류의 기계학습 모형에 입력하여 동해 남서부해역의 고해상도 표층 fCO2 시계열 자료를 산출하였다. 세 모형 중 현장 관측 자료를 가장 잘 재현하는 Random Forest (RF) 모형의 평균제곱근오차는 7.1 μatm이었다. RF 모형을 이용한 fCO2 예측에 중요한 역할을 하는 변수는 수온, 염분과 시간 정보였으며, 엽록소와 혼합층 깊이는 fCO2 예측에 미미한 역할을 하였다. RF 모형에서 예측한 표층 fCO2를 이용하여 계산한 동해 남서부해역의 CO2 교환율은 -0.76 ± 1.15 mol m-2 yr-1로 이전 현장 관측 연구에서 제시한 교환율( -0.66 ~ -2.47 mol m-2 yr-1) 범위 중 작은 값에 해당한다. RF 모형의 표층 fCO2 시계열 자료는 1주일 내외의 짧은 시간 사이에도 CO2 교환율이 상당히 변할 수 있음을 보여주었다. 앞으로 보다 정확한 CO2 교환율 산출을 위해서는 fCO2가 급격하게 변화하는 봄철에 높은 해상도의 현장 관측을 수행할 필요가 있다.

Accurate evaluation of sea-to-air CO2 flux and its variability is crucial information to the understanding of global carbon cycle and the prediction of atmospheric CO2 concentration. fCO2 observations are sparse in space and time in the East Sea. In this study, we derived high resolution time series of surface fCO2 values in the southwest East Sea, by feeding sea surface temperature (SST), salinity (SSS), chlorophyll-a (CHL), and mixed layer depth (MLD) values, from either satellite-observations or numerical model outputs, to three machine learning models. The root mean square error of the best performing model, a Random Forest (RF) model, was 7.1 μatm. Important parameters in predicting fCO2 in the RF model were SST and SSS along with time information; CHL and MLD were much less important than the other parameters. The net CO2 flux in the southwest East Sea, calculated from the fCO2 predicted by the RF model, was -0.76 ± 1.15 mol m-2 yr-1, close to the lower bound of the previous estimates in the range of -0.66 ~ -2.47 mol m-2 yr-1. The time series of fCO2 predicted by the RF model showed a significant variation even in a short time interval of a week. For accurate evaluation of the CO2 flux in the Ulleung Basin, it is necessary to conduct high resolution in situ observations in spring when fCO2 changes rapidly.

Keywords
Machine learning
fCO2
East sea
Random forest

References
1
Breiman, L., 2001. Random Forests. Machine Learning, 45(1): 5-32.
10.1023/A:1010933404324
2
Choi, S.-H., D. Kim, J. Shim, K.H. Kim, H.S. Min and K.-R. Kim, 2012. Seasonal Variations of Surface fCO2 and Sea-Air CO2 Fluxes in the Ulleung Basin of the East/Japan Sea. Terrestrial, Atmospheric and Oceanic Sciences, 23(3): 343-353.
10.3319/TAO.2012.01.19.01(Oc)
3
Ciais, P., C. Sabine, G. Bala, L. Bopp, V. Brovkin, J. Canadell, A. Chhabra, R. DeFries, J. Galloway, M. Heimann, C. Jones, C. Le Quéré, R.B. Myneni, S. Piao and P. Thornton, 2013. Carbon and Other Biogeochemical Cycles. In: Cli- mate Change 2013: The Physical Science Basis. Contribution of Working Group I to the Fifth Assessment Report of the Intergovernmental Panel on Climate Change [Stocker, T.F., D. Qin, G.-K. Plattner, M. Tignor, S.K. Allen, J. Boschung, A. Nauels, Y. Xia, V. Bex and P.M. Midgley (eds.)]. Cambridge University Press, Cambridge, United Kingdom and New York, NY, USA.
4
de Boyer Montegut, C., G. Madec, A.S. Fischer, A. Lazar and D.N. Iudicone, 2004. Mixed layer depth over the global ocean: An examination of profile data and a profile-based climatology. Journal of Geophysical Research-Oceans, 109(C12): C12003.
10.1029/2004JC002378
5
Dickson, A.G., C.L. Sabine and J.R. Christian, 2007. Guide to Best Practices for Ocean CO2 measurements, 1-196 pp.
6
Gregor, L., S. Kok and P.M.S. Monteiro, 2017. Empirical methods for the estimation of Southern Ocean CO2: support vector and random forest regression. Biogeosciences, 14(23): 5551-5569.
10.5194/bg-14-5551-2017
7
Hahm, D., T.S. Rhee, H.-C. Kim, C.J. Jang, Y.S. Kim and J.-H. Park, 2019. An observation of primary production enhanced by coastal upwelling in the southwest East/Japan Sea. Journal of Marine Systems, 195: 30-37.
10.1016/j.jmarsys.2019.03.005
8
Hornik, K., 1991. Approximation capabilities of multilayer feed- forward networks. Neural Networks, 4: 251-257.
10.1016/0893-6080(91)90009-T
9
Jang, E., J. Im, G.-H. Park and Y.-G. Park, 2017. Estimation of Fugacity of Carbon Dioxide in the East Sea Using In Situ Measurements and Geostationary Ocean Color Imager Satellite Data. Remote Sensing, 9(8): 821-823.
10.3390/rs9080821
10
Kim, J.Y., D.J. Kang, T. Lee and K.R. Kim, 2014. Long-term trend of CO2 and ocean acidification in the surface water of the Ulleung Basin, the East/Japan Sea inferred from the underway observational data. Biogeosciences, 11(9): 2443-2454.
10.5194/bg-11-2443-2014
11
Kwak, J.H., S.H. Lee, H.J. Park, E.J. Choy, H.D. Jeong, K.R. Kim and C.K. Kang, 2013. Monthly measured primary and new productivities in the Ulleung Basin as a biological "hot spot" in the East/Japan Sea. Biogeosciences, 10(7): 4405-4417.
10.5194/bg-10-4405-2013
12
Landschützer, P., N. Gruber, D.C.E. Bakker and U. Schuster, 2014. Recent variability of the global ocean carbon sink. Global Biogeochemical Cycles, 28(9): 927-949.
10.1002/2014GB004853
13
Muller, A. and S. Guido, 2017. Introduction to Machine Learning with Python. O'Rielly.
14
Na, J.-Y., J.-W. Seo and S.-K. Han, 1992. Monthly mean sea surface winds over the adjacent seas of the Korea Peninsula. J. Oceangr. Soc. Korea, 27: 1-10.
15
Nakaoka, S., M. Telszewski, Y. Nojiri, S. Yasunaka, C. Miyazaki, H. Mukai and N. Usui, 2013. Estimating temporal and spatial variation of ocean surface pCO2 in the North Pacific using a self-organizing map neural network technique. Biogeosciences, 10(9): 6093-6106.
10.5194/bg-10-6093-2013
16
Oh, D.-C., 1998. A study on the characteristics of fCO2 distributions and CO2 flux at the air-sea interface in the seas around Korea. MS Thesis Seoul National University, 105 p.
17
Oh, D.-C., M.-K. Park, S.-H. Choi, D.-J. Kang, S.Y. Park, J.S. Hwang, A. Andreev, G.H. Hong and K.-R. Kim, 1999. The Air-Sea Exchange of CO2 in the East Sea (Japan Sea). Journal of Oceanography, 55(2): 157-169.
10.1023/A:1007833811440
18
Park, G.-H., K. Lee, P. Tishchenko, D.-H. Min, M.J. Warner, L.D. Talley, D.J. Kang and K.R. Kim, 2006. Large accumulation of anthropogenic CO 2in the East (Japan) Sea and its significant impact on carbonate chemistry. Global Biogeochemical Cycles, 20(4): GB4013.
10.1029/2005GB002676
19
Park, S., T. Lee and Y.-H. Jo, 2016. Sea Surface pCO2 and Its Variability in the Ulleung Basin, East Sea Constrained by a Neural Network Model. The Sea, 21(1): 1-10.
10.7850/jkso.2016.21.1.1
20
Pedregosa, F., G. Varoquaux, A. Gramfort, V. Michel, B. Thirion, O. Grisel, M. Blondel, P. Prettenhofer, R. Weiss, V. Dubourg, J. Vanderplas, A. Passos, D. Cournapeau, M. Brucher, M. Perrot and E. Duchesnay, 2011. Scikit-learn: Machine learning in Python. Journal of Machine Learning Research, 12: 2825-2830.
21
Pierrot, D., C. Neill, K. Sullivan, R. Castle, R. Wanninkhof, H. Lüger, T. Johannessen, A. Olsen, R. Feely and C.E. Cosca, 2009. Recommendations for autonomous underway pCO2 measuring systems and data-reduction routines. Deep-Sea Research Part Ii-Topical Studies in Oceanography, 56(8-10): 512-522.
10.1016/j.dsr2.2008.12.005
22
Takahashi, T., J. Olafsson, J.G. Goddard, D.W. Chipman and S.C. Sutherland, 1993. Seasonal variation of CO2 and nutrients in the high-latitude surface oceans: A comparative study. Global Biogeochemical Cycles, 7(4): 843-878.
10.1029/93GB02263
23
Takahashi, T., S. Sutherland and R. Wanninkhof, 2009. Climatological mean and decadal change in surface ocean pCO2, and net sea-air CO2 flux over the global oceans. Deep-Sea Research, 56(8-10): 554-577.
24
Telszewski, M., A. Chazottes, U. Schuster, A.J. Watson, C. Moulin, D.C.E. Bakker, M. González-Dávila, T. Johannessen, A. Körtzinger, H. Lüger, A. Olsen, A. Omar, X. A. Padin, A. F. Ríos, T. Steinhoff, M. Santana-Casiano, D. W. R. Wallace and R. Wanninkhof, 2009. Estimating the monthly pCO2 distribution in the North Atlantic using a self-organizing neural network. Biogeosciences, 6(8): 1405-1421.
10.5194/bg-6-1405-2009
25
Vapnik, V., 2000. The Nature of Statistical Learning Theory. 2nd ed., Springer, New York.
10.1007/978-1-4757-3264-1
26
Wanninkhof, R., 1992. Relationship between wind speed and gas exchange. Journal of Geophysical Research, 97(C5): 7373-7382.
10.1029/92JC00188
27
Wanninkhof, R., 2014. Relationship between wind speed and gas exchange over the ocean revisited. Limnol. Oceanogr.: Methods, 12(6): 351-362.
10.4319/lom.2014.12.351
28
Yoo, S. and J. Park, 2009. Why is the southwest the most productive region of the East Sea/Sea of Japan?, Journal of Marine Systems, 78(2): 15-15.
10.1016/j.jmarsys.2009.02.014
29
Zeng, J., Y. Nojiri, P. Landschützer, M. Telszewski and S. Nakaoka, 2014. A Global Surface Ocean fCO2 Climatology Based on a Feed-Forward Neural Network. Journal of Atmospheric and Oceanic Technology, 31(8): 1838-1849.
10.1175/JTECH-D-13-00137.1
Information
  • Publisher :The Korean Society of Oceanography
  • Publisher(Ko) :한국해양학회
  • Journal Title :The Sea Journal of the Korean Society of Oceanography
  • Journal Title(Ko) :한국해양학회지 바다
  • Volume : 24
  • No :3
  • Pages :375-388
  • Received Date : 2019-06-04
  • Revised Date : 2019-07-22
  • Accepted Date : 2019-07-23
  • DOI :https://doi.org/10.7850/jkso.2019.24.3.375
Journal Informaiton The Sea Journal of the Korean Society of Oceanography The Sea Journal of the Korean Society of Oceanography
  • NRF
  • KOFST
  • KISTI Current Status
  • KISTI Cited-by
  • crosscheck
  • orcid
  • open access
  • ccl
Journal Informaiton Journal Informaiton - close