Time-lag correction for oceanic in situ CO₂ sensors using a physics-informed hybrid BiLSTM-attention and Kalman filter framework

Abstract

High spatiotemporal-resolution observations of the oceanic partial pressure of carbon dioxide (pCO₂) are essential for advancing our understanding of the global carbon cycle. Nonetheless, oceanic in situ pCO₂ sensors mounted on mobile platforms frequently encounter challenges such as a physical response lag when crossing concentration gradients. To address this issue and consider the inherent noise of the sensors as well as data heterogeneity, including inconsistent data lengths, we propose a physics-informed deep learning framework for sensor time-lag correction. A single-layer bidirectional long short-term memory (BiLSTM) network coupled with an attention mechanism was designed to capture transient concentration gradients without oversmoothing the signal. A Kalman filter was then applied as a post-processing module to suppress the prediction noise without inducing a secondary physical lag. The experimental results demonstrate that this framework corrects the phase delay while minimizing the residual errors. Quantitative results show that the proposed method outperforms the physical linear time-invariant (LTI) model, linear autoregressive with exogenous input (ARX) model, and traditional statistical baselines (e.g., support vector regression, SVR). It dynamically corrects phase delay and reduces system response time by more than 80% without amplifying high-frequency noise, thereby offering a computationally efficient solution for continuous ocean observation.

Keywords

Marine CO₂ sensor / Time-lag correction / Physics-informed deep learning / Kalman filter / BiLSTM-attention

References

[1] Bittig HC, Fiedler B, Scholz R, Krahmann G, Körtzinger A. Time response of oxygen optodes on profiling platforms and its dependence on flow speed and temperature. Limnol Oceanogr Methods, 2014, 12: 617-636.

[2] Doney SC, Fabry VJ, Feely RA, Kleypas JA. Ocean acidification: the other CO₂ problem. Annu Rev Mar Sci, 2009, 1: 169-192.

[3] Fawaz HI, Forestier G, Weber J, Idoumghar L, Muller PA. Deep learning for time series classification: a review. Data Min Knowl Discov, 2019, 33: 917-963.

[4] Fiedler B, Fietzek P, Vieira N, Silva P, Bittig HC, Körtzinger A. In situ CO₂ and O₂ measurements on a profiling float. J Atmos Ocean Technol, 2013, 30: 112-126.

[5] Fietzek P, Fiedler B, Steinhoff T, Körtzinger A. In situ quality assessment of a novel underwater pCO₂ sensor based on membrane equilibration and NDIR spectrometry. J Atmos Ocean Technol, 2014, 31: 181-196.

[6] Gordon C, Fennel K, Richards C, Shay LK, Brewster JK. Can ocean community production and respiration be determined by measuring high-frequency oxygen profiles from autonomous floats?. Biogeosciences, 2020, 17: 4119-4134.

[7] Gruber N, Bakker DCE, DeVries T, Gregor L, Hauch J, Landschützer P, et al.. Trends and variability in the ocean carbon sink. Nat Rev Earth Environ, 2023, 4: 119-134.

[8] Hewage P, Trovati M, Pereira E, Behera A. Deep learning-based effective fine-grained weather forecasting model. Pattern Anal Appl, 2021, 24: 343-366.

[9] Li M, Du BL, Guo JJ, Zhang ZH, Lu ZY, Zheng RE. A low-cost in-situ CO₂ sensor based on a membrane and NDIR for long-term measurement in seawater. J Ocean Limnol, 2022, 40: 986-998.

[10] Lim B, Zohren S. Time-series forecasting with deep learning: a survey. Philos Trans A Math Phys Eng Sci, 2021, 379. ArticleID: 20200209

[11] Lin H, Zhang S, Li Q, Li Y, Li J, Yang Y. A new method for heart rate prediction based on LSTM-BiLSTM-Att. Measurement, 2023, 207. ArticleID: 112384

[12] Mihaylov A, El Naggar H. A comparison of instrument response correction methods: post-processing and real-time methods. Results Geophys Sci, 2021, 8. ArticleID: 100033

[13] Niu Z, Zhong G, Yu H. A review on the attention mechanism of deep learning. Neurocomputing, 2021, 452: 48-62.

[14] Rajwar K, Deep K, Das S. An exhaustive review of the metaheuristic algorithms for search and optimization: taxonomy, applications, and open challenges. Artif Intell Rev, 2023, 56: 13187-13257.

[15] Roemmich D, Johnson GC, Riser S, Davis R, Gilson J, Owens WB, et al.. The Argo Program: observing the global ocean with profiling floats. Oceanography, 2009, 22: 34-43.

[16] Rudnick DL. Ocean research enabled by underwater gliders. Annu Rev Mar Sci, 2016, 8: 519-541.

[17] Sagheer A, Kotb M. Time series forecasting of petroleum production using deep LSTM recurrent networks. Neurocomputing, 2019, 323: 203-213.

[18] Sarmiento JL, Johnson KS, Arteaga LA, Bushinsky SM, Cullen HM, Gray AR, et al.. The Southern Ocean carbon and climate observations and modeling (SOCCOM) project: a review. Prog Oceanogr, 2023, 219. ArticleID: 103130

[19] Sherstinsky A. Fundamentals of recurrent neural network (RNN) and long short-term memory (LSTM) network. Phys D, 2020, 404. ArticleID: 132306

[20] Song T, Pang C, Hou B, Xu G, Xue J, Sun H, et al.. A review of artificial intelligence in marine science. Front Earth Sci, 2023, 11. ArticleID: 1090185

[21] Sutton AJ, Feely RA, Sabine CL, McPhaden MJ, Takahashi T, Chavez FP, et al.. Natural variability and anthropogenic change in equatorial Pacific surface ocean p CO₂ and pH. Glob Biogeochem Cycles, 2014, 28: 131-145.

[22] Tian X, Zhou HX, Zhang HW, Zhang LH, Shi L. LSTM & attention-based meta-reinforcement learning for trajectory tracking of underwater gliders with varying buoyancy loss and current disturbance. Ocean Eng, 2025, 326. ArticleID: 120906

[23] Tomczyk K, Kowalczyk M, Ostrowska K. Procedure proposal for minimising the dynamic error of second-order sensors. Sensors, 2022, 22. ArticleID: 1901

[24] Wang S, Rußwurm M, Körner M, Lobell DB (2020) Meta-learning for few-shot time series classification. In: IGARSS 2020—2020 IEEE International Geoscience and Remote Sensing Symposium. IEEE, pp 7041–7044. https://doi.org/10.1109/IGARSS39084.2020.9441016

[25] Wang Y, Zhang Q, Wang GG, Cheng HL. The application of evolutionary computation in generative adversarial networks (GANs): a systematic literature survey. Artif Intell Rev, 2024, 57: 182.

[26] Wanninkhof R, Park GH, Takahashi T, Sweeney C, Feely R, Nojiri Y, et al.. Global ocean carbon uptake: magnitude, variability and trends. Biogeosciences, 2013, 10: 1983-2000.

[27] Weber M, Auch M, Doblander C, Mandl P, Jacobsen HA. Transfer learning with time series data: a systematic mapping study. IEEE Access, 2021, 9: 165409-165432.

[28] Wilson G, Cook DJ. A survey of unsupervised deep domain adaptation. ACM Trans Intell Syst Technol, 2020, 11: 1-46.

[29] Xiang Z, Yan J, Demir I. A rainfall-runoff model with LSTM-based sequence-to-sequence learning. Water Resour Res, 2020, 56. ArticleID: e2019WR025326

[30] Zhang X, Du ZF, Zheng R, Luan ZD, Qi FJ, Cheng K, et al.. Development of a new deep-sea hybrid Raman insertion probe and its application to the geochemistry of hydrothermal vent and cold seep fluids. Deep Sea Res Part I-Oceanogr Res Pap, 2017, 123: 1-12.

[31] Zhang ZH, Li M, Yang T, Zang ZX, Li N, Zheng R, et al.. Seconds-scale response sensor for in situ oceanic carbon dioxide detection. Anal Chem, 2023, 95(7): 3577-3586.

[32] Zhu XS, Lai ZL, Liang YR, Xiong JP, Wu JH. Generative high-capacity image hiding based on residual CNN in wavelet domain. Appl Soft Comput, 2022, 115. ArticleID: 108170

[33] Zhuang FZ, Qi ZY, Duan KY, Xi DB, Zhu YC, Zhu HS, et al.. A comprehensive survey on transfer learning. Proc IEEE, 2021, 109(1): 43-76.

Funding

National Natural Science Foundation of China(42206194)

Zhejiang Provincial Natural Science Foundation of China(LMS25D060005)

Open Fund Project of Key Laboratory of Ocean Observation Technology, MNR(2024klootA02)

Ministry-Province Cooperative Project under the Ministry of Natural Resources(2024ZRBSHZ102)

Special Project of the Central Government in Guidance of Local Science and Technology Development(2025ZY01111)

RIGHTS & PERMISSIONS

The Author(s)