Years ago, marine expert Eric Prince discovered an anomaly while tracking fish tags: blue marlins typically dive to a depth of 800 meters in the southeastern United States to hunt, while in Costa Rican waters they only operate at the surface of the ocean. Why didn't the marlin, a diving expert, suddenly stop diving? Eric Prince, a longtime long-time study of longmouth fish, was determined to investigate this anomaly.
In fact, this is the self-protection response of marine life in the face of changes in the marine environment, and Eric Prince's exploration is to further promote the phenomenon of "marine hypoxia" behind it to the public - it is precisely because the deep oxygen content in Costa Rican waters is declining, and the hypoxic area is gradually expanding, and the marlin has to move on the surface to avoid suffocation.
The area where a large number of marine organisms die due to hypoxia is often referred to as the marine "death zone", but in fact, the negative effects of marine hypoxia are not only for marine life itself, but also for fisheries and even the social economy. Nowadays, with the intensification of global environmental problems, the lack of oxygen in the oceans is also intensifying.
In 2019, the International Union for Conservation of Nature (IUCN) reported that the current ocean areas with low oxygen concentrations are expanding, and that there are now more than 600 areas affected by low oxygen conditions, compared to 45 hypoxic marine zones in the 60s of the last century. The report notes that the number of oxygen-deficient waters in the global ocean has increased fourfold over the same period.
In order to better maintain the health of marine ecosystems and protect fishery resources, it is essential to carry out scientific measurement of ocean dissolved oxygen levels.
However, the high cost of ocean field observations, the uneven spatial distribution of existing observation data, and the variety of measurement methods for dissolved oxygen concentrations and the large differences in the quality of the obtained data bring certain challenges to the study of the changes in dissolved oxygen levels in the global ocean.
In response, researchers from the GIS Laboratory of Zhejiang University proposed a new approach to combine machine learning techniques with satellite products to develop a comprehensive modeling framework for global dissolved oxygen at sea surface, the DOsurface-Pred Framework, based on which a large-scale SSDO dataset spanning 2010-2018 was generated. The results of the study show a downward trend in dissolved oxygen levels even in oxygen-rich sea surface areas, and this decline is mainly due to changes in sea surface temperature (SST).
Research Highlights:
* A comprehensive modelling framework for global dissolved oxygen at sea surface is proposed
* The SHAP interpreter was introduced, identifying key variables and their impact on predicting dissolved oxygen outcomes
* It is helpful to understand the high dynamics of dissolved oxygen in the global ocean, and to explore the law of deoxygenation and its causes
Address:
https://pubs.acs.org/doi/10.1021/acs.est.3c08833
SSDO dataset address:
https://go.hyper.ai/BBlqA
Follow the official account, and reply to "Sea Surface Dissolved Oxygen" in the background to get the full PDF
Dataset: Satellite + field measurement data
The data used in this study included satellite data, fixed vessel and conductivity meter field measurements.
Satellite datasets include sea surface temperature (SST), sea surface salinity (SSS), chlorophyll-a (Chl-a), sea surface wind (SSW), and sea level anomalies (SLA). SST data from the OISST dataset, SSS data from the European Space Agency (ESA)'s CCI project, Chl-a data from the MODIS Aqua and Terra satellites, SSW data from the VAM method combined with multi-satellite microwave wind and instrument observations with cross-platform cross-calibration, and SLA data from the AVISO project.
Fixed vessel and conductivity meter field measurements are sourced from the Ocean Station Data (OSD) and the high-resolution Conductivity-Temperature-Depth (CTD) subdatabase in the World Oceans Database (WOD) 2018.
The field measurement data used in this study is shown in the figure, with a total of 28,044 records, including 241 records from 2019.
Data distribution and correlation analysis of field measurements
(a) Spatial distribution of field measurement data
(b) Temporal distribution of field measurement data
(c) Distribution of data from field measurements in different oceans
(d) Correlation matrix analysis between the seven major marine hydrological variables
模型构架：DOsurface-Pred Framework 的三大组成部分
An interpretable spatiotemporal machine learning framework for dissolved oxygen in the global ocean
研究人员提出了一种时空信息嵌入的可解释机器学习框架。 该框架由三个主要部分组成：Spatio-temporal Information Embedding Module、Backbone Regression Module 和 SHAP Explainer Module。
The first part, spatiotemporal information data measured with multiple satellites and in situ, i.e., sample Xi= {Xi spatial,
Xi temporal, Xi satellite1, ……, Xi satellite n} 作为输入。 通过时空信息嵌入模块，将空间和月份信息转换为极坐标表示，转换后得到 X'i={X'i spatial, X'i temporal, X'i satellite1, ……, X'i satellite n} 的全局表示。
Subsequently, X'i was passed on to the second part. Evaluate different models with multi-step gridsearch cross-validation.
Performance evaluation table of different models
上表是 backbone models 的性能评价表，与未采用此框架的模型相比，DOsurface-Pred 框架的性能得到了提升。
此外，所有基于树形结构的机器学习模型 (tree-based machine-learning models) 在性能上均优于 benchmark 多元线性回归模型（multiple linear regression model）。 这些模型按性能降序排列如下：ET、RF、GBDT、XGBoost、MLP。 其中，ET 模型在所有评估指标上表现最佳，RMSE 值 为 11.67 μmol/kg，该模型在溶解氧 (DO) 建模中具备更好的泛化能力，可以降低模型的过拟合现象。
In the third part, the researchers utilize the SHAP interpretability method to evaluate the effect of sample eigenvalues on the model output. Through this framework, the prediction results are generated under the optimal model.
SHAP quantifies the contribution of different eigenvalues to the prediction results, making the model's output easier to understand to identify key variables and their impact on the dissolved oxygen prediction results.
Conclusion: SST is the main cause of dissolved oxygen content at sea surface
Based on the DOsurface-Pred Framework, which provides an accurate assessment of global sea surface dissolved oxygen concentrations, researchers have generated a large-scale sea surface dissolved oxygen dataset spanning 2010-2018, which the researchers called the SSDO product.
Uncertainty estimates for DO projections
(a,b) 均方根误差 (RMSE) 和偏差误差 (biaserror)
(c,d) 不同年、月下总不确定性及其组成 (M、R、P)
To evaluate the results generated by the optimal model, the researchers performed uncertainty estimation and spatiotemporal validation.
First, the error and uncertainty of the SSDO product are evaluated. The experimental results show that the total uncertainty of the three errors (measurement error M, representation error R, and prediction error P) is estimated to be ±13.02 μmol/kg.
Evaluation of Independent Buoy (PFL) measurements
(a) Scatter density plots of predicted and PFL measurements
(b) The matching PFL point spatial location, with the serial number corresponding to the plot number provided below (c-h).
(c-h) Temporal trend of oxygen anomalies at each location, compared to PFL measurement outliers
Second, the researchers further tested the accuracy and time series of SSDO using an independent buoy measurement dataset. The results showed that the predicted results had a good fit with the PFL database with an R² value of 0.86.
In addition, the researchers conducted a comparative assessment of the location of long-term buoy observations in different ocean areas, and the results showed that the predictions were consistent with the spatial variations and long-term trends measured by the buoys. These evaluations reliably validate the predictions and help analyze their applicability to different scenarios.
The researchers also performed statistical analyses of SSDO products. The results show that the SSDO data has a similar spatial distribution to the long-term recorded data of WOD. Under the influence of the continuous expansion of the hypoxic zone, the dissolved oxygen level at the sea surface is also undersaturated, and the dissolved oxygen level at the sea surface shows an average annual decreasing trend of 0.22 μmol/kg even in the sea surface with sufficient dissolved oxygen exchange. In addition, the interannual variation of dissolved oxygen at sea surface shows a correlation with typical ocean change phenomena.
Dissolved oxygen modelling factor explainability analysis
(a) Assess the impact of global characteristic mean on model outputs
(b) Evaluate the influence of local features on the model outputs
(c,d) Analyze the impact of SST and SSS on model output
(e,f) Spatial distribution of SST and SSS trait influences
Using the SHAP interpretability approach, the researchers further revealed the driving mechanism of oceanic factors such as thermohaline on dissolved oxygen at the sea surface. Among them, temperature (SST) and salinity (SSS) were the main control factors, which had a negative impact on dissolved oxygen, and SST had the greatest impact on dissolved oxygen levels. This assessment helps to improve the reliability of the modeling, and provides an explainable quantitative result for exploring the temporal and spatial distribution of ocean dissolved oxygen and the causes of deoxygenation.
In summary, the researchers used the DOsurface-Pred framework to generate the SSDO dataset and introduced the SHAP interpretability method to confirm a downward trend in dissolved oxygen content even in oxygen-abundant sea surface areas, and this decline was mainly attributed to SST changes.
AI and the ocean, explore the mysteries of the unknown deep sea
Dan Laffoley, Senior Advisor for Marine Science and Conservation at IUCN's Global Ocean and Polar Programme, has said, "Dissolved oxygen concentrations in the ocean are declining, adding to the pressure of warming and acidification for marine ecosystems. 」
At present, in addition to accelerating global and industry-wide carbon emission reduction, through in-depth research on the phenomenon of ocean hypoxia, the analysis of the important factors affecting climate on life on the earth can also allow people to have a more accurate understanding of the reasons for the extinction or survival of different organisms, and further "prescribe the right medicine".
In this process, the capabilities of emerging technologies such as the Internet of Things and AI have gradually been applied more deeply. In addition to the above-mentioned framework for the analysis of dissolved oxygen in sea markers, some scholars have also carried out research on marine litter control and marine biodiversity conservation based on marine data such as satellite remote sensing, unmanned aerial vehicles, buoys, sonar, and underwater optics.
For example, in 2019, the United Nations Environment Programme (UNEP) released CounterMEASURE, a countermeasure project on plastic pollution in the Mekong and Ganges basins, which uses drones to take aerial imagery of the Mekong River basin and develops an identification and monitoring model to pinpoint the sources and paths of plastic waste in the Mekong River basin with 83.9% accuracy.
In addition, Australia's The Ripper Group has also developed a product that uses drone + machine vision technology to identify sharks. This product is not only used to avoid being disturbed by sharks in coastal areas, but also plays an important role in protecting the shark ecology.
It is true that the high carbon emissions of human society have had a serious impact on the global climate and ecological environment, among which, as the origin of life on the earth, the protection of the marine ecological environment is also urgent.
Resources:
1.https://mp.weixin.qq.com/s/bUbYptqccBXC2T9dvkfOfA
2.https://www.cdstm.cn/gallery/hycx/qyzx/201909/t20190904_923957.html
3.http://www.cbcgdf.org/NewsShow/4854/10658.html