PDF(5177 KB)
高空间分辨率海表CO2分压的卫星遥感反演算法:机器学习在秋季象山港的应用
刘婷宇, 白雁, 朱伯仲, 李腾, 龚芳
海洋学研究 ›› 2023, Vol. 41 ›› Issue (1) : 82-95.
PDF(5177 KB)
PDF(5177 KB)
高空间分辨率海表CO2分压的卫星遥感反演算法:机器学习在秋季象山港的应用
Satellite retrieval algorithm of high spatial resolution sea surface partial pressure of CO2: Application of machine learning in Xiangshan Bay in autumn
近海海湾受人类活动及自然变化影响大,海水碳源汇格局变化影响机制极其复杂。由于海湾空间尺度小,需要使用宽波段的高空间分辨率卫星遥感对海-气CO2通量进行监测评估。相对于传统公里级的水色卫星资料,海-气CO2通量定量估算的关键参数——海表CO2分压(sea surface partial pressure of CO2, pCO2)遥感反演在小尺度海湾具有极大的挑战性。该文以秋季象山港为例,利用走航观测pCO2数据及近5年哨兵2号(Sentinel-2)卫星影像,采用支持向量机(support vector machine, SVM)机器学习的方法,基于Sentinel-2遥感反射率及其比值,建立了海表pCO2的遥感反演算法。算法验证结果显示决定系数为0.92,均方根误差为23.23 μatm,遥感反演结果与实测值具有较高一致性。在此基础上,制作了2017—2021年秋季(9—11月)象山港海表pCO2遥感产品,结果表明,象山港海表pCO2整体上呈现从湾顶向湾口递减的趋势,均值为514.56 μatm,其中,湾内pCO2均值为551.94 μatm,湾外pCO2均值为477.19 μatm,整体呈现为大气CO2的源,且5年间秋季pCO2没有显著趋势性变化。结合多参数走航实测数据分析发现,2021年象山港秋季海表pCO2受物理混合作用及生物活动的共同调控。海表温度(sea surface temperature, SST)与pCO2具有良好的正相关关系,且SST对pCO2的影响主要体现在碳酸盐热力学平衡作用上。此外,平均温度归一化的pCO2(NpCO2)与海水盐度和溶解氧饱和度均具有良好的负相关关系,NpCO2与海水盐度的关系是潮汐作用下湾内和外海水体交换的结果;长时间序列的遥感数据分析也证实湾内、湾外pCO2与平均潮高具有较为一致的变化趋势,且这种趋势在湾外强于湾内。该研究构建了一套海湾小尺度pCO2遥感反演方法,为后续海-气CO2通量大范围、长时序遥感监测奠定了良好基础。
Coastal bays are greatly affected by human activities and natural changes, and the influence mechanism of variation in seawater carbon source and sink patterns is extremely complex. Due to the small spatial scale of the bay, it is necessary to use wide-bands high-spatial resolution satellite remote sensing for monitoring the air-sea CO2 flux. Compared with the traditional kilometer-level ocean color satellite data, the retrieval of the sea surface partial pressure of CO2 (pCO2), the key parameter to calculate air-sea CO2 flux, is extremely challenging in small-scale bays. Taking Xiangshan Bay in Zhejiang Province in autumn as an example, a satellite retrieval algorithm for sea surface pCO2 was proposed based on the in situ pCO2 data and Sentinel-2 satellite images in the past five years, using the machine learning method of support vector machine (SVM). The algorithm validation results showed a good performance with R2 of 0.92 and RMSE of 23.23 μatm, and the satellite-derived results were consistent with the in situ values. On this basis, the satellite products of pCO2 in Xiangshan Bay in autumn from 2017 to 2021 (September to November) were produced. The results revealed that the pCO2 of Xiangshan Bay showed a decreasing trend from the top of the bay to the mouth of the bay, with an average value of 514.56 μatm, of which the average pCO2 in the inner bay was 551.94 μatm and the average pCO2 in the outer bay was 477.19 μatm, which implied that Xiangshan Bay was a source of atmospheric CO2 as a whole. There was no significant trend change of pCO2 in autumn in the past five years. Combined with the analysis of in situ data of multiple parameters, it was found that the sea surface pCO2 of autumn in Xiangshan Bay in 2021 was jointly regulated by physical mixing and biological activities. Sea surface temperature (SST) had a good positive correlation with pCO2, which was mainly reflected by the thermodynamic equilibrium of carbonate system. In addition, the normalized pCO2(NpCO2) with average temperature had a good negative correlation with seawater salinity and dissolved oxygen saturation. The relationship between NpCO2 and salinity resulted from the exchange of sea water inside the bay and offshore coastal water under tidal effect. Long-time series satellite data analysis also confirmed that sea surface pCO2 had a relatively consistent trend with the average tide height inside and outside the bay, and this trend was stronger in the outside bay than that in the inner bay. In this study, a set of pCO2 remote sensing retrieval methods in the small-scale bay was constructed, which laid a good foundation for the subsequent long-time series satellite monitoring of sea-air CO2 fluxes.
近海海湾 / 海表pCO2 / 支持向量机 / 高空间分辨率卫星遥感 / 象山港
coastal bay / sea surface pCO2 / support vector machines / high spatial resolution satellite remote sensing / Xiangshan Bay
| [1] |
. Accurate assessment of anthropogenic carbon dioxide (CO2) emissions and their redistribution among the atmosphere, ocean, and terrestrial biosphere in a changing climate is critical to better understand the global carbon cycle, support the development of climate policies, and project future climate change. Here we describe and synthesize data sets and methodologies to quantify the five major components of the global carbon budget and their uncertainties. Fossil CO2 emissions (EFOS) are based on energy statistics and cement production data, while emissions from land-use change (ELUC), mainly deforestation, are based on land use and land-use change data and bookkeeping models. Atmospheric CO2 concentration is measured directly, and its growth rate (GATM) is computed from the annual changes in concentration. The ocean CO2 sink (SOCEAN) is estimated with global ocean biogeochemistry models and observation-based data products. The terrestrial CO2 sink (SLAND) is estimated with dynamic global vegetation models. The resulting carbon budget imbalance (BIM), the difference between the estimated total emissions and the estimated changes in the atmosphere, ocean, and terrestrial biosphere, is a measure of imperfect data and understanding of the contemporary carbon cycle. All uncertainties are reported as ±1σ. For the year 2021, EFOS increased by 5.1 % relative to 2020, with fossil emissions at 10.1 ± 0.5 GtC yr−1 (9.9 ± 0.5 GtC yr−1 when the cement carbonation sink is included), and ELUC was 1.1 ± 0.7 GtC yr−1, for a total anthropogenic CO2 emission (including the cement carbonation sink) of 10.9 ± 0.8 GtC yr−1 (40.0 ± 2.9 GtCO2). Also, for 2021, GATM was 5.2 ± 0.2 GtC yr−1 (2.5 ± 0.1 ppm yr−1), SOCEAN was 2.9 ± 0.4 GtC yr−1, and SLAND was 3.5 ± 0.9 GtC yr−1, with a BIM of −0.6 GtC yr−1 (i.e. the total estimated sources were too low or sinks were too high). The global atmospheric CO2 concentration averaged over 2021 reached 414.71 ± 0.1 ppm. Preliminary data for 2022 suggest an increase in EFOS relative to 2021 of +1.0 % (0.1 % to 1.9 %) globally and atmospheric CO2 concentration reaching 417.2 ppm, more than 50 % above pre-industrial levels (around 278 ppm). Overall, the mean and trend in the components of the global carbon budget are consistently estimated over the period 1959–2021, but discrepancies of up to 1 GtC yr−1 persist for the representation of annual to semi-decadal variability in CO2 fluxes. Comparison of estimates from multiple approaches and observations shows (1) a persistent large uncertainty in the estimate of land-use change emissions, (2) a low agreement between the different methods on the magnitude of the land CO2 flux in the northern extratropics, and (3) a discrepancy between the different methods on the strength of the ocean sink over the last decade. This living data update documents changes in the methods and data sets used in this new global carbon budget and the progress in understanding of the global carbon cycle compared with previous publications of this data set. The data presented in this work are available at https://doi.org/10.18160/GCP-2022 (Friedlingstein et al., 2022b).
|
| [2] |
|
| [3] |
|
| [4] |
|
| [5] |
|
| [6] |
|
| [7] |
|
| [8] |
. Here we present monthly, basin-wide maps of the partial pressure of carbon dioxide (pCO2) for the North Atlantic on a 1° latitude by 1° longitude grid for years 2004 through 2006 inclusive. The maps have been computed using a neural network technique which reconstructs the non-linear relationships between three biogeochemical parameters and marine pCO2. A self organizing map (SOM) neural network has been trained using 389 000 triplets of the SeaWiFS-MODIS chlorophyll-a concentration, the NCEP/NCAR reanalysis sea surface temperature, and the FOAM mixed layer depth. The trained SOM was labelled with 137 000 underway pCO2 measurements collected in situ during 2004, 2005 and 2006 in the North Atlantic, spanning the range of 208 to 437 μatm. The root mean square error (RMSE) of the neural network fit to the data is 11.6 μatm, which equals to just above 3 per cent of an average pCO2 value in the in situ dataset. The seasonal pCO2 cycle as well as estimates of the interannual variability in the major biogeochemical provinces are presented and discussed. High resolution combined with basin-wide coverage makes the maps a useful tool for several applications such as the monitoring of basin-wide air-sea CO2 fluxes or improvement of seasonal and interannual marine CO2 cycles in future model predictions. The method itself is a valuable alternative to traditional statistical modelling techniques used in geosciences.\n
|
| [9] |
|
| [10] |
吕航宇, 白雁, 李骞, 等. 夏季珠江口海域海水CO2分压的卫星遥感反演[J]. 海洋学研究, 2018, 36(2):1-11.
基于控制因子分析的方法,本研究建立了夏季珠江口海域海水CO<sub>2</sub>分压(pCO<sub>2</sub>)的遥感反演模型。基于珠江水与黑潮水的两端元混合,建立了水平混合和热力学作用的量化模型,并生成了查找表。同时,建立了基于黄色物质(含碎屑)吸收系数的盐度遥感算法,实现珠江口海域表层盐度的遥感反演。利用走航pCO<sub>2</sub>和匹配的遥感叶绿素质量浓度产品,建立了生物作用的量化模型。通过集成水平混合和生物作用,最终实现夏季珠江口海域pCO<sub>2</sub>的遥感反演。与走航pCO<sub>2</sub>比较表明,仅考虑水平混合和热力学作用的遥感结果会显著高估,考虑生物作用后,遥感结果无论在量值和空间变化趋势上均与实测结果相符。此外,遥感反演结果表明,夏季珠江口近岸水域为CO<sub>2</sub>的汇区,而离岸的陆架水域则为CO<sub>2</sub>的弱源。
|
| [11] |
|
| [12] |
|
| [13] |
|
| [14] |
吴凡杰, 吴钢锋, 董平, 等. 基于三维水动力水质数值模型的象山港排污策略研究[J]. 海洋环境科学, 2021, 40(1):24-33.
|
| [15] |
张健, 李佳芮, 刘书明, 等. 象山港生态系统脆弱性及其评价体系构建[J]. 海洋学研究, 2017, 35(2):74-81.
本文利用“暴露-敏感-适应”模型,建立了象山港生态系统脆弱性评价体系,经过分析与计算得出象山港生态系统脆弱性为中等脆弱,影响的主要胁迫因子为临港工业发展和海水养殖活动;系统敏感因子包括海洋生物多样性指数,鱼卵、仔稚鱼密度;最佳适应策略是完善环保法律法规和管理机制、加强入海污染物控制、提高污水处理率、提升公民环保意识和参与程度。
Using “exposure-sensitive-adaptation” model, the ecological vulnerability evaluation system of Xiangshangang Bay was established through the analysis and calculation. The vulnerability of Xiangshangang Bay ecosystem is moderate fragile, and the main stress factors are the development of harbor industry and aquaculture activities. The sensitive factors are marine biodiversity index, fish eggs and larvae density. The best adaptation strategies are to improve the environmental protection laws and regulations and management mechanism, strengthen the control of pollutants into the sea, improve the sewage treatment rate, improve citizen's environmental protection awareness and participation degree.
|
| [16] |
陈华伟, 吴卫飞. 象山港内新增网箱养殖污染物对海水水质的影响预测[J]. 渔业研究, 2021, 43(2):183-192.
象山港是宁波市最重要的网箱养殖港湾，网箱养殖带来丰厚经济利益的同时，也产生了大量的养殖污染物。文章以象山港内某新增大黄鱼网箱养殖基地为例，利用Deflt3D二维对流扩散模型预测养殖废水中的污染物对海水水质的影响。结果表明，新增大黄鱼网箱养殖产生的三种污染物浓度增量均是在网箱养殖点较高，在网箱养殖区外污染物随潮流运动迅速扩散，浓度增量降低，由于本项目规模小、时间短，其产生的养殖污染物对象山港水质的影响不大。但是，由于象山港内活性磷酸盐和无机氮已超标，对新增网箱养殖仍要严格评估其污染物排放浓度，以切实保护象山港海域环境，促进该海域养殖业的可持续发展，本文研究结果可为新增网箱养殖污染物预测提供参考。
|
| [17] |
钱茹茹. 2001—2020年宁波近岸海域赤潮灾害特征分析[J]. 江苏海洋大学学报:自然科学版, 2022, 31(3):11-17.
|
| [18] |
|
| [19] |
|
| [20] |
|
| [21] |
国家海洋局. 海洋调查规范第4部分:海水化学要素调查: GB/T 12763.4—2007[S]. 2007.
State Oceanic Administration. Specifications for oceanographic survey—Part 4: Survey of chemical parameters in sea water: GB/T 12763.4—2007[S]. 2007.
|
| [22] |
陈楚群, 潘志林, 施平. 海水光谱模拟及其在黄色物质遥感反演中的应用[J]. 热带海洋学报, 2003, 22(5):33-39.
|
| [23] |
Based on characteristics analysis about remote sensing reflectance, the Secchi Disk Depth (SDD) in the Qiandao Lake was predicted from the Landsat8/OLI data, and its changing rates on a pixel-by-pixel scale were obtained from satellite remote sensing for the first time. Using 114 matchups data pairs during 2013–2019, the SDD satellite algorithms suitable for the Qiandao Lake were obtained through both the linear regression and machine learning (Support Vector Machine) methods, with remote sensing reflectance (Rrs) at different OLI bands and the ratio of Rrs (Band3) to Rrs (Band2) as model input parameters. Compared with field observations, the mean absolute relative difference and root mean squared error of satellite-derived SDD were within 20% and 1.3 m, respectively. Satellite-derived results revealed that SDD in the Qiandao Lake was high in boreal spring and winter, and reached the lowest in boreal summer, with the annual mean value of about 5 m. Spatially, high SDD was mainly concentrated in the southeast lake area (up to 13 m) close to the dam. The edge and runoff area of the lake were less transparent, with an SDD of less than 4 m. In the past decade (2013–2020), 5.32% of Qiandao Lake witnessed significant (p < 0.05) transparency change: 4.42% raised with a rate of about 0.11 m/year and 0.9% varied with a rate of about −0.09 m/year. Besides, the findings presented here suggested that heavy rainfall would have a continuous impact on the Qiandao Lake SDD. Our research could promote the applications of land observation satellites (such as the Landsat series) in water environment monitoring in inland reservoirs.
|
| [24] |
|
| [25] |
|
| [26] |
|
| [27] |
|
| [28] |
黄菊, 高郭平, 程天宇, 等. 2015年冬季南黄海水文特征及海-气CO2通量分析[J]. 上海海洋大学学报, 2017, 26(5):757-765.
|
| [29] |
|
| [30] |
张龙军. 东海海-气界面CO2通量研究[D]. 青岛: 中国海洋大学, 2003.
|
| [31] |
石晓勇, 陆茸, 张传松, 等. 长江口邻近海域溶解氧分布特征及主要影响因素[J]. 中国海洋大学学报:自然科学版, 2006, 36(2):287-290,294.
|
/
| 〈 |
|
〉 |
