基于GOCI影像的长江口及邻近海域有色溶解有机物（CDOM）遥感反演及其逐时变化分析

发布时间：2019-06-28 09:39

【摘要】：有色溶解有机物(CDOM)是海洋水色组分之一,认识河口近岸海域CDOM的分布、迁移和转化不仅对于海洋水色遥感具有重要意义,而且有着显著的生物学意义和光谱学意义,其与生物地球化学循环有重要联系。本文以CDOM吸收系数为CDOM浓度指标,对CDOM的遥感反演应用进行研究,并在此基础上探究长江口和邻近海域的CDOM分布特征及其日变化特征。对于长江口高浊度水体,已有的反演模型对于CDOM的变化非常不敏感,反演精度很低,因此本文选择BP神经网络算法进行研究,其优势是得到的模型结果中失效的数据比较少,而其他模型可能存在大面积的算法失效。从精度结果看,BP神经网络算法的精度高于GOCI标准软件GDPS提供的Moon算法与YOC算法,但同时BP神经网络算法也存在需要人为参与学习判断的缺点。本文以长江口及其邻近海域野外实测数据为基础,在QAA算法及QAA-E算法基础上,建立了基于BP神经网络算法反演bbp(555)与ap(443)的关系,适用于GOCI卫星数据的反演模型。利用2012年4月26日星地同步数据对模型进行验证,验证结果表明该模型可以应用于GOCI卫星数据的CDOM吸收系数反演。在此基础上分析了长江口及其邻近海域CDOM吸收系数分布及CDOM日变化情况,得到以下结论。(1)基于QAA算法的BP神经元网络法对CDOM吸收系数的反演效果较好,适用于长江口及其邻近海域CDOM反演。但总体而言,高浊度水域CDOM反演精度仍有待提高,原因是由于长江口及其邻近海域沿岸水体中含较高浓度的悬浮物,而悬浮物对后向散射光谱的影响占主导作用,对叶绿素和CDOM光谱的影响较大,从而减弱了 CDOM与后向散射光谱之间的相关性,最终导致算法在复杂水体的反演精度降低。(2)利用2014年3月15日的GOCI影像反演长江口及其邻近海域CDOM吸收系数,并对其日内变化时空特征进行了分析。结果表明,GOCI数据能够清晰展现CDOM吸收系数的空间分布,体现出水体受潮汐等外界因素影响而导致的CDOM吸收系数的变化。从日内变化来看,在涨潮期间,CDOM吸收系数的空间分布为长江口内北支高于南支浓度,北支浓度与口外接近;而在退潮期间,北支CDOM吸收系数明显下降,且低于口外CDOM吸收系数,南港,北港的CDOM吸收系数也出现逐渐下降现象。而海水与淡水的混合稀释,使CDOM的浓度梯度从长江口往外海区呈现沿西北-东南方向降低的趋势。(3)利用GOCI数据高时间分辨率的特点,可以捕捉一天内CDOM的变化特征,这有利于对CDOM循环过程进行实时监测,为进一步研究长江口及其邻近海域CDOM日循环变化特性及其驱动机制及河口演化规律提供了重要的观测数据。(4)定量遥感中分类、校正、反演模型的误差需要进行统计评估,从而确定其性能与效果,而这些评估通常基于一些常见的统计参数。本文通过计算机模拟了误差的各种近似分布,研究了样本量n和统计指标RMSE、MAE与UA之间的关系。结果表明RMSE、MAE与UA随着样本量n的变化呈现不同的趋势:在n小于40左右时,RMSE和MAE往往随样本量增大呈上升趋势,随后趋于平缓;而UA随样本量n的增加总是平滑下降。由此可以看出,在样本量少的情况下,UA比RMSE、MAE更适合评价遥感模型的不确定性(可信赖度),因为基于统计学常识,样本量越大,建立的模型就越可靠(不确定度越小)。
[Abstract]:The color-dissolved organic matter (CDOM) is one of the marine water-color components. It is recognized that the distribution, migration and transformation of the CDOM in the nearshore area of the estuary are not only important for the remote sensing of marine water color, but also have significant biological significance and spectroscopy significance. It is closely related to the biogeochemical cycle. In this paper, the CDOM absorption coefficient is used as the index of the concentration of the CDOM, and the application of the remote sensing inversion of the CDOM is studied, and the distribution characteristics and the diurnal variation of the CDOM in the Changjiang estuary and the adjacent sea area are explored. For the high-turbidity water body of the Yangtze River estuary, the existing inversion model is very insensitive to the change of the CDOM, and the inversion precision is very low, so the BP neural network algorithm is selected for research, and the advantage is that the data of the failure in the obtained model result is less, While other models may have a large area of algorithm failure. The accuracy of the BP neural network algorithm is higher than that of the Moon algorithm and the YOC algorithm provided by the GCI standard software GDPS from the accuracy results, but the BP neural network algorithm also has the disadvantage that the artificial participation of the learning judgment is required. The relationship between bp (555) and ap (443) on the basis of the QAA algorithm and the QAA-E algorithm is established based on the field data of the Changjiang River estuary and its adjacent sea area, and the relationship between bp (555) and ap (443) is established based on the BP neural network algorithm, which is suitable for the inversion model of GOCI satellite data. The results show that the model can be applied to the inversion of the CDOM absorption coefficient of GOCI satellite data. On this basis, the distribution of the absorption coefficient of the CDOM and the diurnal variation of the CDOM in the Changjiang Estuary and its adjacent sea area are analyzed, and the following conclusions are obtained. (1) The BP neural network method based on the QAA algorithm has better inversion effect on the absorption coefficient of the CDOM, and is suitable for the inversion of the CDOM in the Changjiang estuary and its adjacent sea area. However, in general, that accuracy of the inversion of the CDOM in the high-turbidity water area is still to be improved due to the high concentration of suspended matter in the water body along the Yangtze estuary and its adjacent sea area, and the effect of the suspended matter on the back-scattering spectrum of the suspended matter is dominant, and the effect on the spectrum of the chlorophyll and the CDOM is large. So that the correlation between the CDOM and the backward scattering spectrum is reduced, and finally, the inversion accuracy of the algorithm in the complex water body is reduced. (2) Using the GCI image on March 15,2014, the absorption coefficient of the CDOM in the Changjiang Estuary and its adjacent sea area was inverted, and the spatial and temporal characteristics of the diurnal variation were analyzed. The results show that the spatial distribution of the absorption coefficient of the CDOM can be clearly displayed by the GOCI data, and the change of the absorption coefficient of the CDOM due to the influence of the external factors such as the deliquescence of the water is reflected. In the diurnal variation, the spatial distribution of the absorption coefficient of the CDOM is higher than that of the south branch in the Yangtze River estuary during the flood tide, and the concentration of the north branch is close to the outside of the mouth; and during the ebb tide, the absorption coefficient of the CDOM of the north branch is obviously lower, and the absorption coefficient of the CDOM is lower than the external CDOM absorption coefficient, and the south port, The absorption coefficient of CDOM in the northern port also appears to be decreasing. The mixture of seawater and fresh water is diluted, so that the concentration gradient of the CDOM is decreased in the north-west-southeast direction from the outside of the Yangtze River estuary. and (3) utilizing the characteristics of high time resolution of the GOCI data, can capture the change characteristics of the CDOM in a day, which is favorable for real-time monitoring of the CDOM circulation process, In order to further study the characteristics of the cycle change of the CDOM in the Yangtze River estuary and its adjacent sea area and its driving mechanism and the estuary evolution law, it provides important observation data. (4) The error of classification, correction and inversion model in quantitative remote sensing requires statistical evaluation to determine its performance and effect, which is usually based on some common statistical parameters. The relationship between the sample size n and the statistical index RMSE, MAE and UA is studied by computer simulation of various approximate distributions of the error. The results show that RMSE, MAE and UA show a different trend with the change of sample size n: when n is less than 40, the RMSE and MAE tend to increase with the sample size, and then tend to be gentle; and UA increases with the sample size n. As a result, it can be seen that, in the case of a small sample size, the UA is more suitable for evaluating the uncertainty (reliability) of the remote sensing model than the RMSE and MAE, as the larger the sample size is based on the statistical common sense, the more reliable the model is established (the less the degree of certainty).
【学位授予单位】：浙江大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：P734

【参考文献】