• <tr id="yyy80"></tr>
  • <sup id="yyy80"></sup>
  • <tfoot id="yyy80"><noscript id="yyy80"></noscript></tfoot>
  • 99热精品在线国产_美女午夜性视频免费_国产精品国产高清国产av_av欧美777_自拍偷自拍亚洲精品老妇_亚洲熟女精品中文字幕_www日本黄色视频网_国产精品野战在线观看 ?

    Inversion Study on Pollutant Discharges in the Bohai Sea withthe Adjoint Method

    2015-04-01 01:57:08SHENYouliWANGChunhuiWANGYonggangandLVXianqing
    Journal of Ocean University of China 2015年6期

    SHEN Youli, WANG Chunhui, WANG Yonggang, and LV Xianqing), *

    ?

    Inversion Study on Pollutant Discharges in the Bohai Sea withthe Adjoint Method

    SHEN Youli1), 2), WANG Chunhui3), WANG Yonggang4), and LV Xianqing1), *

    1),,266100,2),536000,3),,266033,4),,266061,

    The adjoint methodis presented which could be used to estimate the time-varying contamination concentration (CC) from pollution source (PS). Then the pollutant flux is calculated.In order to guarantee the continuity of pollutant distribution and make the calculated results more accurate,the independent point (IP)scheme is proposed. The contamination concentrations (CCs)atsome time steps are selected as the independent points (IPs), and only CCs at these IPs are optimized while CCsat other points are calculated through linear interpolation of the independent CCs.In twin numerical experiments, all the given distributions are successfully inverted with the adjoint method. The cost functions and the mean absolute errors (MAEs) in concentrations and pollutant fluxes decrease greatly after assimilation,and the cost functions are reduced by about 5 orders of magnitude compared with their initial values. The results indicate that the adjoint method is computationally efficient to recover CCs from PS. It is easier to invert the given distribution which is less complex.The inversion efficiency with IP scheme is raised compared to that without this scheme. The IP scheme is significant for the inversion result, in which appropriate IP number couldyieldbetter inversion results. More work will be done to apply this method to real experiment.

    adjoint method; time-varying variable; contamination concentration; pollutant flux; independent point scheme

    1 Introduction

    The Bohai Sea is China’s only inland sea, which receives large volumes of domestic and industrial discharges and even accidental or intentional spills from moving ships. According to statistics, there are more than 80 rivers flowing into the Bohai Sea, including 40 rivers which have important effects on the Bohai Sea. Due to limited flushing capacity, the multiple pollutant discharges in the Bohai Sea can deteriorate the water quality significantly (Zheng, 2011). Internationally there is a clear consensus that source pollution, especially land source pollution, is having a profound effect on coastal and marine ecosystems(Waterhouse, 2012). Preventing point source pollution is among the most cost-effective means of marine management.

    Obtaining precise information on pollutant discharge from point sources such as wastewater effluent and runoff from oil fields and unsewered industrial sites, is important forpollution prediction and effective control. However, the information is hard to obtainbecause of the complex variation of sources, especially for large-scale and natural sources (Zhu and Wang, 2006). One way to solve this problem is to study variety of sources and to construct pollutant dischargeor emission model. Gupta(2004) carried out water quality simulation of Thane creek using a two-dimensional advection-diffusion model to determine the wastewater assimilative capacity of the creek considering 11 point-source discharges. Shulka(2002) considered advection and reaction when studying the steady nonconservative pollutant with time-dependent periodic waste discharge concentration. Milnes and Perrochet (2007) predicted the concentration at the unknown PS when a reversed flow field transport simulation was performed. However, with limited observations, this approachusually causes the uncertainty and even deviation from the reality.

    Another approach is to invert the pollutant discharges or emissions using the observation and the pollutant discharge or emission model.This approachmainly includes two categories: gradient based methods and gradient free methods. In the first category, there is variational method based on an adjoint, reduced adjoint, simple finite difference method and stochastic estimation of gradient. For thesecond category, there is ensemble Kalman filter (EnKF), Bayesian methodsand sequential Monte-Carlo methods like particle filtering. Gilliland(2001) used the discrete Kalman filter (DKF) in an adaptive-iterative mode to deduce time-varying air quality emission.Their result shows that if the initial condition assumption is violated, the adaptive-iterative DKF can produce biased emissions to compensatefor the initial modeled and observed concentration difference. Heeminkand Segers (2002) used the Kalman filter to estimate and predict ozone concentrations in a part of North West Europe. Their work focused onthe development of fast filter algorithms to make Kalman filtering feasible for high dimensional space-time models. Zhu and Wang (2006) presented some ensemble based statistical estimation methods for inverting modeling of pollution emissions. The results show that EnKF can estimate the time-variant emission effectively at every time step when the observations are not available. Bousquet(1999) presented a 3-D Bayesian inverse methodology to infer annual net carbon dioxide (CO2) sources and sinks using atmospheric measurements of CO2. Elbern(2000) studied emissions using a regulatory air quality model with four dimension variational (4DVAR) data assimilation. It is shown that the space-time variational method is able to analyze emission rates of Nitric oxide (NO) directly, as for volatile organic compounds (VOC), regularization techniques must be introduced. 4DVAR has the advantage of directly assimilating variousobservations distributed in time and space into numerical models while maintaining dynamicaland physical consistency with the model. In this paperthe adjoint method is used to invert CC from point source and the pollutant flux is also computed. The CC in this paper is taken as a time-varying variable, rather than a constant in traditional studies.

    The paper is organized as follows. In the next section the model is presented, leading to the relationship between the target concentrations and the flow field. In Section 3 the method we used in this paper is illustrated. The selectionand the optimization of the independent CCs areshown in Section 4. In Section 5 numericalexperiments arecarried out and the results are analyzed. Finally conclusions are presented in Section 6.

    2 Model Introduction

    2.1 The Background Flow Field

    The background flow field is provided by the 3-D Regional Ocean Model System (ROMS). ROMS is a free-surface, hydrostatic, and primitive equation ocean model that uses stretched, terrain-following coordinates in the vertical and orthogonal curvilinear coordinates in the hori- zontal. It has been expanded to include a variety of new features including high-order advection schemes, accurate pressure gradient algorithms, several subgrid-scale param- eterizations, atmospheric, oceanic, benthic boundary layers, biological modules, and radiation boundary conditions (Shchepetkin and McWilliams, 2005).

    In the present study the horizon resolution, the vertical resolutionand the integral time step is similar to Wang(2013). The open boundary is installed on 122.5?E. The open boundary condition is from the measured data.Surface forcing functions are drawn from the COADS climatology of Da Silva. (1994). These forcing fields include surface wind stresses, heat and freshwater fluxes, and heat flux sensitivity to sea surface temperature. The model is run for 10 years and result of the last year serves as the hydrodynamic background field for ecological model. The simulated circulation is Eulerian residual current. Mean surface and bottomcirculation of the Bohai Sea in July are shown in Fig.1. As for the mean surface circulation, the maximum current is about8.0cms?1, while the minimum current is less than 2.0cms?1.A counterclockwise gyre is in the northeast corner of the Liao- dong Bay and the flow direction is from south to north in the Bohai Bay. In the Laizhou Bay the flow becomes weaker. Meanwhile the circulation near the Bohai Strait is inflow in the north and outflow in the southern part of the Bohai Strait, which is similar to Wei(2001) and Wei(2003). While the bottom circulation is much weaker than that of surface except in the Bohai Strait, and the maximum current is less than 1.0cms?1.

    Fig.1 Mean surface and bottom circulation of the Bohai Sea in July. (a) is the mean surface circulation, and (b) is the mean bottom (the third layer) circulation.

    2.2 Governing Equation and Its Difference Scheme

    Based on the hydrodynamic background field, an advection-diffusion model is constructed, and the basic equation system is written in the following form:

    , (1)

    whereis the concentration of contamination,,,are components of the Cartesian coordinate system,,,are velocities in the Cartesian directions,is time,AandKare the horizontal and vertical turbulent diffusion coefficients, respectively.

    The petroleum pollutants are considered as pollutants for example, and the initial concentration is set to be 25μgL?1. Radiation boundary condition is given on the open boundary, which is described as:

    wheredenotes the normal direction of the open boundary. While the closed wall boundary condition is in the following form:

    . (3)

    The difference equation of governing Eq. (1) is written as follows:

    , (4)

    where,

    , (6)

    , (7)

    The stabilization condition is:

    where,,are components of the Cartesian coordinate system, ?x, ?(where ?is constant),?zare grid spacings in the Cartesian directions, respectively.

    2.3 The Adjoint Equation and Its Difference Scheme

    The cost function which is usually used to measure the misfit between the observations and the simulation results is defined as follows:

    where Tdenotes transposition andthe observations.Kis the weight of the observation, which is defined as:

    . (11)

    By introducing Lagrangianfunction, the adjoint model, which provides a method of calculating the gradient of cost function with respect to CC from PS, can be constructed.is defined as:

    , (12)

    where*is called adjoint variable of. In order to minimize, let ?/?=0, then the adjoint equation can be obtained:

    . (13)

    The corresponding open boundary condition is defined as:

    While the closed wall boundary condition is in the fol- lowing form:

    . (15)

    The difference equation of adjoint Eq. (13) is asfollows:

    , (16)

    where

    , (18)

    , (19)

    2.4 Test of the Adjoint Model

    It is essential to verify the adjoint model before proceeding with the minimization runs. As the work of Zhang and Lv (2008) and Chen(2013), let

    be a Taylor expansion of the cost function, which is defined in Section 2.3, whereis a general value of CC,is a small scalar andis an arbitrary vector of unit length. Rewriting (21), a function ofcan be defined as

    . (22)

    As mentioned in the work of Smedstad and O’Brien (1991), Das and Lardner (1991), Lardner and Song (1995) and Navon(1991), ifis chosen close to machine zero, one cannot expect to be able to verify that the correct gradient has been found. For values ofwhich are not too close to the machine zero, one should expect to obtain a value ofΦ() which is close to 1. In Fig.2, the values Φ() are plotted when the CCs are taken as an example. In this test, the basic pointis set to be 25μgL?1, which is equal to the initial values of CCs in the following numerical experiments. It shows that forbetween 10?1and 10?8, Φ() is close to 1 (the dashed line), which means the adjoint model is verified.

    Fig.2 Values of the function Φ(α)versus α.

    3 The Adjoint Method

    The variational method with adjoint equations was first demonstrated by Penenko and Obraztsov (1976) and has been used in manyprevious studies(Lv and Zhang, 2006; Seiler, 1993; Fan and Lv, 2009; Wang, 2013; Cao., 2013; Chen., 2013). It is an advanced data assimilationmethod that involves the adjoint method and has the advantage of directly assimilating variousobser- vations distributed in time and space into numerical mod- els while maintaining dynamicaland physical consistency with the model (Zhang and Lv, 2008). The basic idea of variational adjoint method is to define a cost function in a desirable way that measures the misfit of the model con- sidered (Robertson and Langner, 1998).

    In this study, the calculation process of the adjoint method (Fig.3) is as follows:

    1) An initial value of the CCs from PS is given empiri- cally;

    2) Perform the simulation by running the forward model, and the simulation results are obtained;

    3) The adjoint model is driven backward in time with the misfit between the model result and the observation;

    4) Calculate the gradient of cost function with respect to the CCs from PS;

    5) Adjust the CCs from PS on the basis of the gradient above;

    6) Return to step (1) while the initial values are updated with the aboveconcentrations, and repeat the iteration procedure;

    7) The process is stopped when the number of iterations is 200 (We tried different numbers of iterationsand found that 200 is a good choice mainly based on the following considerations: (1) The cost functions and the MAEs are almost no longer falling after 200 numbers of iterations; (2) The calculation required is acceptable for the 200 numbers of iterations.) or the target<(whereis a small number, such as 0.0001) is reached.

    Fig.3 The calculation process of the adjoint method.

    4 Twin Experiment (TE)

    4.1 Experimental Design

    The calculated regionis the Bohai Sea (37?N–41?N, 117.5?E–122.5?E). The horizontal resolution of the model is 4?×4?, and the open boundary is installed on 122.5?E. In the vertical direction the water is divided into 6 layers, and the thicknesses of each from top to bottom are 10m, 10m, 10m, 20m, 25m and 25m, respectively. The integral time step is 1 hour and the integral period is 7 days.

    A bathymetry map of the Bohai Sea is shown in Fig.4, in which the PS and the observations are also located. Observations are marked from 1 to 20, Nos. 1–4 are observed once per day during 8:00–18:00from the second to the fourth day, Nos. 5–12 are observed once per day during 8:00–18:00 on the fourth, the fifth and the seventh day, and Nos. 13–20 are observed once per day during 8:00–18:00on the sixth and seventh day.

    Fig.4 The topography of the Bohai Sea and the locations of the observation sites. ‘+’is the location of the PS, and the dots are the observations.

    4.2 Correction of CCs at IPs

    In order to guarantee the continuity of pollutant distribution and make the calculated results more accurate,the IP scheme is proposed.is the value of CC at the IPs,is the CC from the PS after linear interpola-tion (denotesthe calculated point,denotesthe IP), and the relationship between them is as follows:

    wherewis the weight coefficient in the Cressman form (Cressman, 1959) and the specific form is as follows:

    , where, (24)

    whereis the influence radius,ris the distance from the IP to thecalculated point.

    After derivation, the gradient of cost functionwithrespecttois

    , (26)

    , (28)

    , (29)

    Since the variable is adjusted in the inverse direction of the gradient above, thecalibration relationship of independent CC from PS is as follows:

    , (31)

    , (32)

    4.3 The Given Variation of CC from PS

    Considering the real variation of CC from PS, three types of time-varying CCs are prescribed as follows:

    Type 1:; (34)

    Type 2:; (35)

    Type 3:. (36)

    4.4 Design Strategies of IP

    Six IP strategies are designed to invert the three types of prescribed CCs (Fig.5). The strategies are presented as follows:

    Strategy (A): Three evenly distributed grid points are taken as IPs, and the influence radius is the length of 72 grids.

    Strategy (B) to Strategy (F): The numbers of IPs are 5, 7, 9, 13 and 19, respectively, which are distributed evenly. Radiuses are the lengths of 38, 24, 18, 12 and 8 grids corresponding to 5 strategies, respectively.

    All strategies are used to invert CCs with the distributions of Type 1 and 2, and Strategies (B)–(F) are used to invert CCs with the distributions of Type 3.

    Fig.5 Six IP strategies. Larger points and smaller ones represent IPs and other grid points, respectively.

    5 Experimental Results and Discussion

    5.1 The Inversion Result of Type 1

    The cost functions and MAEsin CCs and pollutant fluxes are two important convergence criteria for data assimilation in this model. The errors after assimilation for Type 1 are shown in Table 1, in which ‘None’ denotes the result without IP scheme. The results show that the cost functions, and the MAEs inCCs and pollutant fluxeswith strategies (A)–(F) decrease significantly, and the cost functions are reduced by about 5 orders of magnitude compared with their initial values, indicating that the given time-varying distribution is successfully inverted with these strategies. The result with Strategy (A) is the most satisfactory.The MAE in CC declines by 97.7% from 14.98μgL?1to 0.41μgL?1. Meanwhile the MAE in pollutant flux decreases by 98.9%from 22.74t to 0.25t. The MAEs with ‘None’ strategy are generally bigger than those with IP strategies except strategy (C), which shows that IP scheme is a more effective technique to invert the prescribed CCs.

    Table 1 Errors after assimilation

    The inverted distributions of CCs and pollutant fluxes with using Strategy (A) are shown in Fig.6. In this figure, the inverted distribution of CCs shows a good agreement with the prescribed one, and the maximum deviation is 1.54μgL?1. The pollutant flux is also well inverted, and the error at any moment is within 1.73t. It is concluded that the given time-varying distribution is successfully inverted using Strategy (A).

    Fig.6 The given distribution (solid lines) and the corresponding inversion results (dotted lines). (a) is the given time-varying concentration and the inversion result, (b) is the given time-varying flux and the inversion result.

    Then observations are increased, and Nos. 1–4 are observed from the second to the seventh day, while other observations are the same as the former. The Strategy (A) is taken as an example, and the inversion result is given in Fig.7. The original observations are marked observed (1), and latter ones are marked observed (2). It is clear that the concentration after assimilation using observed (2) in the last few hours is better than that using observed (1), and the MAE inCC is reduced to 0.32μgL?1. Apparently, the inversion efficiency is raised by increasing observations, which is in accordance with the reality.

    5.2 The Inversion Result of Type2

    The errors after assimilation in Type 2 are shown in Table 2. The MAEsin CCs with strategies (A)–(F) decline from 21.62μgL?1to 1.45μgL?1or smaller, while the MAEsin pollutant fluxes with these strategies decline from 37.63t to 0.81t or smaller, indicating that the given distribution is successfully inverted. The result with strategy(C) is the best. The errors without IP scheme are a little bigger than those with this strategy, which also shows the inversion result can be improved with IP scheme.

    Fig.7 The given distribution and the inversion results.

    Table 2 Errors after assimilation

    The given distribution and the inversion result with strategy (B) are shown in Fig.8, which shows the given concentration is well inverted with deviation remaining within 2% except those in the last few hours. From Fig.8 it is clear that the adjoint method also makes it possible to invert the given pollutants flux with errors less than 4t.

    Fig.8 The given distributions (solid lines) and the corresponding inversion results (dotted lines). (a) is the Given time-varying concentration and the inversion result, (b) is the given time-varying flux and the inversion result.

    5.3 The Inversion Result of Type 3

    The errors after assimilation in Type 3 are shown in Table 3 from which it is seen that the MAEsinCCs and pollutant fluxes using Strategies (B)–(F) can reach 2.52μgL?1and 1.90tor smaller, respectively.The cost function can even reach the magnitude of 10?5of its initial value, indicating that the given distribution is successfully inverted with all these strategies. However, the MAE in CC and pollutant flux without IP scheme are 3.32μgL?1and 1.79t, respectively.

    The inversion results of the given time-varying con-centration and errors between the given time-varying concentration and the inversion results using Strategies (B)–(F) are shown in Fig.9, in which deviation is smaller than 5% except in the last few hours, indicating that the given time-varying concentration is successfully inverted with all these strategies. But the result is not as good as the above inversion result, because the variation is more complex and it is hard to invert. From Fig.9 it also can be seen that the IP strategy has significant influence on inversion results, for inappropriate IP number can cause the result to deviate from the given distribution greatly.

    Table 3 Errors after assimilation

    Fig.9 The inversion results of the given time-varying concentration (a) and the errors between the given time-varying concentration and the inversion results (b).

    The inversion results of the given time-varying pollutant flux and errors between the given time-varying pollutant flux and the inversion results using Strategies (B)–(F) are shown in Fig.10, where deviation is smaller than 3% except in the last few hours, indicating that the given time-varying pollutant flux is also successfully inverted. Asfor the time-varying CC, the inversion result is also a little worse than the above two results.

    The MAEsinCCsand pollutant fluxes versus assimilation step using Strategies (B)-(F) are shown in Fig.11. From this figure, we can find the MAEs in CCs decrease rapidly in the beginning, and can decrease to 2.52or smaller after 200 interaction steps. The MAEs in pollutant fluxes are similar to those in CCs. It is evident that the given distributions are successfully inverted using all these strategies.

    Fig.10 The inversion results of the given time-varying pollutant flux (a) and the errors between the given time-varying pollutant flux and the inversion results at any moment (b).

    Fig.11 TheMAEs in CCs (a) and pollutant fluxes (b) versus assimilation step.

    6 Conclusions

    The PS information is hard to get due to various reasons, so one of the objectives of this study is to invert the time-varying CCs and then calculate the pollutant fluxes from PS by using the adjoint method. Based on the hydrodynamic background field, anadjoint assimilation modelof pollutantin the Bohai Sea is presented. In order to guarantee the continuity of pollutant distribution over time and make the calculated results more reasonable,theIP scheme is proposed.The CCs atsome time steps are selected as the IPs, and only CCs at these IPs are optimized while thoseat other points are calculated through linear interpolation of the independent CCs. In this way, the number of control variables is reduced.

    It is shown that all the given distributions could be estimated by the adjoint method. The cost functions and the MAEs decrease greatly after assimilation, and the cost functions are reduced by about 5 orders of magnitude compared with their initial values. The results indicate that the adjoint method is an effective tool to recover CCs from PS. It is also shown that the IP scheme is significant for the inversionresult, in which appropriate IP number couldyieldbetter inversion results.

    The adjoint method can invert the time-varying point source information, which is the first application in oceanography. In the present study only ideal experiments are carried out in which observations are obtained by the model itself, and data in our hand are available to support this study. In the further study we would estimate the source information by assimilating the routine monitoring data in the Bohai Sea, which gives a reference to the scientific countermeasures to control pollution discharge.

    Acknowledgements

    Partial support for this research was provided by the National Natural Science Foundation of China (Grant Nos. 41072176 and 41371496), the State Ministry of Science and Technology of China (Grant Nos. 2013AA121203 and 2013BAK05B04), and the Fundamental Research Funds for the Central Universities (201262007).

    Bousquet, P., Ciais, P., Peylin, P., Ramonet, M., and Monfray, P., 1999. Inverse modeling of annual atmospheric CO2sources and sinks. 1. Method and control inversion., 104 (21): 26161-26178.

    Cao, A. Z., Chen, H. B., Zhang, J. C., and Lv, X. Q., 2013. Optimization of open boundary conditions in a 3D internal tidal model with the adjoint method around Hawaii., DOI: 10.1155/2013/950926.

    Chen, H. B., Cao, A. Z., Zhang, J. C., Miao, C. B., and Lv, X. Q., 2013. Estimation of spatially varying open boundary conditions for a numerical internal tidal model with adjoint method., DOI: 10.1016/j.matcom.2013.08.005.

    Cressman, G. P., 1959. An operational objective analysis system., 87: 367-374.

    Da Silva, A. M., Young, C. C., and Levitus, S., 1994.. NOAA Atlas NESDIS 6, US Department of Commerce, NOAA, NESDIS, USA.

    Das, S. K., and Lardner, R. W., 1991. On the estimation of parameters of hydraulic models by assimilation of periodic tidal data., 96: 15187-15196.

    Elbern, H., Schmidt, H., Talagrand, O., and Ebel, A.,2000. 4D-variational data assimilation with an adjoint air quality model for emission analysis.,15: 539-548.

    Fan, W., and Lv, X. Q., 2009.Data assimilation in a simple marine ecosystem model based on spatial biological parameterizations.g, 220(17): 1997-2008.

    Gilliland, A., 2001. A sensitivity study of the discrete Kalman filter (DKF) to initial condition discrepancies., 106 (16): 17939-17952.

    Gupta, I., Dhage, S., Chandorkar, A.A., and Srivastav, A., 2004. Numerical modeling for Thane creek.,19: 571-579.

    Heemink, A. W., and Segers, A. J., 2002. Modeling and prediction of environmental data in space and time using Kalman filtering., 16: 225-240.

    Lardner, R. W., and Song, Y., 1995. Optimal estimation of eddy viscosity and friction coefficients for a quasi-three-dimensional numerical tidal model.,33 (3): 581-611.

    Lv, X. Q., and Zhang, J. C., 2006. Numerical study on spatially varying bottom friction coefficient of a 2D tidal model with adjoint method., 26: 1905-1923.

    Milnes, E.,and Perrochet, P., 2007. Simultaneous identification of a single pollution point-source location and contamination time under known flow field conditions., 30: 2439-2446.

    Navon, I. M., Zou, X., Derber, J., and Sela, J., 1991. Variational data assimilation with an adiabatic version of the NMC spectral model., 120: 1433-1446.

    Penenko, V. V., and Obraztsov, N. N., 1976. A variational initialization method for the fields of meteorological elements (English translation)., 1: 1-11.

    Robertson, L., and Langner, J., 1998. Source function estimate by means of variational data assimilation applied to the ETEX-I tracer experiment., 32 (24): 4219-4225.

    Seiler, U., 1993. Estimation of open boundary conditions with the adjoint method., 98: 22855-22870.

    Shchepetkin, A. F., and McWilliams, J. C., 2005. The regional oceanic modeling system (ROMS): A split-explicit, free-surface, topography-following-coordinate oceanic model.,9: 347-404.

    Shulka, P., 2002. Analytical solutions for steady transport dispersion of nonconservative pollutant with time-dependent periodic waste discharge concentration.,129 (9): 866-869.

    Smedstad, O. M., and O’Brien, J. J., 1991. Variational data assimilation and parameter estimation in an equatorial Pacific Ocean model., 26: 179-241.

    Wang, C. H., Li, X. Y., and Lv, X. Q., 2013. Numerical study on initial field of pollution in the Bohai Sea with an adjoint method., DOI: 10.1155/2013/104591.

    Waterhouse, J., Brodie, J., Lewis, S., and Mitchell, A., 2012. Quantifying the sources of pollutants in the Great Barrier Reef catchmentsand the relative risk to reef ecosystems., 65: 394-406.

    Wei, H., Wu, J. P., and Pohlmann, T., 2001. A simulation on the seasonal variation of the circulation and transport in the Bohai Sea., 19 (2): 1-9.

    Wei, Z. X., Li, C.Y., Fang, G. H., and Wang, X. Y., 2003. Numerical diagnosticstudyof thesummertimecirculationinthe Bohai Seaandthe water transport inthe Bohai Strait., 21 (4): 454-464 (in Chinese with English abstract).

    Wolfe, P., 1969. Convergence conditions for ascent methods., 11: 226-235.

    Wolfe, P., 1971. Convergence conditions for ascent methods. II. Some corrections., 13: 185-188.

    Zhang, J. C., and Lv, X. Q., 2008. Parameter estimation for a three-dimensional numerical barotropic tidal model with adjoint method., 57 (1): 47-92.

    Zheng, B. H., Zhao, X. R., Liu, L. S., Li, Z. C., Lei, K., Zhang, L., Qin, Y. W., Gan, Z., Gao, S. Z., and Jiao, L. X.,2011. Effects of hydrodynamics on the distribution of trace persistent organic pollutants and macrobenthic communities in Bohai Bay., 84: 336-341.

    Zhu, J.,and Wang, P., 2006. Ensemble Kalman smoother and ensemble Kalman filter approaches to the joint air quality state and emission estimation problem., 30: 5871-882 (in Chinense).

    (Edited by Xie Jun)

    DOI 10.1007/s11802-015-2501-8

    ISSN 1672-5182, 2015 14 (6): 941-950

    ? Ocean University of China, Science Press and Springer-Verlag Berlin Heidelberg 2015

    (September 30, 2013; revised January 14, 2014; accepted September 2, 2015)

    * Corresponding author. Tel: 0086-532-66782971 E-mail: xqinglv@ouc.edu.cn

    观看av在线不卡| 91精品三级在线观看| 91在线精品国自产拍蜜月| 国产免费现黄频在线看| 一级a做视频免费观看| 亚洲国产av新网站| 能在线免费看毛片的网站| 少妇被粗大猛烈的视频| 超碰97精品在线观看| videossex国产| 亚洲综合色网址| 亚洲成色77777| 国产无遮挡羞羞视频在线观看| 国产亚洲欧美精品永久| 日韩人妻高清精品专区| 18+在线观看网站| 国产精品蜜桃在线观看| 多毛熟女@视频| 在线观看免费日韩欧美大片 | 亚洲精品久久午夜乱码| 22中文网久久字幕| 国产午夜精品一二区理论片| 久久99热6这里只有精品| 视频中文字幕在线观看| 少妇人妻 视频| 免费黄色在线免费观看| av免费观看日本| 久久狼人影院| 五月天丁香电影| 亚洲伊人久久精品综合| 人妻制服诱惑在线中文字幕| 一边摸一边做爽爽视频免费| 天堂8中文在线网| 国产一区二区在线观看日韩| 日本黄大片高清| 国产伦理片在线播放av一区| 麻豆乱淫一区二区| 在线观看美女被高潮喷水网站| 大香蕉久久成人网| 亚洲不卡免费看| 我要看黄色一级片免费的| 伦理电影免费视频| av免费观看日本| 国产高清三级在线| 国产成人精品无人区| 国产精品嫩草影院av在线观看| 婷婷成人精品国产| 国产精品.久久久| 十八禁高潮呻吟视频| av电影中文网址| 最近最新中文字幕免费大全7| 亚洲精品自拍成人| 人妻一区二区av| 精品亚洲成a人片在线观看| 久久久精品区二区三区| 最近手机中文字幕大全| 亚洲精品久久午夜乱码| 午夜激情久久久久久久| 2022亚洲国产成人精品| 精品酒店卫生间| 精品人妻在线不人妻| 香蕉精品网在线| 中文字幕亚洲精品专区| 免费黄网站久久成人精品| 香蕉精品网在线| 十八禁高潮呻吟视频| 精品亚洲乱码少妇综合久久| 亚洲人成网站在线播| 成年av动漫网址| 日日啪夜夜爽| 亚洲精品一区蜜桃| 丝袜喷水一区| 久久精品国产鲁丝片午夜精品| 亚洲不卡免费看| 久久精品久久久久久久性| 美女xxoo啪啪120秒动态图| 国产亚洲午夜精品一区二区久久| 欧美日韩成人在线一区二区| 女性生殖器流出的白浆| 欧美另类一区| 夜夜看夜夜爽夜夜摸| 日韩电影二区| av黄色大香蕉| 欧美精品亚洲一区二区| 国产精品久久久久久av不卡| 精品少妇久久久久久888优播| 又粗又硬又长又爽又黄的视频| 国产国语露脸激情在线看| 视频在线观看一区二区三区| 色网站视频免费| h视频一区二区三区| 久久精品国产亚洲av天美| 一级黄片播放器| 丰满少妇做爰视频| 人人澡人人妻人| 爱豆传媒免费全集在线观看| 2022亚洲国产成人精品| av网站免费在线观看视频| 丰满饥渴人妻一区二区三| 一本久久精品| 啦啦啦视频在线资源免费观看| 国产精品女同一区二区软件| 日韩av免费高清视频| 国产熟女午夜一区二区三区 | 内地一区二区视频在线| 久久午夜福利片| 一级毛片电影观看| 一边亲一边摸免费视频| 一本色道久久久久久精品综合| 国产伦精品一区二区三区视频9| 亚洲熟女精品中文字幕| 22中文网久久字幕| 插逼视频在线观看| 尾随美女入室| 国产精品蜜桃在线观看| 2021少妇久久久久久久久久久| 日韩av免费高清视频| 亚洲怡红院男人天堂| 久久久亚洲精品成人影院| 99国产精品免费福利视频| 亚洲av成人精品一区久久| 欧美日韩亚洲高清精品| 久久人人爽人人片av| 欧美少妇被猛烈插入视频| 久久99精品国语久久久| 人妻夜夜爽99麻豆av| 亚洲经典国产精华液单| 国产日韩欧美在线精品| 99热这里只有是精品在线观看| 国产免费一区二区三区四区乱码| xxx大片免费视频| 日本黄大片高清| 91精品三级在线观看| 婷婷成人精品国产| 韩国av在线不卡| 在线免费观看不下载黄p国产| 黄色毛片三级朝国网站| 国产免费一级a男人的天堂| 9色porny在线观看| 在线看a的网站| 亚洲国产精品国产精品| 日本av免费视频播放| 美女中出高潮动态图| 亚洲第一av免费看| 日日啪夜夜爽| 亚洲av在线观看美女高潮| 热re99久久精品国产66热6| 男人爽女人下面视频在线观看| 青春草亚洲视频在线观看| 欧美日韩精品成人综合77777| 中文字幕人妻丝袜制服| 国产欧美亚洲国产| 午夜影院在线不卡| 日本91视频免费播放| 黑人巨大精品欧美一区二区蜜桃 | 免费日韩欧美在线观看| 我要看黄色一级片免费的| 午夜激情福利司机影院| 自线自在国产av| 亚洲国产日韩一区二区| 免费高清在线观看日韩| 一级黄片播放器| 丰满迷人的少妇在线观看| 中国国产av一级| 80岁老熟妇乱子伦牲交| 亚洲精品aⅴ在线观看| 2018国产大陆天天弄谢| 最近2019中文字幕mv第一页| 老女人水多毛片| 看免费成人av毛片| 久久久国产精品麻豆| 国产成人精品在线电影| 久久人妻熟女aⅴ| 久久精品国产自在天天线| xxx大片免费视频| 久久久欧美国产精品| 晚上一个人看的免费电影| 亚洲怡红院男人天堂| 如何舔出高潮| 满18在线观看网站| 亚洲av免费高清在线观看| 国产成人免费无遮挡视频| 精品久久久精品久久久| 高清欧美精品videossex| 亚洲人与动物交配视频| 熟女人妻精品中文字幕| 亚洲欧美一区二区三区黑人 | 亚洲精品亚洲一区二区| 日韩,欧美,国产一区二区三区| 夜夜骑夜夜射夜夜干| 美女国产高潮福利片在线看| 少妇熟女欧美另类| 丝瓜视频免费看黄片| 色5月婷婷丁香| av.在线天堂| 成人黄色视频免费在线看| 欧美另类一区| 国模一区二区三区四区视频| 国产精品久久久久久精品古装| 免费黄网站久久成人精品| 国产精品国产av在线观看| 成人亚洲欧美一区二区av| 亚洲美女黄色视频免费看| 欧美国产精品一级二级三级| 久久影院123| 亚洲成人手机| 热re99久久精品国产66热6| 在线观看一区二区三区激情| 夜夜爽夜夜爽视频| 国产精品99久久久久久久久| 亚洲精品aⅴ在线观看| 日韩不卡一区二区三区视频在线| 有码 亚洲区| a级毛片免费高清观看在线播放| 亚洲精品乱久久久久久| 中国国产av一级| 下体分泌物呈黄色| 国产色爽女视频免费观看| 精品少妇黑人巨大在线播放| 国产黄色免费在线视频| 亚洲综合色惰| 午夜福利视频精品| 婷婷色av中文字幕| 久久这里有精品视频免费| 日本黄大片高清| www.色视频.com| 国产精品一区二区在线不卡| 国产成人a∨麻豆精品| 国产亚洲一区二区精品| 一级毛片aaaaaa免费看小| 免费av不卡在线播放| 日本av免费视频播放| 男女国产视频网站| av女优亚洲男人天堂| 97在线人人人人妻| 少妇高潮的动态图| 中文天堂在线官网| 国产精品国产三级国产av玫瑰| 久久久久久久精品精品| 免费观看的影片在线观看| 美女视频免费永久观看网站| 亚洲无线观看免费| 久久久久久久久久久久大奶| 日韩av免费高清视频| av电影中文网址| 精品亚洲乱码少妇综合久久| 欧美老熟妇乱子伦牲交| 久久久午夜欧美精品| 伦理电影免费视频| 欧美日韩av久久| 国产精品麻豆人妻色哟哟久久| 亚洲av二区三区四区| 99热国产这里只有精品6| 91在线精品国自产拍蜜月| 飞空精品影院首页| 另类亚洲欧美激情| 国产有黄有色有爽视频| 成人国产麻豆网| 一区二区三区四区激情视频| 一级黄片播放器| 久久毛片免费看一区二区三区| 国产精品 国内视频| 亚洲国产欧美在线一区| 亚洲高清免费不卡视频| 中文字幕亚洲精品专区| 色5月婷婷丁香| 亚洲精品视频女| 国产精品一区二区在线不卡| 日本黄色片子视频| 午夜影院在线不卡| av电影中文网址| 成人黄色视频免费在线看| 亚洲综合精品二区| h视频一区二区三区| 国产av码专区亚洲av| 日韩一区二区视频免费看| 黄色配什么色好看| 蜜臀久久99精品久久宅男| 日韩强制内射视频| 免费黄色在线免费观看| 在线观看免费视频网站a站| 满18在线观看网站| 国产男女内射视频| a 毛片基地| 精品视频人人做人人爽| 国产女主播在线喷水免费视频网站| 2018国产大陆天天弄谢| 久久99热6这里只有精品| 麻豆成人av视频| 大香蕉久久成人网| 国产精品一区二区在线观看99| 国产午夜精品久久久久久一区二区三区| 日本爱情动作片www.在线观看| 在线亚洲精品国产二区图片欧美 | 在线免费观看不下载黄p国产| 日韩电影二区| av网站免费在线观看视频| 激情五月婷婷亚洲| 最近的中文字幕免费完整| 日韩成人伦理影院| 22中文网久久字幕| 精品午夜福利在线看| 中文字幕精品免费在线观看视频 | 亚洲成人一二三区av| 精品人妻熟女毛片av久久网站| 国产精品无大码| 亚洲欧美日韩卡通动漫| 成人二区视频| 成人亚洲精品一区在线观看| 26uuu在线亚洲综合色| 男男h啪啪无遮挡| av在线老鸭窝| 色94色欧美一区二区| 美女国产视频在线观看| 性色avwww在线观看| 国产高清三级在线| 国产成人精品无人区| 啦啦啦在线观看免费高清www| 亚洲精品第二区| 99国产精品免费福利视频| 一级,二级,三级黄色视频| 中国国产av一级| 99热这里只有是精品在线观看| 国产欧美亚洲国产| 亚洲精品日韩av片在线观看| 国产免费视频播放在线视频| a级毛色黄片| 九九爱精品视频在线观看| 在线天堂最新版资源| 婷婷色综合大香蕉| 9色porny在线观看| 亚洲综合精品二区| 99热这里只有精品一区| 18+在线观看网站| videosex国产| 七月丁香在线播放| 一区二区三区四区激情视频| 中文字幕亚洲精品专区| 国产精品一区二区三区四区免费观看| 美女国产视频在线观看| 亚洲av免费高清在线观看| 99久久精品一区二区三区| 超色免费av| 午夜福利视频精品| 一个人免费看片子| 永久网站在线| 免费av中文字幕在线| 丝袜在线中文字幕| 十八禁网站网址无遮挡| 最新中文字幕久久久久| 一区二区三区乱码不卡18| 免费人成在线观看视频色| 亚洲精品视频女| 久久精品久久精品一区二区三区| 午夜精品国产一区二区电影| 精品熟女少妇av免费看| 一级毛片电影观看| 在线亚洲精品国产二区图片欧美 | 久久人人爽av亚洲精品天堂| 精品99又大又爽又粗少妇毛片| 亚洲经典国产精华液单| 亚洲精品乱久久久久久| 久久久久国产精品人妻一区二区| 不卡视频在线观看欧美| 日本vs欧美在线观看视频| 少妇 在线观看| 大香蕉久久网| 久久久久久伊人网av| 色婷婷久久久亚洲欧美| 欧美激情 高清一区二区三区| 69精品国产乱码久久久| 91精品国产国语对白视频| 天天影视国产精品| 男女啪啪激烈高潮av片| 青春草亚洲视频在线观看| 高清在线视频一区二区三区| 夜夜看夜夜爽夜夜摸| 久久人人爽av亚洲精品天堂| 黑人巨大精品欧美一区二区蜜桃 | 人体艺术视频欧美日本| av国产久精品久网站免费入址| 久久精品夜色国产| 美女cb高潮喷水在线观看| 男女无遮挡免费网站观看| 久久精品国产亚洲av涩爱| 成年女人在线观看亚洲视频| 这个男人来自地球电影免费观看 | 久久影院123| 久久精品国产a三级三级三级| 中文天堂在线官网| 亚洲欧美日韩卡通动漫| 亚洲国产毛片av蜜桃av| 亚洲精品久久久久久婷婷小说| 毛片一级片免费看久久久久| 一级毛片aaaaaa免费看小| 亚洲熟女精品中文字幕| 亚洲欧美清纯卡通| 赤兔流量卡办理| 少妇人妻 视频| 最近2019中文字幕mv第一页| 少妇高潮的动态图| 欧美精品国产亚洲| 狠狠精品人妻久久久久久综合| 精品人妻偷拍中文字幕| 成人影院久久| 97在线视频观看| 婷婷色av中文字幕| 美女大奶头黄色视频| 美女福利国产在线| 亚洲精品久久成人aⅴ小说 | 欧美三级亚洲精品| 在线看a的网站| 亚洲精品国产色婷婷电影| 丝瓜视频免费看黄片| a级毛片在线看网站| 久久青草综合色| 午夜免费观看性视频| 国产伦精品一区二区三区视频9| 人妻少妇偷人精品九色| 欧美人与善性xxx| 亚州av有码| 欧美 日韩 精品 国产| 边亲边吃奶的免费视频| 亚洲精品av麻豆狂野| 日韩精品免费视频一区二区三区 | tube8黄色片| 成人手机av| av不卡在线播放| 国产精品一区二区三区四区免费观看| 国产精品人妻久久久影院| 成人二区视频| 91久久精品电影网| 男男h啪啪无遮挡| 久久久欧美国产精品| www.色视频.com| 香蕉精品网在线| 亚洲av欧美aⅴ国产| 国产精品一区二区在线不卡| 精品久久久久久久久av| 亚洲av综合色区一区| 日本黄大片高清| 亚洲欧洲精品一区二区精品久久久 | 最黄视频免费看| 婷婷色综合www| 精品人妻熟女毛片av久久网站| 日韩av免费高清视频| 十八禁高潮呻吟视频| 国产精品一国产av| 亚洲欧美一区二区三区国产| 大陆偷拍与自拍| 国产av国产精品国产| 精品人妻在线不人妻| 亚洲av男天堂| 欧美亚洲日本最大视频资源| 国产女主播在线喷水免费视频网站| 免费观看性生交大片5| 另类亚洲欧美激情| 成人国语在线视频| 超碰97精品在线观看| 天天操日日干夜夜撸| 精品久久久精品久久久| 黄色配什么色好看| 国产亚洲最大av| 国产色婷婷99| 搡女人真爽免费视频火全软件| 国产精品国产三级国产专区5o| 成人毛片60女人毛片免费| 18禁在线无遮挡免费观看视频| 日日摸夜夜添夜夜添av毛片| 日本猛色少妇xxxxx猛交久久| 亚洲精品一区蜜桃| 3wmmmm亚洲av在线观看| 日韩一区二区三区影片| 母亲3免费完整高清在线观看 | 亚洲精品久久成人aⅴ小说 | 2021少妇久久久久久久久久久| 熟女电影av网| 婷婷色综合大香蕉| 久久久久国产精品人妻一区二区| 人体艺术视频欧美日本| 日本爱情动作片www.在线观看| 少妇被粗大的猛进出69影院 | 欧美xxⅹ黑人| 91精品伊人久久大香线蕉| 麻豆成人av视频| av一本久久久久| 国产一区有黄有色的免费视频| 精品久久久久久电影网| 亚洲精品456在线播放app| 亚洲一区二区三区欧美精品| 欧美日韩av久久| 免费观看av网站的网址| 久久国产亚洲av麻豆专区| 国产成人freesex在线| 18禁在线播放成人免费| 成年女人在线观看亚洲视频| 在线观看美女被高潮喷水网站| 国产精品麻豆人妻色哟哟久久| 九九久久精品国产亚洲av麻豆| 亚洲精品中文字幕在线视频| 欧美精品人与动牲交sv欧美| 久久国内精品自在自线图片| 日韩熟女老妇一区二区性免费视频| 18禁在线播放成人免费| 99久久精品一区二区三区| 精品久久蜜臀av无| 精品亚洲成a人片在线观看| 亚洲怡红院男人天堂| 内地一区二区视频在线| 免费观看a级毛片全部| 一级黄片播放器| 天天躁夜夜躁狠狠久久av| av不卡在线播放| 夜夜看夜夜爽夜夜摸| 午夜91福利影院| 成人亚洲精品一区在线观看| 人人妻人人添人人爽欧美一区卜| 人妻 亚洲 视频| 日韩av免费高清视频| 国产成人精品一,二区| 久久久久久伊人网av| 特大巨黑吊av在线直播| 亚洲美女黄色视频免费看| 国产视频内射| 国产极品天堂在线| 少妇人妻精品综合一区二区| 国产精品 国内视频| 亚洲欧美中文字幕日韩二区| 99视频精品全部免费 在线| 99久久综合免费| 大香蕉久久网| 久久97久久精品| av有码第一页| 91在线精品国自产拍蜜月| 久久精品国产亚洲av涩爱| 青青草视频在线视频观看| 午夜福利,免费看| 制服人妻中文乱码| 欧美老熟妇乱子伦牲交| 欧美少妇被猛烈插入视频| 日韩欧美一区视频在线观看| 日韩中文字幕视频在线看片| 乱码一卡2卡4卡精品| 美女国产视频在线观看| 亚洲人成网站在线播| 最黄视频免费看| 欧美日韩亚洲高清精品| 欧美日韩精品成人综合77777| 蜜桃国产av成人99| 日本色播在线视频| 美女视频免费永久观看网站| 国产 一区精品| 如何舔出高潮| 26uuu在线亚洲综合色| 国产欧美日韩一区二区三区在线 | 九色亚洲精品在线播放| 男的添女的下面高潮视频| 最近2019中文字幕mv第一页| 91午夜精品亚洲一区二区三区| 久久精品夜色国产| 亚洲一区二区三区欧美精品| 亚洲精品成人av观看孕妇| 国产精品国产三级专区第一集| 韩国av在线不卡| 亚洲美女黄色视频免费看| 飞空精品影院首页| 最黄视频免费看| 亚洲美女搞黄在线观看| 国产欧美日韩综合在线一区二区| 成年av动漫网址| 亚洲av.av天堂| 一级爰片在线观看| 有码 亚洲区| 两个人免费观看高清视频| 免费久久久久久久精品成人欧美视频 | 18禁裸乳无遮挡动漫免费视频| 午夜福利视频在线观看免费| 精品久久蜜臀av无| 日韩一区二区视频免费看| av一本久久久久| 高清视频免费观看一区二区| 免费大片18禁| 免费大片黄手机在线观看| 热re99久久国产66热| 黑丝袜美女国产一区| 欧美 亚洲 国产 日韩一| 国产成人精品无人区| 在线免费观看不下载黄p国产| 日本与韩国留学比较| 嘟嘟电影网在线观看| 日韩伦理黄色片| videos熟女内射| 亚洲av福利一区| 国产精品国产av在线观看| 青春草亚洲视频在线观看| 久久久久久久久久成人| 国产黄片视频在线免费观看| 亚洲成色77777| 亚洲四区av| 久久热精品热| 免费看av在线观看网站| 国产欧美日韩综合在线一区二区| 高清在线视频一区二区三区| 精品午夜福利在线看| 日韩欧美一区视频在线观看| 美女xxoo啪啪120秒动态图| 岛国毛片在线播放| 亚洲国产最新在线播放| 天天操日日干夜夜撸| 少妇人妻精品综合一区二区| 欧美日韩视频精品一区| 乱人伦中国视频| av又黄又爽大尺度在线免费看| 久久精品熟女亚洲av麻豆精品| 国产高清国产精品国产三级|