【作者】 张娜；

【导师】 张庆河；

【作者基本信息】 天津大学 ， 海洋资源与环境， 2012， 博士

【摘要】 渤海海域冬季海水温度较低，每年都会结冰。冰情严重时会导致海冰灾害，给人类在海上的生命财产安全造成危害。通过卫星遥感获取海冰资料，结合气象条件深入研究海冰生消演变规律，利用数值模型模拟海冰演化，建立和完善海冰预报模型，对于防灾减灾具有十分重要的意义。为了获取更为精确的海冰预报基础资料，本文采用CART决策树方法从MODIS卫星遥感数据反演海冰，同时对可见光、近红外和热红外多波段进行自动分类计算，有效消除了传统阈值法反演高悬沙等特殊海洋环境时出现的海冰误判。反演结果采用高分辨率中国环境与灾害监测预报小卫星（HJ-1A/1B）进行校验，确保了海冰预报基础资料的精度。根据本文遥感反演的6个冬季辽东湾海冰面积，结合锦州与营口两个气象观测站的实测气象资料分析，可以认为辽东湾海冰面积增长主要受低气温影响，其中低温天气时发生的离岸向大风累积次数直接影响年度海冰面积极值。通过回归分析发现低于-5°C的累积冻冰温度与冻冰阶段的海冰面积具有显著的线性相关关系。基于最小二乘法开发了辽东湾海冰面积的长期和短期两种预报公式，长期预测公式只需提供气象条件即可进行预报。短期预报公式在预报下一期海冰面积时除需给出气象条件外，还需给出上一期的海冰面积观测值。拟合及预测结果表明两个公式均能实现较为准确的海冰面积预报。通过进一步引入非线性逼近能力更为强大的BP神经网络模型用于开发海冰预报模型，预报精度得到进一步提高。本文最后通过提出适合渤海特点的模型参数，考虑海洋-海冰耦合的海冰动力过程和热力过程，将基于有限体积算法的非结构化网格FVCOM三维海洋模型用于2003~2004年冬季渤海海冰数值模拟。模拟结果表明，水动力、温盐以及海冰面积和厚度计算值与实测值吻合均较好，所建立的三维海冰数学模型输入参数合理，适合用于渤海中长期海冰数值模拟，具有较高的计算精度。FVCOM所采用的非结构化网格模型可实现对复杂岸线的准确描述及重点区域的局部加密，与气象预报相结合，未来可实现海冰的精细化模拟与预报。更多还原

【Abstract】 The water temperature in the Bohai Sea is low in winter and the sea ice occursevery year. In heavy ice years, the sea ice disaster would be induced and be harmful tothe safety of human life and property in the sea. Therefore, it is of great importancefor preventing and reducing sea disaster to develop prediction models of sea icethrough obtaining the sea ice information by satellite remote sensing, investigating theprocesses of sea ice formation and disappearance using meteorological data,simulating the sea ice evolution applying the mumerical model.To obtain basic data with high accuracy for sea ice forecasting, the methodologyof CART decision tree was utilized to retrieve sea ice from MODIS satallite remotingsensing data. The multi-bands, including visible light, near infrared light and thermalinfrared imaging were automatically used to classify sea ice and other category, andthe misjudgment of sea ice was effectively eliminated comparing with the traditionalthreshold methods when retrieving sea ice from MODIS data with special seaenvironment such as high sediment suspension. The retrieved results are verified bythe Small Satellite Constellation for Environment and Disaster Monitoring andForecasting(HJ-1A/1B)with high spatial resolution.Based on the sea ice acreage of6winters in the Liaodong Bay retrieved by remotesensing, combined with the analysis of measured data by meteorological stations inJinzhou and Yingkou, it is concluded that the accruement of sea ice area in theLiaodong Bay is mainly affected by the low air temperature, and especially theaccumulated times of strong wind offshore in days with lower air temperature directlyinfluences the annual maximum area of sea ice. It is found that the accumulatedfreezing air temperature lower than-5°C has significant linear correlation with seaice acreage in the frozen phase.The formulae for long term and short term forecasting of sea ice acreage in theLiaodong Bay were developed based on the least-square-method. The long termforecasting formula can run just on the given weather conditions. The short termforecasting formula should run on the given weather conditions and the last measuredvalue of sea ice acreage. The fitting and forecasting results show that both formulaecan rather accurately forecast the sea ice acreage. The BP neural networks model withstronger non-linear approximation capabilities was further used to develop the sea iceforecast model, and the accuracy is improved.In the end, a three-dimensional unstructered grid FVCOM model based on thefinite-volume solver was used for numerical simulation of sea ice in2003~2004winter in the Bohai Sea. The parameters of the model were adjusted to be suitable for the Bohai Sea, and the dynamical processes and thermodynamic processes of sea icewere fully incorporated through coupling of ocean and ice model. The simulatedresults show good agreement with measured data such as hydrodynamics, sea watertemperature, sea water salinity, sea ice area and thickness. It is indicated that thedeveloped3D sea ice model in the present dissertation is reasonable for parametersand suitable for long-term numerical simulation of sea ice with high accuracy. Theunstructured triangular meshes used in FVCOM can accurately describe the complexshoreline and be locally refined for the interesting region, therefore, combined withthe weather foarcast, the model can be applied in the refined simulation and forecastof sea ice in the future.更多还原