基于正交匹配追踪的声层析方法

声速剖面的变化会对声传播产生较大的影响,经验正交函数模型经常用来实现对声速剖面数据的简化描述。然而在内波、湍流等海水不均匀性存在时,这种正则化操作会造成声速重构精度的大幅降低。本文利用字典学习生成声速剖面的非正交原子,在稀疏编码时采用正交匹配追踪（OMP,Orthogonal Matching Pursuit）算法,更新字典则使用KSVD （Kernel Singular Value Decomposition）的字典更新算法。由于字典学习不需要强制使用正交条件,对于训练数据更加灵活,从而可以使用少数的原子组合达到更高的重构精度。利用一次浅海声学实验多次测量的声速剖面研究了海水声速剖面的经验正交函数表示和字典学习,研究表明：相比于正交函数表示,学习字典可以利用少数原子（甚至一个原子）更好的解释声速剖面扰动。字典学习可以提高声速剖面的稀疏性,从而提高声速剖面的反演精度。     

时间尺度重构EEMD-GRNN改进模型预测PM2.5的研究

合理构建PM_2.5浓度预测模型是科学、准确地预测PM_2.5浓度变化的关键。传统PM_2.5预测EEMD-GRNN模型具有较好的预测精度,但是存在过于关注研究数据本身而忽略其物理意义的不足。本研究基于南京市2014-2017年PM_2.5浓度时间序列数据,分析PM_2.5浓度多尺度变化特征及其对气象因子和大气污染因子的尺度响应,基于时间尺度重构进行EEMD-GRNN模型的改进与实证研究。南京市样本数据PM_2.5浓度变化表现为明显的天际尺度和月际尺度,从重构尺度（天际、月际）构建GRNN模型更具有现实意义;同时,PM_2.5对PM₁₀、NO₂、O₃、RH、MinT等因子存在多尺度响应效应,以其作为GRNN模型中的输入变量更具有时间序列上的解释意义。改进后的EEMD-GRNN模型具有更高的PM_2.5浓度预测精度,MAE、MAPE、RMSE和R²分别为6.17、18.41%、8.32和0.95,而传统EEMD-GRNN模型的模型有效性检验结果分别为8.37、27.56%、11.56、0.91。对于高浓度天（PM_2.5浓度大于100 μg/m³）的预测,改进模型更是全面优于传统EEMD-GRNN模型,MAPE为12.02%,相较于传统模型提高了9.03%。     

声速剖面EOF表示的第一模态解析

为实现对声速剖面EOF表示后第一模态时间系数和空间函数变化规律的解析,提出了一种简化的声速剖面变化模型,即"拐点"深度值和声速值的变化;声速梯度的变化和表层海水温度周期性变化所引起的海水声速变化,通过将4种因素所引起的第一模态空间函数的变化规律与实际声速剖面簇第一模态空间函数的变化规律对比,分析引起实测声速剖面变化的主要因素,最后,分别用深海和浅海实测声速剖面数据对其进行验证。     

梅雨降水季节预测的多方法比较

基于1961—2000年逐月降水观测资料和全球大气再分析资料,分析了6—7月长江中下游(108°～123°E,27°～33°N)梅雨的时空分布特征。通过观测诊断和数值试验确定了影响梅雨异常偏多的3个前期因子:4—5月平均的西北太平洋海平面气压正异常;3月至5月北大西洋海平面气压负变压倾向;1月至4月西伯利亚的2 m温度负倾向。利用这3个具有物理意义的影响因子构建了梅雨季节预测模型,该模型在训练期(1961—2000年)和独立预测期(2001—2022年)均具有显著的预测技巧(相关系数分别为0.79和0.77,均方根误差分别为0.59和0.68)。同时,基于相似的潜在预测因子,对比了利用偏最小二乘回归方法和5种机器学习方法(随机森林、轻量级梯度提升机、自适应提升、类别型特征提升、极端梯度提升)建立的预测模型的技巧。虽然训练期(1961—2000年)偏最小二乘回归和机器学习建模拟合效果更高,但在独立预测期(2001—2022年)上述模型的预测技巧显著降低(相关系数均低于0.44,均方根误差均大于0.93),出现了明显的过拟合问题。本研究强调梅雨的短期气候预测应建立在物理机制基础之上,而使用机器学习方法需谨慎。     

一种非监督快速端元提取方法

正交子空间投影(OSP)方法广泛用于目标与背景的分离之中,对于高光谱影像,OSP可用于目标提取和混合像元分解,但缺点是需要端元的先验知识。针对这一问题,本文基于OSP的原理提出了一种非监督快速端元提取方法。实验使用模拟高光谱数据和由OM ISⅠ获取的真实高光谱数据,结果精度令人满意,证明了本文算法进行端元自动提取的可行性。     

轮廓波变换在遥感目标图像检索中的应用

除具有纹理属性外,遥感图像中还包含大量的结构性边缘特征。如何有效捕捉这些特征中的信息进行检索,成为提高遥感图像检索效率的关键。依据Contourlet在离散域的多尺度几何分析的功能,提出一种利用Contourlet子带能量分布特性提取原始图像在多个尺度下的边缘方向信息进行检索的方法。针对Cont-ourlet变换响应不同特征方向能力的差别,采用正交补偿法加以改正,经过傅里叶算子的处理,从而实现旋转不变结构性边缘特征的检索。最后,在遥感目标图像上所作的实验证实了该算法的有效性。     

Statistical analysis and forecasts of long-term sandbank evolution at Great Yarmouth, UK

A data-driven model has been developed to analyse the long-term evolution of a sandbank system and to make ensemble predictions in a period of 8 years. The method uses a combination of empirical orthogonal function (EOF) analysis, (to define spatial and temporal patterns of variability), jack-knife resampling, (to generate an ensemble of EOFs), a causal auto-regression technique, (to extrapolate the temporal eigenfunctions), and straightforward statistical analysis of the resulting ensemble of predictions to determine a ‘forecast’ and associated uncertainty. The methodology has been applied to a very demanding site which includes a curved shoreline and a group of mobile nearshore sandbanks. The site is on the eastern coast of the UK and includes the Great Yarmouth sandbanks and neighbouring shoreline. A sequence of 33 high quality historical survey charts reaching back to 1848 have been used to analyse the patterns and to predict morphological evolution of the sandbank system. The forecasts demonstrate an improved skill relative to an assumption of persistence, but suffer in locations where there are propagating features in the morphology that are not well-described by EOFs.     

Water level variations in the Neuse and Pamlico Estuaries,North Carolina due to local and remote forcing

Water level time series records from the Neuse and Pamlico River Estuaries were statistically compared to local and distant wind field data, water level records within the Pamlico Sound and also coastal ocean sites to determine the relative contribution of each time series to water levels in the Neuse and Pamlico Estuaries. The objectives of this study were to examine these time series data using various statistical methods (i.e. autoregressive, empirical orthogonal function analysis (EOF), exploratory data analysis (EDA)) to determine short- and long-time-scale variability, and to develop predictive statistical models that can be used to estimate past water level fluctuations in both the Neuse Estuary (NE) and Pamlico Estuary (PE). Short- and long-time-scale similarities were observed in all time series of estuarine, Pamlico Sound and subtidal coastal ocean water level and wind component data, due to events (nor'easters, fronts and tropical systems) and seasonality. Empirical orthogonal function analyses revealed a strong coastal ocean and wind field contribution to water level in the NE and PE. Approximately 95% of the variation was captured in the first two EOF components for water level data from the NE, sound and coastal ocean, and 70% for the PE, sound and coastal ocean. Spectral density plots revealed strong diurnal signals in both wind and water level data, and a strong cross correlation and coherency between the NE water level and the North/South wind component. There was good agreement between data and predictions using autoregressive statistical models for the NE (R² = 0.92) and PE (R² = 0.76). These methods also revealed significant autoregressive lags for the NE (days 1 and 3) and for the PE (days 1, 2 and 3). Significant departures from predictions are attributed to local meteorological and hydrological events. The autoregressive techniques showed significant predictive improvement over ordinary least squares methods. The results are considered within the context of providing long time-scale hindcast data for the two estuaries, and the importance of these data for multidisciplinary researchers and managers.     

上游建库引水对健跳港冲淤影响的经验预测

从分析健跳港水文和泥沙特性出发,探讨了健跳港水动力和河床演变的基本特征,结果表明,尽管健跳港内落潮最大流速大于涨潮最大流速,但该港泥沙净输沙方向还是朝向港内;流域径流对健跳港海床稳定起着重要作用,维护健跳港冲淤平衡的主要动力来自洪水期的径流.利用水沙关系和河相关系模型预测了健跳港上游引水后对该港海床淤积的影响.     

黄东海表观光学量与固有光学量经验模式研究

表观光学特性和固有光学特性是海洋光学研究领域的两大主要研究内容。文中通过2003年黄东海试验和2007年908专项海洋光学调查所获取的表观光学参数中的遥感反射率Rrs(λ)、水下遥感反射率rrs(λ)、海水透明度Zs、下行辐照度漫衰减系数Kd(490)、固有光学参数中的后向散射系数bb(λ)和水体吸收系数a(λ),建立了区域性、季节性的表观光学量和固有光学量之间的经验模式,并对模式进行了误差分析和改进。