基于YOLO算法的海洋中尺度涡三维结构构建

海洋中尺度涡是一种常见的中尺度海洋现象,研究海洋中尺度涡的分布及运动特性对航运、气候、军事等具有重要作用,海洋中尺度涡的识别是海洋学和计算机科学领域的一个热门研究课题。运用深度学习的方法和框架,对中尺度涡的二维识别和三维结构构建展开研究分析。首先,获取全球海洋再分析数据并进行流线可视化,构建涡旋流线数据集;其次,利用YOLO v5s卷积神经网络对涡旋流线数据集进行训练,并对南海区域中尺度涡进行有效检测。实验结果表明,YOLO v5s训练后得到最优模型经过测试,平均检测精度均值达到了86.10%;最后,根据涡旋检测结果,对检测出的同时刻不同深度的涡旋判断是否属于同一涡旋,确定后进行该涡旋的三维结构构建。     

能见度集合预报及后处理技术应用

基于污染物情况、环流系统和时空分布特征分析,利用神经网络对历史数据进行建模,生成了能见度集合预报产品。在2022年冬季的TS评分检验中,预报产品优于欧洲中期数值预报中心模式(ECMWF)的能见度预报产品。利用概率匹配、最优百分位和神经网络三种后处理方法生成后处理产品,这些产品的TS评分优于集合预报产品。预报输入的ECMWF模式2 m湿度与实况的偏差是误差的主要来源。利用集成方法对三种后处理产品进行集成,其TS评分结果在低能见度区间总体接近或略优于原始产品。生成的能见度集合预报后处理最优集成预报产品成功提高了对中期延伸期能见度天气的预测准确性。     

Effect of the Initial Vortex Structure on Intensity Change During Eyewall Replacement Cycle of Tropical Cyclones: A Numerical Study

This study investigates the effect of the initial tropical cyclone (TC) vortex structure on the intensity change during the eyewall replacement cycle (ERC) of TCs based on two idealized simulations using the Weather Research and Forecasting (WRF) model. Results show that an initially smaller TC with weaker outer winds experienced a much more drastic intensity change during the ERC than an initially larger TC with stronger outer winds. It is found that an initially larger TC vortex with stronger outer winds favored the development of more active spiral rainbands outside the outer eyewall, which slowed down the contraction and intensification of the outer eyewall and thus prolonged the duration of the concentric eyewall and slow intensity evolution. In contrast, the initially smaller TC with weaker outer winds corresponded to higher inertial stability in the inner core and weaker inertial stability but stronger filamentation outside the outer eyewall. These led to stronger boundary layer inflow, stronger updraft and convection in the outer eyewall, and suppressed convective activity outside the outer eyewall. These resulted in the rapid weakening during the formation of the outer eyewall, followed by a rapid re-intensification of the TC during the ERC. Our study demonstrates that accurate in- itialization of the TC structure in numerical models is crucial for predicting changes in TC intensity during the ERC. Additionally, monitoring the activity of spiral rainbands outside the outer eyewall can help to improve short-term intensity forecasts for TCs experiencing ERCs.     

广西钦江河口枯季水动力特征及其对冷空气的响应

河口作为陆海相互作用的关键带,是河流入海泥沙及污染物的主要归宿地。钦江作为广西第二大河流,在茅尾海资源开发利用和生态环境保护与修复中发挥着重要作用。为进一步认识钦江河口地区的沉积动力过程,2021年10月27日至12月8日在此展开水动力和水体环境要素的连续观测。现场观测与分析结果表明,观测期间钦江河口潮汐类型为正规全日潮,浅水分潮显著,平均潮差为3.07 m,具有落潮优势;潮流以全日分潮流为主导,平均流速为0.12 m/s,运动形式主要为往复流;余流流向主要为西南方向。钦江河口潮汐和潮流性质较外湾和茅尾海中部海域全日潮特征更显著。枯季期间河口表层沉积物随潮汐运动表现为侵蚀-沉降交替的变化规律,垂向上侵蚀通量大于沉降通量,水平方向上悬沙净向海输运。冷空气过境带来的降雨使得钦江河口水体盐度降低、浊度增大,悬沙浓度及输沙量增加,同时冷空气南下时北风增强引起钦江河口减水效应并使潮差略有增加,水动力强度增大,增加表层沉积物的活动性,从而引起底部沉积物再悬浮强度和频率增加。     

预测有效波高的深度学习模型研究

研究基于RNN、LSTM、GRU深度学习模型,针对NOAA浮标数据集中的44013、44014、44017浮标的数据,通过斯皮尔曼相关性分析提高模型预测效果。实验结果表明,在进行相关性分析后,S-RNN、S-LSTM、 S-GRU的预测效果均比原始RNN、LSTM、GRU模型预测效果好。此外,提出一种基于LSTM的LSTM-Attention 波高预测模型,并进行相关实验,量化LSTM-Attention模型的预测效果,实验结果表明LSTM-Attention模型有更好的预测效果。为评估模型的泛化能力,研究还提出了一种采用邻近浮标数据进行学习,预测浮标缺失数据的方 法。实验结果表明,该方法的预测精度可以达到97.93%。本研究为海浪预测提供了新的方法和思路,也为未来深 度学习模型在海浪预测中的应用提供了参考。     

可解释性长短期记忆模型用于预测湖泊总磷浓度变化

对湖泊总磷的变化预测和来源识别对水资源调度和流域生态治理有着重要的意义,然而复杂的生化反应和水动力条件导致的非平稳性给湖泊总磷浓度的准确预测带来极大的困难。为克服这一挑战,本文引入了基于加权回归的季节趋势分解(seasonal and trend decomposition using Loess,STL)技术和夏普利加法(SHapley additive exPlanations,SHAP)结合长短期记忆网络(long short-term memory neural network,LSTM)和门控循环单元(gated recurrent unit,GRU)构建了一个可解释的预测框架,以增强对湖泊总磷浓度演变的预测并提高其可解释性。研究表明:(1)在骆马湖总磷浓度的预测中,该框架拥有较好的预报精度(R²=0.878),优于LSTM和卷积长短期记忆模型(convolutional neural networks and long short term memory network,CNN-LSTM)。当预测时间步长增加到8 h时,该框架有效提高了总磷浓度的预测精度,平均相对误差和均方根误差分别降低了47.1%和33.3%。从预测趋势来看,骆马湖在汛期的总磷平均浓度为0.158 mg/L,相较于非汛期的平均浓度,增加了202.1%。(2)运河来水是骆马湖总磷浓度最重要的影响因素,贡献权重为60.0%,并且不同断面(三湾、三场)的污染源受水动力、气象等因素的影响存在显著的时空差异。本文凸显了神经网络模型在预警水体污染方面的可实施性,并且为提高传统神经网络的学习能力和可解释性的开发与验证提供了重要方向。     

1996-2022年黄河三角洲行水河口岸线稳定性研究

黄河清八汊现行河口自改汊以来发生了巨大变化,监测其岸线变化,探讨其稳定程度对海岸带可持续发展以及海岸带韧性评估具有重要意义。本研究利用GPS、GIS、RS技术从1996—2022年黄河三角洲行水河口的220幅遥感影像中推断出年平均海岸线位置,同时根据行水河口摆动次数划分为5个阶段,并以此为基础对海岸线变迁及其稳定性进行定量分析。结果表明：行水河口岸线长期处于动态变化过程中,整体呈淤进状态,各岸段岸线时空变化特征不同,最大侵蚀（-73.89m/a）区出现在两丁坝之间,最大淤积（393.20m/a）区出现在河口区附近。研究区90%的岸线表现为较强淤积至严重淤积,稳定性指数由两丁坝之间（0.135）、2007年出汊前旧河口（0.068）、2007年出汊后新生河口（0.006）依次降低。入海水沙量、河口位置变迁以及沿岸输沙是影响岸线稳定性出现时空差异的主要原因。     

0～6 h定量降雪(雨)预报技术及其在温哥华冬奥会中的应用

利用中尺度数值预报模式与雷达回波外推技术相结合,围绕SNOWV10项目的科学目标,在自动判识降水相态(雨、雪或雨雪混合物)基础上,利用雷达反演降雪(雨)关系,建立短时(0～6 h)定量降雪(雨)预报方法,并集成到GRAPES-SWIFT临近预报系统,为温哥华冬奥会提供实时场馆预报产品.从2010年2月冬奥会使用情况看,所建立的定量降雪(雨)预报技术,一定程度满足高纬度冬季降雪(雨)临近预报预警的需求,但降水(雨或雪)预报量级偏小,针对场馆的降水预报起止时间节奏偏差较大,各时次预报一致性有待改进.     

季节内印度洋-西太平洋对流涛动对次季节-季节尺度大气可预报性的影响

利用非线性局部Lyapunov指数和条件非线性局部Lyapunov指数定量估计了季节内印度洋-西太平洋对流涛动（IPCO）和实时多变量Madden-Julian指数（RMM指数）可预报期限,量化了季节内IPCO对S2S尺度大气可预报性的贡献,深入研究了季节内IPCO演变下S2S尺度可预报期限空间分布的变化规律。结果表明：（1）与RMM指数相比,季节内IPCO指数可预报性更强,可预报期限达到31天左右,比RMM指数高出2周以上;（2）印度洋-西太平洋区域S2S尺度大气可预报性最强,可预报期限达到30天以上,其中季节内IPCO是该地区的主要可预报性来源之一,其贡献达到6天,占总可预报期限的25%以上;（3）随着季节内IPCO的演变,印度洋-西太平洋地区S2S尺度大气可预报性有空间结构变化,表现为可预报期限异常的传播和振荡。S2S尺度大气可预报期限正负异常沿季节内IPCO传播路径,一支以赤道中西印度洋为起点北传至印度半岛,一支向东传播,经过海洋性大陆到赤道西太平洋后向北传播,到达日本南部。同时,可预报性异常的传播在在东印度洋和西太平洋表现出反向变化的特征,形成东西两极振荡,当季节内IPCO向正位相发展时,东印度洋具有更强的可预报性,西太平洋具有更弱的可预报性,反之亦然。季节内IPCO的发展（衰退）可使东印度洋（西太平洋）S2S尺度大气可预报性更强,表明模式预报技巧对此具有更大的提升空间。     

Vertical structures of PM10 and PM 2.5 and their dynamical character in low atmosphere in Beijing urban areas

The vertical structures and their dynamical character of PM2.5 and PM10 over Beijing urban areas are revealed using the 1 min mean continuous mass concentration data of PM2.5 and PM10 at 8, 100, and 320 m heights of the meteorological observation tower of 325 m at Institute of Atmospheric Physics, Chinese Academy of Sciences (IAP CAS tower hereafter) on 10―26 August, 2003, as well as the daily mean mass concentration data of PM2.5 and PM10 and the continuous data of CO and NO2 at 8, 100 (low layer), 200 (middle layer), and 320 m (high layer) heights, in combination with the same period meteorological field observation data of the meteorological tower. The vertical distributions of aerosols observed on IAP CAS tower in Beijing can be roughly divided into two patterns: gradually and rapidly decreasing patterns, I.e. The vertical distribution of aerosols in calm weather or on pollution day belongs to the gradually decreasing pattern, while one on clean day or weak cold air day belongs to the rapidly decreasing pattern. The vertical distributive characters of aerosols were closely related with the dynamical/thermal structure and turbulence character of the atmosphere boundary layer. On the clean day, the low layer PM2.5 and PM10 concentrations were close to those at 8 m height, while the concentrations rapidly decreased at the high layer, and their values were only one half of those at 8 m, especially, the concentration of PM2.5 dropped even more. On the clean day, there existed stronger turbulence below 150 m, aerosols were well mixed, but blocked by the more stronger inversion layer aloft, and meanwhile, at various heights, especially in the high layer, the horizontal wind speed was larger, resulting in the rapid decrease of aerosol concentration, I.e. Resulting in the obvious vertical difference of aerosol concentrations between the low and high layers. On the pollution day, the concentrations of PM2.5 and PM10 at the low, middle, and high layers dropped successively by, on average, about 10% for each layer in comparison with those at 8 m height. On pollution days, in company with the low wind speed, there existed two shallow inversion layers in the boundary layer, but aerosols might be, to some extent, mixed below the inversion layer, therefore, on the pollution day the concentrations of PM2.5 and PM10 dropped with height slowly; and the observational results also show that the concentrations at 320 m height were obviously high under SW and SE winds, but at other heights, the concentrations were not correlated with wind directions. The computational results of footprint analysis suggest that this was due to the fact that the 320 m height was impacted by the pollutants transfer of southerly flow from the southern peripheral heavier polluted areas, such as Baoding, and Shijiazhuang of Hebei Province, Tianjin, and Shandong Province, etc., while the low layer was only affected by Beijing's local pollution source. The computational results of power spectra and periods preliminarily reveal that under the condition of calm weather, the periods of PM10 concentration at various heights of the tower were on the order of minutes, while in cases of larger wind speed, the concentrations of PM2.5 and PM10 at 320 m height not only had the short periods of minute-order, but also the longer periods of hour order. Consistent with the conclusion previously drawn by Ding et al., that air pollutants at different heights and at different sites in Beijing had the character of "in-phase" variation, was also observed for the diurnal variation and mean diurnal variation of PM2.5 and PM10 at various heights of the tower in this experiment, again confirming the "in-phase" temporal/spatial distributive character of air pollutants in the urban canopy of Beijing. The gentle double-peak character of the mean diurnal variation of PM2.5 and PM10 was closely related with the evident/similar diurnal variation of turbulent momentum fluxes, sensible heat fluxes, and turbulent kinetic energy at various heights in the urban canopy. Besides, under the condition of calm weather, the concentration of PM2.5 and PM10 declined with height slowly, it was 90% of 8 m concentration at the low layer, a little lesser than 90% at the middle layer, and 80% at the high layer, respectively. Under the condition of weak cold air weather, the concentration remarkably dropped with height, it was 70% of 8 m concentration at the low layer, and 20%―30% at the middle and high layers, especially the concentration of PM2.5 was even lower.