基于自组织网络的黄河口浑浊模式研究

传统浑浊度时空变化模式研究依靠野外观测实验,需要投入大量人力物力,模型适用范围亦非常有限。本文利用自组织网络（SOM）,直接从长时序遥感影像中提取典型浑浊模式,分析浑浊度年内、年际变化特征。以黄河河口附近海域为研究区,提取出近15年的6类典型浑浊模式。典型特征显示,研究区内主要有2个浑浊区,位于渤海湾西部和南部,以及河口外和莱州湾西北部;6类模式中四类以年为周期过渡更替,冬春季浑浊度较高,夏秋季浑浊度较低;多年浑浊模式逐渐由中浑浊向清澈模式变化,整体浑浊度有降低趋势。浑浊水体分布主要受河口潮流、环流等海洋动力和风浪影响,结合研究区气象观测数据分析,海面风浪变化是造成浑浊模式更替的主要原因,黄河入海泥沙影响范围仅局限于河口口门周边。利用统计参数分析和2007年各月悬浮泥沙浓度反演结果比较,评价SOM分类效果,结果表明SOM提取的模式间具有显著差异,一定程度上能够代替经验模型反映区域浑浊特征。SOM神经网络能够从长时序遥感影像中直接提取浑浊水体典型分布模式,分析海岸带地区水体浑浊度变化的时空特征,对了解复杂水体泥沙输运及优化水资源利用具有重要应用价值。

Turbidity Patterns Identification Based on Self-organizing Maps at Yellow River Estuary

In the study of spatial and temporal changes of the turbidity, traditional methods largely depend on field survey, which needs considerable manpower and materials. And these models are limited in different regions and time periods. With the help of self-organizing map (SOM) clustering, typical turbidity patterns can be extracted from plenty of remote sensing imageries covering long time periods. It can also facilitate the analysis of intra-annual and inter-annual variations. Taking the Yellow River estuary as our study area, six turbidity patterns were revealed from 2000-2015 MODIS data. Two main turbid areas appeared on the turbidity feature maps, located in the western and southern Bohai Bay, and outside the estuary in north-western Laizhou Bay. Four patterns appeared annually, of which the turbid level in winter and spring is higher than that in autumn and summer. And during the last fifteen years, turbidity patterns have gradually changed from middle turbidity to Clean, showing a declining trend of the overall turbid level. The hydrological data from Yellow River, meteorological observation data of wind and wave on the sea surface and ocean dynamics in the estuary were combined to detect the contributing factors of turbidity patterns. Spatial distribution was mainly influenced by wind waves and ocean dynamics, such as tide and circulating current. And intra-annual changes of dominant turbidity pattern were mainly caused by wind and wave on the sea surface, while the influence of sediment transportation from Yellow River is only limited around the estuary. SOM clustering results were evaluated from two perspectives: calculation of statistic parameters for quantitative analysis and inversion of the concentration of total suspended matter in the study area in 2007 for comparison with empirical model. It showed that there were significant differences between SOM patterns, and this method could reveal similar turbid features as the empirical models do. Thus, SOM is an effective and indispensable method to identify turbidity patterns and can directly extract typical features from long time series remote sensing imageries. This method significantly facilitates the study on spatial and temporal variation of water turbidity in coastal areas, which is of great practical value for research on sediment transport and water utilization in complex water bodies.