Abstract：
The transfer of dissolved silicate (DSi) from land to coastal environments is a crucial part of global biogeochemical cycling. However, the retrieval of coastal DSi distribution is challenging due to the spatiotemporal non-stationarity and nonlinearity of modeling processes and the low resolution of in situ sampling. To explore the coastal DSi changes in a higher spatiotemporal resolution, this study developed a spatiotemporally weighted intelligent method based on a geographically and temporally neural network weighted regression (GTNNWR) model, a Data-Interpolating Empirical Orthogonal Functions (DINEOF) model, and satellite observations. For the first time, the complete surface DSi concentrations of 2182 days at the 500-meter and 1-day resolution in the coastal sea of Zhejiang Province, China, were obtained (Testing R2 = 78.5 %) by using 2901 in situ records with concurrent remote sensing reflectance. The long-term and large-scale distributions of DSi reflected the changes in coastal DSi under the influences of rivers, ocean currents, and biological effects across multiple spatiotemporal scales. Benefiting from the high-resolution modeling, this study found that the surface DSi concentration had at least 2 declines during a diatom bloom process, which can provide crucial signals for the timely monitoring and early warning of diatom blooms and guide the management of eutrophication. It was also indicated that the correlation coefficient between the monthly DSi concentration and the Yangtze River Diluted Water velocities reached -0.462**, quantitatively revealing the significant influence of the terrestrial input. In addition, the daily-scale DSi fluctuations resulting from typhoon transits were finely characterized, which greatly reduces the monitoring cost compared with the field sampling. Therefore, this study developed an effective data-driven-based method to help explore the fine-scale dynamic changes of surface DSi in coastal seas.
摘要：
溶解硅酸盐（DSi）从陆地到沿海环境的转移是全球生物地球化学循环的重要组成部分。然而,由于建模过程的时空非平稳性和非线性以及原位采样的低分辨率，沿海DSi分布的检索具有挑战性。以更高的时空分辨率探索沿海DSi的变化，这项研究开发了一种基于地理和时间神经网络加权回归（GTNNWR）模型的时空加权智能方法，数据插值经验正交函数(DINEOF)模型，和卫星观测。第一次,在浙江省沿海海域的500米和1天分辨率下，2182天的完整表面DSi浓度，中国,通过使用2901个具有并发遥感反射率的原位记录获得(测试R2=78.5%)。DSi的长期和大规模分布反映了河流影响下沿海DSi的变化，洋流,和跨多个时空尺度的生物效应。受益于高分辨率建模，这项研究发现，在硅藻开花过程中，表面DSi浓度至少下降了2次，可以为硅藻水华的及时监测和预警提供关键信号，指导富营养化管理。还表明，月DSi浓度与长江稀释水速的相关系数达到-0.462**，定量揭示了地面输入的显著影响。此外,台风过境导致的每日DSi波动得到了很好的表征，与现场采样相比,大大降低了监测成本。因此,这项研究开发了一种有效的基于数据驱动的方法，以帮助探索沿海海域表面DSi的精细尺度动态变化。
