基于海洋表层遥感观测的次表层温度异常遥感反演研究

Currently multisource satellite remote sensing provides diverse sea surface observations at multi-spatial-temporal scales, but is limited to the surface layer of the ocean. The satellite remote sensing is unable to detect subsurface dynamic environmental information since the electromagnetic wave is confined to the surface layer, restricting the study of the ocean interior spatial-temporal variations. Since many important ocean processes and features are located well below the sea surface and at considerable depths, it is essential to determine the extent to which such surface remote sensing observations can be used to develop information about the ocean's interior. Many subsurface phenomena have surface manifestations that can be interpreted with the help of satellite measurements to derive key parameters of subsurface and deeper ocean processes. This study aims to break the sea surface bound for satellite measurements, and extend the satellite remote sensing observation range from sea surface to subsurface, so as to derive the subsurface dynamic parameters. This study proposes novel approaches based on machine learning to estimate subsurface temperature anomaly (STA) at large scale from satellite remote sensing observations including sea surface temperature (SST), sea surface height (SSH), sea surface salinity (SSS), sea surface wind field (SSW) as well as their derived features with the help of Argo in situ data for model training and testing. By setting up reliable empirical statistical and machine learning models, we can derive the subsurface temperature anomaly (STA) over large scale, and then employ Argo gridded data for result validation and accuracy evaluation. Subsurface and deeper ocean remote sensing has the ability to derive ocean interior dynamic parameters (especially the thermohaline structure) and enables us to characterize subsurface and deeper ocean processes and features and their implications for the climate change.

目前多源卫星遥感提供了丰富的多时空尺度对海观测资料。然而，受海表对电磁波束缚的影响，海洋次表层的动力环境信息传统遥感难以探测，大都依靠现场观测获取，制约了对海洋内部时空变化的研究。虽然卫星遥感仅能直接获取海表参数信息，但许多海洋次表层现象都会在海洋表面有所表征，因此结合海表表征现象与机制模型，可推演海洋次表层关键动力参量，如温度异常。本项目旨在打破海表对卫星遥感的束缚，将卫星对海观测拓展到海表以下，以感知海洋次表层热力异常，拓展卫星遥感对海观测范围。充分利用表层多源卫星观测资料，如海表温度、海表高度、海表盐度及海表风场等参量，通过构建合理可靠的经验统计与机器学习模型，反演海洋次表层温度异常，用Argo实测数据做验证开展精度评价。该项目的研究有助于发展深海遥感技术，可为海洋内部三维动力过程研究提供数据支持，对于认识海洋内部的热力变异、动力过程及其在全球气候变化过程中的作用有重要意义。

基于卫星遥感观测准确反演海洋次表层温度信息对于深入了解海洋内部的动力过程与暖化现象有重要意义。如何应用现有的卫星遥感资料结合浮标观测资料反演海洋内部关键动力环境信息场是具有挑战性的海洋遥感技术前沿。本研究融合多源卫星遥感（海表高度、海表温度、海表盐度与海表风场）和浮标观测资料（Argo浮标现场观测），创新应用高级机器学习（随机森林、XGBoost等）和地理空间建模方法（地理加权回归），发展了大尺度海洋次表层温度遥感反演模型，准确反演了海洋次表层温度场信息。利用Argo次表层现场观测开展反演验证、精度评价与适用性评估，并有效分析了海洋次表层温度遥感反演的季节时空变化特征。研究表明集成学习模型可以很好地解释海洋的非线性过程，能够通过卫星遥感观测结合浮标观测准确反演海洋次表层温度场。而地理加权回归模型考虑了海洋过程的空间非平稳性，能够显著改进普通线性回归的反演精度，且模型具有很好的可解释性。不论是机器学习方法还是地理加权回归模型均能取得较高的反演精度，可满足海洋内部观测的精度需求。本研究大大拓展了卫星对海观测维度，将卫星对海观测范围从海洋表层延伸至次表层乃至中深层。本研究可为长时序、大尺度海洋内部参量信息遥感反演与时序重建提供依据，有助于进一步发展基于卫星观测的深海遥感反演方法，对于认识海洋内部暖化过程及其全球气候变化意义有重要作用。