基于机器学习与半分析模型的珊瑚礁底栖类型遥感监测方法研究

在全球气候变化的背景下，珊瑚礁生态系统面临的威胁不断增加，而现场监测的高成本和低效率给珊瑚礁底栖的监测带来了困难。本文基于Sentinel-2 卫星影像及冰、云和陆地高程卫星2 号（Ice, Cloud and land Elevation Satellite-2，ICESat-2）激光测高数据，以印度洋、太平洋的4 座岛礁作为研究区，利用校正后的遥感水深数据对半分析模型进行约束，得到轨道处底质反射率。基于机器学习构建底质反射率与卫星影像蓝绿波段对数比之间的关系，得到整座岛礁的底质反射率。以此为基础，提出了一种从已有的底栖分类产品中自动提取可靠底栖训练样本的方法，结合底质反射率等特征参数，建立了不依赖实测数据的珊瑚礁底栖分类模型。精度评价结果显示，该分类方法在 4 座岛礁的总体精度均达到90%以上，Kappa 系数大于0.88。研究表明：该方法能够实现珊瑚礁底栖类型的高效监测，为珊瑚礁的管理和保护提供数据支撑。

英文摘要:

In the context of global climate change, the threat to coral reef ecosystems is increasing, while the high cost and low efficiency of on-site monitoring bring difficulties to the monitoring of coral reef benthic. Based on Sentinel-2 satellite image and ICESat-2 laser altimetry data, this study takes four islands and reefs in the Indian Ocean and the Pacific Ocean as the research area, and uses the corrected remote sensing water depth data to constrain the semi-analytical model to obtain the sediment reflectance at the track. Then, the relationship be tween substrate reflectance and the logarithmic ratio (Ci) of the blue and green bands of satellite imagery is constructed to obtain the reflectance of the whole island. Based on this, a method for automatically extracting reliable benthic training samples from existing benthic classification products was developed. Combined with characteristic parameters such as sediment reflectance, a coral reef benthic classifi cation model that does not rely on measured data was established. The accuracy evaluation results show that the overall accuracy of the classification method in the five islands and reefs is more than 90%, and the Kappa coefficient is greater than 0.88. It means that this method can realize the efficient monitoring of coral reef benthic types and provide data support for the management and protection of coral reefs.

查看全文查看/发表评论下载PDF阅读器